RSS2026

Abstract:
Compliance control is essential for safe physical interaction, yet its adoption is limited by hardware requirements such as force/torque sensors. While recent reinforcement learning approaches aim to bypass these constraints, they often suffer from sim-to-real gaps, lack safety guarantees, and add system complexity. We propose Minimalist Compliance Control, which enables compliant behavior using only motor current or voltage signals readily available in modern servos and quasi-direct-drive motors—without force sensors, current control, or learning. External wrenches are estimated from actuator signals and Jacobians and incorporated into a task-space admittance controller, preserving sufficient force measurement accuracy for stable and responsive compliance control. Our method is embodiment-agnostic and plug-and-play with diverse high-level planners. We validate our approach on a robot arm, a dexterous hand, and a humanoid robot across multiple contact-rich tasks, using vision-language models, imitation learning, and model-based planning. The results demonstrate robust, safe, and compliant interaction across embodiments and planning paradigms.

Abstract:
Diffusion-based policies have recently achieved remarkable success in robotics by formulating action prediction as a conditional denoising process. However, the standard practice of sampling from random Gaussian noise often requires multiple iterative steps to produce clean actions, leading to high inference latency that incurs a major bottleneck for real-time control. In this paper, we challenge the necessity of uninformed noise sampling and propose Action-to-Action flow matching (A2A), a novel policy paradigm that shifts from random sampling to initialization informed by the previous action. Unlike existing methods that treat proprioceptive action feedback as static conditions, A2A leverages historical proprioceptive sequences, embedding them into a high-dimensional latent space as the starting point for action generation. This design bypasses costly iterative denoising while effectively capturing the robot's physical dynamics and temporal continuity. Extensive experiments demonstrate that A2A exhibits high training efficiency, fast inference speed, and improved generalization. Notably, A2A enables high-quality action generation in as few as a single inference step (0.56 ms latency), and exhibits superior robustness to visual perturbations and enhanced generalization to unseen configurations. Lastly, we also extend A2A to video generation, demonstrating its broader versatility in temporal modeling.

Abstract:
Autoregressive policies offer a compelling foundation for scalable robot learning by enabling discrete abstraction, token-level reasoning, and flexible inference. However, applying autoregressive modeling to continuous robot actions requires an effective action tokenization scheme. Existing approaches either rely on analytical discretization methods that produce prohibitively long token sequences, or learned latent tokenizers that lack structure, limiting their compatibility with next-token prediction. In this work, we identify three desiderata for action tokenization — high compression, total decodability, and a left-to-right causally ordered token space — and introduce Ordered Action Tokenization (\textcolorMaroon\textttOAT), a learned action tokenizer that satisfies all three. \textcolorMaroon\textttOAT discretizes action chunks into an ordered sequence of tokens using transformer with registers, finite scalar quantization, and ordering-inducing training mechanisms. The resulting token space aligns naturally with autoregressive generation and enables prefix-based detokenization, yielding an anytime trade-off between inference cost and action fidelity. Across more than 20 tasks spanning four simulation benchmarks and real-world settings, autoregressive policies equipped with \textcolorMaroon\textttOAT consistently outperform prior tokenization schemes and diffusion-based baselines, while offering significantly greater flexibility at inference time.

Abstract:
This paper studies the problem of robot performance evaluation, focusing on how to obtain accurate and efficient estimates of real-world behavior under severe constraints on physical experimentation. Such estimates are essential for benchmarking algorithms, comparing design alternatives, validating controllers, and supporting certification or regulatory decision-making, yet real-world testing with physical robots is often expensive, time-consuming, and safety-limited. To mitigate the scarcity of real-world trials, sim-to-real methodologies are commonly employed, using low-cost simulators to inform, supplement, or prioritize physical experiments. Departing from (and complementing to) existing approaches in variance reduction (e.g., importance-sampling variants) or bias-correction (e.g., through prediction-powered inference or learned control variates), we examine this performance-evaluation problem through the lens of betting. We establish theoretical conditions under which a betting mechanism can yield accurate and efficient estimates (provably outperforming the Monte Carlo estimator) and we characterize how such bets should be constructed. We further develop theoretically grounded yet practically implementable approximations of the ideal bet, and we provide concrete decision rules that diagnose when these approximate betting strategies are working as intended. We demonstrate the effectiveness of the proposed methods using both synthetic examples and cross-fidelity computational simulators. Notably, we also showcase an illustrative case in which a group of synthetic distributions are used to infer the real-world pick-and-place accuracy of a robotic manipulator, a seemingly unconventional sim-to-real transfer that becomes natural and feasible under the proposed betting perspective. Programs for reproducing empirical results are available at https://github.com/ISUSAIL/Bet4Sim2Real

Abstract:
Inference-time guidance is essential for steering generative robot policies toward dynamic objectives without retraining, yet existing methods are largely confined to chunk-based architectures that exhibit high latency and lack the reactivity needed for test-time preference alignment or obstacle avoidance. In this work, we formally derive the optimal guidance term for Stochastic Interpolants (SI) by analyzing the value function's time evolution via the Backward Kolmogorov Equation, establishing a modified drift that theoretically guarantees sampling from a target distribution. We apply this framework to real-time control through the Streaming Stochastic Interpolant Policy (SSIP), which generalizes the deterministic Streaming Flow Policy (SFP). Unifying this guidance law with the streaming architecture enables fast and reactive control. To support diverse deployment needs, we propose two complementary mechanisms: training-free Stochastic Trajectory Ensemble Guidance (STEG) that computes gradients on-the-fly for zero-shot adaptation, and training-based Conditional Critic Guidance (CCG) for amortized inference. Empirical evaluations demonstrate that our guided streaming approach significantly outperforms conventional chunk-based policies in reactivity and provides superior, physically valid guidance for dynamic, unstructured environments.

Abstract:
While imitation learning (IL) has achieved impressive success in dexterous manipulation through generative modeling and pretraining, state-of-the-art approaches like Vision-Language-Action (VLA) models still struggle with adaptation to environmental changes and skill transfer. We argue this stems from mimicking raw trajectories without understanding the underlying intent. To address this, we propose explicitly disentangling behavior intent from execution details in end-2-end IL: ``Mimic Intent, Not just Trajectories'' (MINT) . We achieve this via multi-scale frequency-space tokenization , which enforces a spectral decomposition of action chunk representation. We learn action tokens with a multi-scale coarse-to-fine structure, and force the coarsest token to capture low-frequency global structure and finer tokens to encode high-frequency details. This yields an abstract Intent token that facilitates planning and transfer, and multi-scale Execution tokens that enable precise adaptation to environmental dynamics. Building on this hierarchy, our policy generates trajectories through next-scale autoregression , performing progressive intent-to-execution reasoning , thus boosting learning efficiency and generalization. Crucially, this disentanglement enables one-shot transfer of skills, by simply injecting the Intent token from a demonstration into the autoregressive generation process. Experiments on several manipulation benchmarks and on a real robot demonstrate state-of-the-art success rates, superior inference efficiency, robust generalization against disturbances, and effective one-shot transfer.

Abstract:
This work highlights that video world modeling, alongside vision-language pre-training, establishes a distinct foundation for robot learning. By capturing environmental dynamics, video world models enable the robot to "imagine" near future states—a capability essential for effective planning. Inspired by this, we introduce CauVA, an autoregressive diffusion framework that unifies frame prediction and action inference. Our approach features three key innovations: (1) an autoregressive Mixture-of-Transformers (MoT) that processes visual frames and actions as a single causal sequence; (2) a history integration mechanism using KV cache to maintain temporal context from real-world interactions; and (3) a noisy latent augmentation strategy that enables decoding actions directly from intermediate denoised videos for fast inference. Evaluations on simulation and real-world benchmarks demonstrate that CauVA excels in complex manipulation, including long-horizon, high-precision, and deformable object tasks. Our code and models are publicly available.

Abstract:
Connectivity maintenance is critical for networked robot teams to exchange information and coordinate actions, yet enforcing persistent global connectivity can be unnecessarily restrictive when communication ranges are limited and teams operate at scale. In practice, robots must be allowed to temporarily disconnect so as to spread out efficiently, and then deliberately reconfigure their motions for local reconnection to maintain time-aggregated connectivity that allows information to flow across the team over a window of time. In this paper, we propose a novel motion coordination framework for robots to maintain time-aggregated connectivity while progressing to their individual goals. As a modularized layer built on top of robots' nominal motion plans, it encompasses (a) a time-window aggregated minimum spanning tree (TWA-MST) approach to dynamically decide which robot pairs should be reconnected and when, and (b) a novel notion of adaptive prescribed-time control barrier functions (adaptive PT-CBF) that enforce assured reconnection by motion reconfiguration. This allows robots to follow their nominal plans while minimally adjusting motions to ensure time-aggregated connectivity over a defined time horizon. Both theoretical analysis and experimental results are provided to demonstrate the effectiveness of the proposed methods.

Abstract:
Zero-Shot Object Goal Navigation (ZSON) is an important task for robots. While Multimodal Large Language Models (MLLMs) have empowered robots with significant semantic reasoning capabilities, current RGB-only navigation methods still struggle to align high-level discrete logic with low-level continuous physical execution, often due to a lack of explicit geometric constraints and spatial memory. In this paper, we propose MVP-Nav (Multi-layer ValueMap Planner Navigator), a hierarchical framework designed for robust RGB-only ZSON. Our approach utilizes a 3D foundation model to recover the physical scale and spatial occupancy of semantic instances from monocular observations, representing them as Oriented Bounding Boxes (OBB) within a dynamic Spatial Semantic List. At the core of our system is the Multi-layer Value Maps (MVM) mechanism, which serves as a navigation hub: the MLLM acts as a high-level planner to assign semantic weights and determine navigation modes, while the low-level controller performs precise geometric path planning within a cost space fused with physical constraints. Experimental results demonstrate that MVP-Nav achieves state-of-the-art (SOTA) success rates and exploration efficiency among depth-free methods, even surpassing several depth-based benchmarks.

Abstract:
Embodied Chain-of-Thought (CoT) reasoning has significantly enhanced Vision-Language-Action (VLA) models, yet current methods rely on rigid templates to specify reasoning primitives (e.g., objects in the scene, high-level plans, structural affordances). These templates can force policies to process irrelevant information that distracts from critical action-prediction signals. This creates a bottleneck: without successful policies, we cannot verify reasoning quality; without quality reasoning, we cannot build robust policies. We introduce R&B-EnCoRe, which enables models to bootstrap embodied reasoning from internet-scale knowledge through self-supervised refinement. By treating reasoning as a latent variable within importance-weighted variational inference, models can generate and distill a refined reasoning training dataset of embodiment-specific strategies without external rewards, verifiers, or human annotation. We validate R&B-EnCoRe across manipulation (Franka Panda in simulation, WidowX in hardware), legged navigation (bipedal, wheeled, bicycle, quadruped), and autonomous driving embodiments using various VLA architectures with 1B, 4B, 7B, and 30B parameters. Our approach achieves 28% gains in manipulation success, 101% improvement in navigation scores, and 21% reduction in collision-rate metric over models that indiscriminately reason about all available primitives. R&B-EnCoRe enables models to distill reasoning that is predictive of successful control, bypassing manual annotation engineering while grounding internet-scale knowledge in physical execution.

Abstract:
Reliable benchmarking is critical for advancing Vision–Language–Action (VLA) models, as it reveals their generalization, robustness, and alignment of perception with language-driven manipulation tasks. However, existing benchmarks often provide limited or misleading assessments due to insufficient evaluation protocols that inadequately capture real-world distribution shifts. This work systematically rethinks VLA benchmarking from both evaluation and data perspectives, introducing LIBERO-X, a benchmark featuring: 1) A hierarchical evaluation protocol with progressive difficulty levels targeting three core capabilities: spatial generalization, object recognition, and task instruction understanding. This design enables fine-grained analysis of performance degradation under increasing environmental and task complexity; 2) A high-diversity training dataset collected via human teleoperation, where each scene supports multiple fine-grained manipulation objectives to bridge the train-evaluation distribution gap. Experiments with representative VLA models reveal significant performance drops under cumulative perturbations, exposing persistent limitations in scene comprehension and instruction grounding. By integrating hierarchical evaluation with diverse training data, LIBERO-X offers a more reliable foundation for assessing and advancing VLA development.

Abstract:
Robust trajectory optimization enables autonomous systems to operate safely under uncertainty by computing control policies that satisfy the constraints for all bounded disturbances. However, these problems often lead to large Second Order Conic Programming (SOCP) constraints, which are computationally expensive. In this work, we propose the CUDA Nonlinear Robust Trajectory Optimization (cuNRTO) framework by introducing two dynamic optimization architectures that have direct application to robust decision-making and are implemented on CUDA. The first architecture, NRTO-DR, leverages the Douglas-Rachford (DR) splitting method to solve the SOCP inner subproblems of NRTO, thereby significantly reducing the computational burden through parallel SOCP projections and sparse direct solves. The second architecture, NRTO-FullADMM, is a novel variant that further exploits the problem structure to improve scalability using the Alternating Direction Method of Multipliers (ADMM). Finally, we provide a GPU implementation of the proposed methodologies using custom CUDA kernels for SOC projection steps and cuBLAS GEMM chains for feedback gain updates. We validate the performance of cuNRTO through simulated experiments on unicycle, quadcopter, and Franka manipulator models, demonstrating speedup up to 139.6×.

Abstract:
Existing learning-based occupancy prediction methods rely on large-scale 3D annotations and generalize poorly across environments. We present FreeOcc, a training-free framework for open-vocabulary occupancy prediction from monocular or RGB-D sequences. Unlike prior approaches that require voxel-level supervision and ground-truth camera poses, FreeOcc operates without 3D annotations, pose ground truth, or any learning stage. FreeOcc incrementally builds a globally consistent occupancy map via a four-layer pipeline: a SLAM backbone estimates poses and sparse geometry; a geometrically consistent Gaussian update constructs dense 3D Gaussian maps; open-vocabulary semantics from off-the-shelf vision–language models are associated with Gaussian primitives; and a probabilistic Gaussian-to-occupancy projection produces dense voxel occupancy. Despite being entirely training-free and pose-agnostic, FreeOcc achieves over 2× improvements in IoU and mIoU on EmbodiedOcc-ScanNet compared to prior self-supervised methods. We further introduce ReplicaOcc, a benchmark for indoor open-vocabulary occupancy prediction, and show that FreeOcc transfers zero-shot to novel environments, substantially outperforming both supervised and self-supervised baselines. Code will be released.

Abstract:
vent cameras report asynchronously per-pixel brightness changes with microsecond latency, encoding dynamic visual information as a sparse stream of events. However, their extreme temporal resolution floods perception systems with entangled events from ego-motion and indepen- dently moving objects (IMOs), which existing solutions fail to efficiently decouple, relying instead on prohibitive dense 3D reconstructions or limited bio-inspired filters. In this work, we introduce the first framework for Motion- aware Event Suppression, which learns to filter events triggered by IMOs and ego-motion in real time. Our model jointly segments IMOs in the current event stream while predicting their future motion, enabling anticipatory suppression of dynamic events before they occur. Our lightweight architecture achieves 173 Hz inference on consumer-grade GPUs with less than 1 GB of memory usage, outperforming previous state-of-the-art methods on the challenging EVIMO benchmark by 67% in segmentation accuracy while operating at a 53% higher inference rate. Moreover, we demonstrate significant benefits for down- stream applications: our method accelerates Vision Transformer inference by 83% via token pruning and improves event-based visual odometry accuracy, reducing Absolute Trajectory Error (ATE) by 13%.

Abstract:
Imitation learning has achieved remarkable success in robotic manipulation, yet its application to surgical robotics remains challenging due to data scarcity, constrained workspaces, and the need for an exceptional level of safety and predictability. We present a supervised Mixture-of-Experts (MoE) architecture designed for phase-structured surgical manipulation tasks, which can be added on top of any autonomous policy. Unlike prior surgical robot learning approaches that rely on multi-camera setups or thousands of demonstrations, we show that a lightweight action decoder policy like Action Chunking Transformer (ACT) can learn complex, long-horizon manipulation from less than 150 demonstrations using solely stereo endoscopic images, when equipped with our architecture. We evaluate our approach on the collaborative surgical task of bowel grasping and retraction, where a robot assistant interprets visual cues from a human surgeon, executes targeted grasping on deformable tissue, and performs sustained retraction. We benchmark our method against state-of-the-art Vision-Language-Action (VLA) models and the standard ACT baseline. Our results show that generalist VLAs fail to acquire the task entirely, even under standard in-distribution conditions. Furthermore, while standard ACT achieves moderate success in-distribution, adopting a supervised MoE architecture significantly boosts its performance, yielding higher success rates in-distribution and demonstrating superior robustness in out-of-distribution scenarios, including novel grasp locations, reduced illumination, and partial occlusions. Notably, it generalizes to unseen testing viewpoints and also transfers zero-shot to ex vivo porcine tissue without additional training, offering a promising pathway toward in vivo deployment. To support this statement, we present qualitative preliminary results of policy roll-outs during in vivo porcine surgery. These results demonstrate that supervised MoE architectures provide a data-efficient approach for learning multi-step dexterous manipulation in visually constrained environments. Code and dataset will be released upon acceptance.

Abstract:
Human demonstrations collected by wearable devices (e.g., tactile gloves) provide fast and dexterous supervision for policy learning, and are guided by rich, natural tactile feedback. However, a key challenge is how to transfer human-collected tactile signals to robots despite the differences in sensing modalities and embodiment. Existing human-to-robot (H2R) approaches that incorporate touch often assume identical tactile sensors, require paired data, and involve little to no embodiment gap between human demonstrator and the robots, limiting scalability and generality. We propose TactAlign, a cross-embodiment tactile alignment method that transfers human-collected tactile signals to a robot with different embodiment. TactAlign transforms human and robot tactile observations into an shared latent representation using a rectified flow, without paired datasets, manual labels, or privileged information. Our method enables low-cost latent transport guided by hand-object interaction-derived pseudo-pairs. We demonstrate that TactAlign improves H2R policy transfer across multiple contact-rich tasks (pivoting, insertion, lid closing), generalizes to unseen objects and tasks with human data (approximately 5 minutes), and enables zero-shot H2R transfer on a highly dexterous tasks (light bulb screwing).

Abstract:
Underwater robotic grasping is difficult due to degraded, highly variable imagery and the expense of collecting diverse underwater demonstrations. We introduce a system that (i) autonomously collects successful underwater grasp demonstrations via a self-supervised data collection pipeline and (ii) transfers grasp knowledge from on-land human demonstrations through a depth-based affordance representation that bridges the on-land–to–underwater domain gap and is robust to lighting and color shift. An affordance model trained on on-land handheld demonstrations is deployed underwater zero-shot via geometric alignment, and an affordance-conditioned diffusion policy is then trained on underwater demonstrations to generate control actions. In pool experiments, our approach improves grasping performance and robustness to background shifts, and enables generalization to objects seen only in on-land data, outperforming RGB-only baselines.

Abstract:
Despite the sustained scaling on model capacity and data acquisition, Vision–Language–Action (VLA) models remain brittle in contact-rich and dynamic manipulation tasks, where minor execution deviations can compound into failures. While reinforcement learning (RL) offers a principled path to robustness, on-policy RL in the physical world is constrained by safety risk, hardware cost, and environment reset. To bridge this gap, we present RISE, a scalable framework of robotic reinforcement learning via imagination. At its core is a Compositional World Model that (i) predicts multi-view future via a controllable dynamics model, and (ii) evaluates imagined outcomes with a progress value model, producing informative advantages for the policy improvement. Such compositional design enables state and value to be modeled by best-suited yet distinct architectures and objectives. These components are integrated into a closed-loop self-improving pipeline that continuously generates imaginary rollouts, estimates advantages, and updates the policy in imaginary space without costly physical interaction. Across three challenging real-world tasks, RISE yields significant improvement over prior art, with more than +35% absolute performance increase in dynamic brick sorting, +45% for backpack packing, and +35% for box closing, respectively.

Abstract:
Multi-agent navigation in unknown and cluttered environments has broad applications, yet remains fundamentally challenging. In particular, dense agent-agent and agent-obstacle reactive interactions can exacerbate the inherent competition between collision-avoidance constraints and goal-reaching objectives. Most existing approaches mitigate this by applying per-step safety filtering on top of a predefined goal-reaching controller or by designing heuristic loss functions that penalizes safety constraints violation gradient. While effective in sparse environments, these methods still suffer from overly-conservative behaviors when interactions become dense. To overcome these limitations, we propose HJB-GNN, a Hamilton-Jacobi-Bellman (HJB)-based learning framework that jointly learns a graph neural network (GNN)-parameterized control barrier function for explicit safety enforcement, a distributed GNN-based navigation policy, and a value function that induces goal-reaching behavior. By exploiting the analytical solution of the constrained HJB equation, the proposed method derives graph-dependent Lagrange multipliers that adaptively balance collision-avoidance and goal-reaching across diverse multi-agent navigation scenarios. Moreover, HJB-GNN supports centralized training with distributed deployment. Extensive simulations and real-world experiments with Crazyflie drone swarms demonstrate its superior safety and goal-reaching performance, as well as strong scalability and generalizability to large-scale teams operating in previously unseen, dense environments.

Abstract:
Localization and mapping of an environment are crucial tasks for any robot operating in unstructured environments. Time-of-flight (ToF) sensors (e.g.,~lidar) have proven useful in mobile robotics, where high-resolution sensors can be used for simultaneous localization and mapping. In soft and continuum robotics, however, these high-resolution sensors are too large for practical use. This, combined with the deformable nature of such robots, has resulted in continuum robot (CR) localization and mapping in unstructured environments being a largely untouched area. In this work, we present a localization technique for CRs that relies on small, low-resolution ToF sensors distributed along the length of the robot. By fusing measurement information with a robot shape prior, we show that accurate localization is possible despite each sensor experiencing frequent degenerate scenarios. We achieve an average localization error of 2.5cm in position and 7.2° in rotation across all experimental conditions with a 53cm long robot. We demonstrate that the results are repeated across multiple environments, in both simulation and real-world experiments, and study robustness in the estimation to deviations in the prior map.

Abstract:
We present VGGT-SLAM 2.0, a real-time RGB feed-forward SLAM system which substantially improves upon VGGT-SLAM for incrementally aligning submaps created from VGGT. Firstly, we remove high-dimensional 15-degree-of-freedom drift and planar degeneracy from VGGT-SLAM by creating a new factor graph design while still addressing the reconstruction ambiguity of VGGT given unknown camera intrinsics. Secondly, by studying the attention layers of VGGT, we show that one of the layers is well suited to assist in image retrieval verification for free without additional training, which enables both rejecting false positive matches and allows for completing more loop closures. Finally, we conduct a suite of experiments which includes showing VGGT-SLAM 2.0 can easily be adapted for open-set object detection and demonstrating real-time performance while running online onboard a ground robot using a Jetson Thor. We test in environments ranging from cluttered indoor apartments and office scenes to a 4,200 square foot barn, and we also demonstrate VGGT-SLAM 2.0 achieves the highest accuracy on the TUM dataset with about 23 percent less pose error than VGGT-SLAM. Code will be released upon publication.

Abstract:
In safety-critical robotics applications, guaranteed and practical uncertainty quantification (UQ) in perception is vital. Many existing works either offer no formal containment guarantee, rely on restrictive modeling assumptions, or focus only on pose estimation rather than a complete SLAM pipeline. This paper presents provably guaranteed UQ algorithms for 3D-3D landmark-based SLAM. The algorithms consist of three basic UQ modules: forward UQ for mapping, backward UQ for pose tracking, and pose compound. Each module produces a certified uncertainty set; when the input uncertainty bounds are deterministic, the output sets inherit deterministic guarantees, i.e., they provably contain the true poses and landmarks. Specifically, we use polytopes to represent uncertainty sets, enabling tractable computations and a unified treatment of pose uncertainty. To enhance algorithms' practical usability, we incorporate conformal prediction to calibrate measurement uncertainty from data with prescribed probability. Simulations and experiments demonstrate that the proposed algorithms provide both strong theoretical guarantees and practical usability.

Abstract:
Diffusion-based policies have established a new standard for precise robotic manipulation but face a critical scalability bottleneck: high-performance models are computationally expensive, while lightweight alternatives often fail to generalize across diverse multi-task environments. Mixture-of-Experts (MoE) architectures offer a promising path to efficiency by activating only a subset of parameters. However, existing MoE routing mechanisms typically rely on low-level noise or latent statistics, ignoring the compositional nature of manipulation tasks. This results in redundant experts that fail to capture reusable skills, limiting both interpretability and transfer. We introduce Semantically Structured Mixture-of-Experts Diffusion Policy for Compositional Robotic Manipulation (SMoDP), a framework that grounds expert specialization in semantic task structure. SMoDP leverages a lightweight, inference-time skill predictor—distilled from offline VLM annotations—to route action chunks to experts specialized for specific behavioral phases. To ensure robust assignment, we propose a dual contrastive alignment strategy that grounds multi-modal observations in language-defined skill semantics (Inter-modal) while enforcing routing consistency across visually distinct but functionally identical behaviors (Intra-modal). Our approach achieves state-of-the-art performance on multi-task benchmarks with significantly improved parameter efficiency and demonstrates effective compositional transfer to novel tasks through parameter-efficient fine-tuning.

Abstract:
Low speed does not always guarantee safety in off-road driving. For instance, crossing a ditch may be risky at a low speed due to the risk of getting stuck, yet safe at a higher speed with a controlled, accelerated jump. Achieving such behavior requires path planning that explicitly models complex motion dynamics, whereas existing methods often neglect this aspect and plan solely based on positions or a fixed velocity. To address this gap, we introduce Motion-aware Traversability (MAT) representation to explicitly model terrain cost conditioned on actual robot motion. Instead of assigning a single scalar score for traversability, MAT models each terrain region as a Gaussian function of velocity. During online planning, we decompose the terrain cost computation into two stages: (1) predict terrain-dependent Gaussian parameters from perception in a single forward pass, (2) efficiently update terrain costs for new velocities inferred from current dynamics by evaluating these functions without repeated inference. We develop a system that integrates MAT to enable agile off-road navigation and evaluate it in both simulated and real-world environments with various obstacles. Results show that MAT achieves real-time efficiency and enhances the performance of off-road navigation, reducing path detours by 75% while maintaining safety across challenging terrains.

Abstract:
Open-world navigation requires robots to make decisions in complex everyday environments while adapting to flexible task requirements. Conventional navigation approaches often rely on dense 3D reconstruction and hand-crafted goal metrics, which limits their generalization across tasks and environments. Recent advances in vision-language navigation (VLN) and vision-language-action (VLA) models enable end-to-end policies conditioned on natural language, but typically require interactive training, large-scale data collection, or task-specific fine-tuning with a mobile agent. We formulate navigation as a sparse subgoal identification and reaching problem and observe that providing visual anchoring targets for high-level semantic priors enables highly efficient goal-conditioned navigation. Based on this insight, we select navigation frontiers as semantic anchors and propose OpenFrontier, a training-free navigation framework that seamlessly integrates diverse vision-language prior models. OpenFrontier enables efficient navigation with a simple system design, without dense 3D mapping, policy training, or model fine-tuning. We evaluate OpenFrontier across multiple navigation benchmarks and demonstrate strong zero-shot performance, as well as effective real-world deployment on a mobile robot.

Abstract:
Vision–Language–Action (VLA) models have shown strong potential for general-purpose robotic manipulation by leveraging large pretrained vision-language backbones. However, most existing VLAs rely primarily on 2D visual representations, which limits their ability to reason about fine-grained geometry and spatial grounding - capabilities that are essential for precise and robust manipulation in 3D environments. In this paper, we propose PointACT, a dual-system 3D-aware VLA policy that integrates hierarchical 3D point cloud representations directly into the action decoding process. PointACT employs a multi-scale point-action interaction mechanism with efficient bottleneck window self-attention, enabling evolving action tokens to densely attend to both local geometric detail and global scene structure. We evaluate PointACT on the LIBERO and RLBench benchmarks and systematically compare it against monolithic and dual-system VLA baselines, including variants augmented with point cloud inputs. PointACT achieves consistent improvements across both benchmarks, increasing success rates by 10% on the challenging RLBench-10Tasks suite over state-of-the-art pretrained VLAs, with even larger gains when the vision–language backbone is frozen and the action expert is trained from scratch. Extensive ablation studies demonstrate that tightly coupling hierarchical 3D geometry with pretrained 2D semantic representations is critical for robust and spatially grounded robot control. Our results also highlight the promise of pretrained 3D representations for 3D-aware VLA policies.

Abstract:
We propose a standalone autoregressive (AR) Action Expert that generates actions as a continuous causal sequence while conditioning on refreshable vision-language prefixes. In contrast to existing Vision-Language-Action (VLA) models and diffusion policies that reset temporal context with each new observation and predict actions reactively, our Action Expert maintains its own history through a long-lived memory and is inherently context-aware. This structure addresses the frequency mismatch between fast control and slow reasoning, enabling efficient independent pretraining of kinematic syntax and modular integration with heavy perception backbones, naturally ensuring spatio-temporally consistent action generation across frames. To synchronize these asynchronous hybrid V-L-A modalities, we utilize a re-anchoring mechanism that mathematically accounts for perception staleness during both training and inference. Experiments on simulated and real-robot manipulation tasks demonstrate that the proposed method can effectively replace traditional chunk-based action heads for both specialist and generalist policies. AR-VLA exhibits superior history awareness and substantially smoother action trajectories while maintaining or exceeding the task success rates of state-of-the-art reactive VLAs. Overall, our work introduces a scalable, context-aware action generation schema that provides a robust structural foundation for training effective robotic policies.

Authors: Ali Amin, Raichelle Aniceto, Ashwin Balakrishna, Kevin Black, Ken Conley, Grace B. Connors, James Darpinian, Karan Dhabalia, Jared Di Carlo, Danny Driess, Michael Robert Equi, Adnan Esmail, Yunhao Fang, Chelsea Finn, Catherine Glossop, Thomas Godden, Ivan Goryachev, Lachy Groom, Hunter Hancock, Karol Hausman, Gashon Hussein, Brian Ichter, Szymon Jakubczak, Rowan Jen, Tim Jones, Benjamin Katz, Liyiming Ke, Chandra Kuchi, Marinda Lamb, Devin Leblanc, Sergey Levine, Adrian Li-Bell, Yao Lu, Vishnu Mano, Mohith Mothukuri, Suraj Nair, Karl Pertsch, Allen Z. Ren, Charvi Sharma, Lucy Xiaoyang Shi, Laura Smith, Jost Tobias Springenberg, Kyle Stachowicz, Will Stoeckle, Alexander Swerdlow, James Tanner, Marcel Torne, Quan Vuong, Anna Walling, Haohuan Wang, Blake Williams, Sukwon Yoo, Lili Yu, Ury Zhilinsky, Zhiyuan Zhou

Abstract:
Vision–language–action (VLA) models offer a promising path toward general-purpose robots, but achieving the reliability and speed required for practical deployment remains challenging. We present a general-purpose method, RL with Experience and Corrections via Advantage-conditioned Policies (RECAP) that improves the efficiency and reliability of VLA policies by utilizing their real-world experience. Our method introduces value-based advantage conditioning during both pre-training and post-training phases, enabling VLA policies to ingest highly heterogeneous real-world experience, including human demonstrations, policy rollouts, and online correction data. We show that the π 0.6 model, trained with RECAP, achieves hours-long deployment of folding diverse laundry in real homes, can reliably assemble boxes in a factory, and make espresso drinks using a professional espresso machine. On some of the hardest tasks, RECAP more than doubles task throughput and roughly halves the task failure rate.

Abstract:
While Vision-Language-Action (VLA) models excel in generalist manipulation, they often lack fine-grained spatial awareness and struggle with viewpoint generalization. This limitation largely stems from the reliance on pretrained RGB encoders, which lack explicit geometric cues and prioritize semantic alignment over geometric representation. We argue that effective visual representations for VLA models must jointly encode both semantic and geometric information. In this paper, we introduce StereoVLA, the first VLA model to incorporate rich geometric cues from large-scale synthetic stereo data. StereoVLA employs a Geometric-and-Semantic (GeoSem) vision encoder that extracts geometric cues from subtle stereo-view disparities for precise spatial perception, while simultaneously capturing semantic features from pixel observations to support language-conditioned manipulation. Additionally, we introduce two synergistic co-training objectives: Interaction-Region Depth Estimation for precise spatial reasoning, and Camera Parameter Estimation to implicitly align perception and action coordinate systems. Compared with baselines that employ various input modalities, StereoVLA achieves a 33.4% improvement in real-world experiments and demonstrates robust generalization to near-hemispheric camera perspectives.

Abstract:
Matching the swimming efficiency and agility of fish has remained an elusive goal in underwater robotics. Such locomotion capabilities rely on complex vortex interactions between the robot's body and the surrounding fluid. However, simulating these dynamics, which are governed by coupled ordinary and partial differential equations, is significantly more difficult than the multi-body dynamics of classical rigid robotic systems. We present a differentiable framework for simulating strongly coupled fluid-robot multiphysics as a unified optimization problem. The coupled manipulator and incompressible Navier-Stokes equations are derived together from a single Lagrangian using the principle of least action. We employ discrete variational mechanics to derive a stable, well-conditioned, and physically accurate scheme for jointly simulating articulated bodies and the surrounding fluid. We leverage the implicit function theorem to compute derivatives of the fully coupled dynamics. Using this simulator and its gradients, we realize undulating swimming gaits and optimize a highly dynamic C-start escape maneuver for a bioinspired eel robot. We validate both gaits on physical hardware, demonstrating successful sim-to-real transfer.

Abstract:
Existing Vision-Language-Action (VLA) models often suffer from feature collapse and low training efficiency because they entangle high-level perception with sparse, embodiment-specific action supervision. Since these models typically rely on VLM backbones optimized for Visual Question Answering (VQA), they excel at semantic identification but often overlook subtle 3D state variations that dictate distinct action patterns. To resolve these misalignments, we propose Pose-VLA, a decoupled paradigm that separates VLA training into a pre-training phase for extracting universal 3D spatial priors in a unified camera-centric space, and a post-training phase for efficient embodiment alignment within robot-specific action space. By introducing discrete pose tokens as a universal representation, Pose-VLA seamlessly integrates spatial grounding from diverse 3D datasets with geometry-level trajectories from robotic demonstrations. Our framework follows a two-stage pre-training pipeline, establishing fundamental spatial grounding via poses followed by motion alignment through trajectory supervision. Extensive evaluations demonstrate that Pose-VLA achieves state-of-the-art results on RoboTwin 2.0 with a 79.5% average success rate and competitive performance on LIBERO at 96.0%. Real-world experiments further showcase robust generalization across diverse objects using only 100 demonstrations per task, validating the efficiency of our pre-training paradigm.

Abstract:
Offline reinforcement learning (RL) should be ideal for robotics, allowing learning from dataset without risky exploration. Yet, offline RL's performance often hinges on a brittle trade-off between (1) return maximization, which can push policies outside dataset support, and (2) behavioral constraints, which typically require sensitive hyperparameter tuning. Latent steering offers a structural way to stay within dataset support during RL but, in order to approximate action values, existing offline adaptations commonly rely on latent-space critics learned via indirect distillation, which can lose information and hinder convergence. We propose Latent Policy Steering (LPS), which enables high-fidelity latent policy improvement by backpropagating original-action-space Q-gradients through a differentiable one-step MeanFlow policy to update a latent-action-space actor. By eliminating proxy latent critics, LPS allows an original-action-space critic to guide end-to-end latent-space optimization, while the one-step MeanFlow policy serves as a behavior-constrained generative prior. This decoupling yields a robust method that works out-of-the-box with minimal tuning. Across OGBench and physical-world robotic tasks, LPS achieves state-of-the-art performance and consistently outperforms behavioral cloning and strong latent steering baselines.

Abstract:
Adaptation to damages and in-situ physical repairs is essential for long-term robot autonomy, yet challenging outside of narrowly defined and well-anticipated bounds. In this work we proprioceptively adapt to catastrophic damage in soft-actuated systems in under one minute. Architected materials are well equipped for adaptation: actuator failure occurs gradually rather than acutely, and damage can be described in a low-dimensional, discrete coordinate space. Surprisingly, latent damage representations plus a simple yet robust ensemble method is sufficient for adapting to unseen damage in real-time. We demonstrate LEAP, our method for adaptive proprioception, via a tracing task for a 6DoF soft wrist based on Handed Shearing Auxetic (HSA) actuators. Our algorithm is able to adapt to cuts, burns, and actuator repairs, enabling simulation-free real-time adaptation that is critical for realizing the promise of soft robots outside the lab.

Abstract:
Diffusion-based trajectory planners can synthesize rich, multimodal robot motions from demonstrations, but their iterative denoising makes online planning and control prohibitively slow. Existing accelerations either modify the sampler or compress the network--sacrificing plan quality or requiring retraining without accounting for downstream control risk. We address the problem of making diffusion-based trajectory planners fast enough for real-time robot use without retraining the model or sacrificing trajectory quality, and in a way that works across diverse state-space diffusion architectures. Our key insight is that diffusion trajectory planners already expose two signals we can exploit: a cheap probe of how their internal trajectory representation changes across steps, and analytic coefficients that describe how denoiser errors affect the sampler's state update. By calibrating the first signal against the second on offline runs, we obtain a per-step score that upper-bounds how far the final trajectory can deviate when we reuse a cached denoiser output, and we treat this bound as an uncertainty budget that we can spend over the denoising process. Building on this insight, we present Muninn, a training-free caching wrapper that tracks this uncertainty budget during sampling and, at each diffusion step, chooses between reusing a cached denoiser output when the predicted deviation is small and recomputing the denoiser when it is not. Across standard benchmarks spanning offline RL planning (D4RL), configuration-space motion planning, and visuomotor diffusion policies, Muninn delivers up to 4.6× wall-clock speedups across several trajectory diffusion planners and diffusion policies by reducing denoiser evaluations, while preserving task performance and safety metrics. Muninn further certifies---at a user-chosen deviation tolerance and risk level---that cached rollouts remain within a specified distance of their full-compute counterparts, and we validate these gains in real-time closed-loop navigation and manipulation hardware deployments.

Abstract:
Generating collision-free and smoothly executable motions is a persistent challenge in robotic manipulation, especially in cluttered workspaces and narrow passages where the feasible set is highly nonconvex and fragmented. We propose a trajectory optimization method that performs geometry-aware updates directly in function space via natural functional gradients. Our approach optimizes a Gaussian-smoothed surrogate objective that regularizes the landscape through trajectory perturbations while preserving trajectory-level structure. Because updates are defined intrinsically in function space, trajectory regularity is controlled independently of the time grid, avoiding discretization-tuned smoothness penalties. We derive a practical Monte-Carlo estimator of the natural functional gradient that requires only black-box cost evaluations, making the method applicable when analytic gradients are unavailable or unreliable due to collision checking and contact-rich simulation. Across manipulation benchmarks with dense clutter and narrow clearances, the proposed optimizer achieves higher success rates and produces trajectories with lower acceleration and jerk than representative state-of-the-art baselines.

Abstract:
In the presence of occlusions and measurement noise, geometrically accurate scene reconstructions---which fit the sensor data---can still be physically incorrect. For instance, when estimating the poses and shapes of objects in the scene and importing the resulting estimates into a simulator, small errors might translate to implausible configurations including object interpenetration or unstable equilibrium. This makes it difficult to predict the dynamic behavior of the scene using a digital twin, an important step in simulation-based planning and control of contact-rich behaviors. In this paper, we posit that object pose and shape estimation requires reasoning holistically over the scene (instead of reasoning about each object in isolation), accounting for object interactions and physical plausibility. Towards this goal, our first contribution is Picasso, a physics-constrained reconstruction pipeline that builds multi-object scene reconstructions by considering geometry, non-penetration, and physics. Picasso relies on a fast rejection sampling method that reasons over multi-object interactions, leveraging an inferred object contact graph to guide samples. Second, we propose the Picasso dataset, a collection of 10 contact-rich real-world scenes with ground truth annotations, as well as a metric to quantify physical plausibility, which we open-source as part of our benchmark. Finally, we provide an extensive evaluation of Picasso on our newly introduced dataset and on the YCB-V dataset, and show it largely outperforms the state of the art while providing reconstructions that are both physically plausible and more aligned with human intuition.

Abstract:
We present Viser, a toolkit for 3D visualization in robotics and computer vision. Viser aims to bring easy and extensible 3D visualization to Python: we provide comprehensive 3D scene and 2D GUI primitives, which can be used independently with minimal setup or composed to build specialized interfaces. This paper details features and key design choices—an imperative-style API and a web-based viewer—which improve compatibility with modern programming patterns. We then discuss system architecture, adoption, and limitations.

Abstract:
Active 3D reconstruction of moving objects requires selecting informative viewpoints while accounting for object motion during the decision-to-execution delay. However, most next-best-view (NBV) planners assume static objects, while motion-aware active perception for moving targets typically prioritizes tracking over surface coverage. We present a motion-uncertainty-aware NBV framework for reconstructing an unknown rigid object undergoing planar translation, using only noisy planar position measurements of the object and depth observations from a separate mobile robot. Our key idea is to plan over a predictive distribution of future camera-object configurations. We maintain a predictive object-state belief over planar position and velocity using a fixed-lag Gaussian Process smoother, propagate it one step forward, and generate candidate viewpoints around the predicted object location. We filter candidates by single-step reachability, then evaluate feasible viewpoints by expected coverage gain under the predictive belief via Monte Carlo sampling of induced camera-object configurations, and execute the highest-utility feasible view. Simulations and real-world experiments demonstrate improved surface coverage and reconstruction completeness over non-predictive and tracking-only baselines, bridging tracking-driven prediction and coverage-driven NBV.

Abstract:
Environmental monitoring robots often need to reconstruct spatial fields (e.g., salinity, temperature, bathymetry) under tight distance and energy constraints. Classical boustrophedon lawnmower surveys provide geometric coverage guarantees but can waste effort by oversampling predictable regions. In contrast, informative path planning (IPP) methods leverage spatial correlations to reduce oversampling, yet typically offer no guarantees on reconstruction quality. This paper bridges these approaches by addressing informative path planning with guaranteed estimation uncertainty: computing the shortest path whose measurements ensure that the Gaussian-process (GP) posterior variance—an intrinsic uncertainty measure that lower-bounds the mean-squared prediction error under the GP model—falls below a user-specified threshold over the monitoring region. We propose a three-stage approach: (i) learn a GP model from available prior information; (ii) transform the learned GP kernel into binary coverage maps for each candidate sensing location, indicating which locations’ uncertainty can be reduced below a specified target; and (iii) plan a near-shortest route whose combined coverage satisfies the global uncertainty constraint. To address heterogeneous phenomena, we incorporate a nonstationary kernel that captures spatially varying correlation structure, and we accommodate non-convex environments with obstacles. Algorithmically, we present methods with provable approximation guarantees for sensing-location selection and for the joint selection-and-routing problem under a travel budget. Experiments on real-world topographic data show that our planners meet the uncertainty target using fewer sensing locations and shorter travel distances than a recent baseline, and field experiments with bathymetry-mapping autonomous surface and underwater vehicles demonstrate real-world feasibility.

Abstract:
A long-standing goal in robotics is a generalist policy that can be deployed zero-shot on new robot embodiments without per-embodiment adaptation. Despite large-scale multi-embodiment pre-training, existing Vision--Language--Action models (VLAs) remain tightly coupled to their training embodiments and typically require costly fine-tuning. We introduce Language-Action Pre-training (LAP), a simple recipe that represents low-level robot actions directly in natural language, aligning action supervision with the pre-trained vision--language model’s input--output distribution. LAP requires no learned tokenizer, no costly annotation, and no embodiment-specific architectural design. Based on LAP, we present LAP-3B, which to the best of our knowledge is the first VLA to achieve substantial zero-shot transfer to previously unseen robot embodiments without any embodiment-specific fine-tuning. Across multiple novel robots and manipulation tasks, LAP-3B attains over 50% average zero-shot success, delivering roughly a 2× improvement over the strongest prior VLAs. We further show that LAP enables efficient adaptation and favorable scaling, while unifying action prediction and VQA in a shared language-action format that yields additional gains through co-training.

Abstract:
Human demonstrations offer rich environmental diversity and scale naturally, making them an appealing alternative to robot teleoperation. While this paradigm has advanced robot-arm manipulation, its potential for the more challenging, data-hungry problem of humanoid loco-manipulation remains largely unexplored. We present EgoHumanoid, the first framework to co-train a vision-language-action policy using abundant egocentric human demonstrations together with a limited amount of robot data, enabling humanoids to perform loco-manipulation across diverse real-world environments. To bridge the embodiment gap between humans and robots, including discrepancies in physical morphology and viewpoint, we introduce a systematic alignment pipeline spanning from hardware design to data processing. A portable system for scalable human data collection is developed, and we establish practical collection protocols to improve transferability. At the core of our human-to-humanoid alignment pipeline lies two key components. The view alignment reduces visual domain discrepancies caused by camera height and perspective variation. The action alignment maps human motions into a unified, kinematically feasible action space for humanoid control. Extensive real-world experiments demonstrate that incorporating robot-free egocentric data significantly outperforms robot-only baselines by 51%, particularly in unseen environments. Our analysis further reveals which behaviors transfer effectively and the potential for scaling human data.

Abstract:
Learning dexterous bimanual manipulation policies critically depends on large-scale, high-quality demonstrations, yet current paradigms face inherent trade-offs: teleoperation provides physically grounded data but is prohibitively labor-intensive, while simulation-based synthesis scales efficiently but suffers from sim-to-real gaps. We present BiDemoSyn, a framework that synthesizes contact-rich, physically feasible bimanual demonstrations from a single real-world example. The key idea is to decompose tasks into invariant coordination blocks and variable, object-dependent adjustments, then adapt them through vision-guided alignment and lightweight trajectory optimization. This enables the generation of thousands of diverse and feasible demonstrations within several hour, without repeated teleoperation or reliance on imperfect simulation. Across six dual-arm tasks, we show that policies trained on BiDemoSyn data generalize robustly to novel object poses and shapes, significantly outperforming recent strong baselines. Beyond the one-shot setting, BiDemoSyn naturally extends to few-shot-based synthesis, improving object-level diversity and out-of-distribution generalization while maintaining strong data efficiency. Moreover, policies trained on BiDemoSyn data exhibit zero-shot cross-embodiment transfer to new robotic platforms, enabled by object-centric observations and a simplified 6-DoF end-effector action representation that decouples policies from embodiment-specific dynamics. By bridging the gap between efficiency and real-world fidelity, BiDemoSyn provides a scalable path toward practical imitation learning for complex bimanual manipulation without compromising physical grounding.

Abstract:
Generalizing tool manipulation requires both semantic planning and precise physical control. Modern generalist robot policies, such as Vision-Language-Action (VLA) models, often lack the high-fidelity physical grounding required for contact-rich tool manipulation. Conversely, existing contact-aware policies that leverage tactile or haptic sensing are typically instance-specific and fail to generalize across diverse tool geometries. Bridging this gap requires learning unified contact representations from diverse data, yet a fundamental barrier remains: diverse real-world tactile data are prohibitive at scale, while direct zero-shot sim-to-real transfer is challenging due to the complex dynamics of nonlinear deformation of soft sensors. To address this, we propose Semantic-Contact Fields (SCFields), a unified 3D representation fusing visual semantics with dense contact estimates. We enable this via a two-stage Sim-to-Real Contact Learning Pipeline: first, we pre-train on a large simulation data set to learn general contact physics; second, we fine-tune on a small set of real data, pseudo-labeled via geometric heuristics and force optimization, to align sensor characteristics. This allows physical generalization to unseen tools. We leverage SCFields as the dense observation input for a diffusion policy to enable robust execution of contact-rich tool manipulation tasks. Experiments on scraping, crayon drawing, and peeling demonstrate robust category-level generalization, significantly outperforming vision-only and raw-tactile baselines.

Abstract:
Generalizing robotic manipulation across object poses, viewpoints, and dynamic disturbances is difficult, especially with only a few demonstrations. End-to-end visuomotor policies are expressive but data-hungry, while planning and optimization satisfy explicit constraints but do not directly capture the interaction strategies demonstrated by humans. We propose Sliding into Distribution (SID), a structured framework that learns an object-centric motion field from canonicalized demonstrations to iteratively slide the system toward the demonstrated manifold and into the reliable operating region of a lightweight egocentric execution policy, mitigating out-of-distribution (OOD) execution. The motion field provides large corrective motions when far from the demonstration manifold and naturally vanishes near convergence, enabling robust reaching under substantial pose and viewpoint shifts. Within the reached regime, an egocentric policy trained with conditioned flow matching performs task-specific manipulation, supported by kinematically consistent point-cloud reprojection augmentation that preserves action--observation consistency. Across six real-world tasks, SID achieves approximately 90% success under OOD initializations with only two demonstrations, with under a 10% drop under distractors and external disturbances. Overall, SID provides a new paradigm for few-shot manipulation: explicitly managing distribution shift via online distribution recovery.

Abstract:
Specifying robotic manipulation tasks in a manner that is both expressive and precise remains a central challenge. While visual goals provide a compact and unambiguous task specification, existing goal-conditioned policies often struggle with long-horizon manipulation due to their reliance on single-step action prediction without explicit modeling of task progress. We propose Act2Goal, a general goal-conditioned manipulation policy that integrates a goal-conditioned visual world model with multi-scale temporal control. Given a current observation and a target visual goal, the world model generates a plausible sequence of intermediate visual states that captures long-horizon structure. To translate this visual plan into robust execution, we introduce Multi-Scale Temporal Hashing, which decomposes the imagined trajectory into dense proximal frames for fine-grained closed-loop control and sparse distal frames that anchor global task consistency. The policy couples these representations with motor control through end-to-end cross-attention, enabling coherent long-horizon behavior while remaining reactive to local disturbances. Extensive experiments in both simulation and real-world environments demonstrate that Act2Goal, initialized from large-scale imitation learning, achieves strong performance across a wide range of long-horizon manipulation tasks. Beyond offline generalization, Act2Goal also supports reward-free online autonomous improvement via hindsight goal relabeling and LoRA-based finetuning, enabling rapid adaptation during deployment without external supervision. In real world experiments, it improves success rates from 30% to 90% on challenging out-of-distribution tasks within minutes of autonomous interaction, validating the effectiveness of online training as a powerful complement to offline learning.

Abstract:
While current humanoid whole-body control frameworks predominantly rely on the static environment assumptions, addressing tasks characterized by high dynamism and complex interactions presents a formidable challenge. In this paper, we address humanoid skateboarding, a highly challenging task requiring stable dynamic maneuvering on an underactuated wheeled platform. This integrated system is governed by non-holonomic constraints and tightly coupled human-object interactions. Successfully executing this task requires simultaneous mastery of hybrid contact dynamics and robust balance control on a mechanically coupled, dynamically unstable skateboard. To overcome the aforementioned challenges, we propose HUSKY, a learning-based framework that integrates humanoid-skateboard system modeling and physics-aware whole-body control. We first model the coupling relationship between board tilt and truck steering angles, enabling a principled analysis of system dynamics. Building upon this, HUSKY leverages Adversarial Motion Priors (AMP) to learn human-like pushing motions and employs a physics-guided, heading-oriented strategy for lean-to-steer behaviors. Moreover, a trajectory-guided mechanism ensures smooth and stable transitions between pushing and steering. Experimental results on the Unitree G1 humanoid platform demonstrate that our framework enables stable and agile maneuvering on a skateboard in real-world scenarios.

Abstract:
We present a framework for distributed Pose Graph Optimization (PGO) by formulating the problem as a second-order continuous-time dynamical system evolving on Lie groups. By modeling pose variables as massive particles subject to damping, the equilibrium points of the resulting Riemannian dynamics coincide with first-order critical points of the original PGO problem. Using the governing damped Euler-Poincare equations and a semi-implicit geometric integrator, we design an optimization algorithm that generalizes existing algorithms such as Riemannian gradient descent and Gauss-Newton. In multi-robot settings, we present a fully distributed and parallel method based on block-diagonal mass and damping matrices, where each robot solves an ordinary differential equation for its own poses with minimal communication overhead. Moreover, modeling both state and velocity enables principled neighbor prediction that significantly improves convergence under delayed communication. Theoretically, we present an analysis and establish sufficient condition that ensures energy dissipation under the employed geometric discretization scheme. Experiments on benchmark PGO datasets demonstrate that the proposed solver achieves superior performance compared to state-of-the-art distributed baselines in both synchronous and asynchronous regimes.

Abstract:
Realizing interactive whole-body control for multi-humanoid systems is critical for unlocking complex collaborative capabilities in shared environments. Although recent advancements have significantly enhanced the agility of individual robots, bridging the gap to physically coupled multi-humanoid interaction remains challenging, primarily due to severe kinematic mismatches and complex contact dynamics. To address this, we introduce Rhythm, the first unified framework enabling real-world deployment of dual-humanoid systems for complex, physically plausible interactions. Our framework integrates three core components: (1) an Interaction-Aware Motion Retargeting (IAMR) module that generates feasible humanoid interaction references from human data; (2) an Interaction-Guided Reinforcement Learning (IGRL) policy that masters coupled dynamics via graph-based rewards; and (3) a real-world deployment system that enables robust transfer of dual-humanoid interaction. Extensive experiments on physical Unitree G1 robots demonstrate that our framework achieves robust interactive whole-body control, successfully transferring diverse behaviors such as hugging and dancing from simulation to reality.

Abstract:
Multi-robot systems provide the parallelism and redundancy necessary for long-horizon tasks, while Large Language Models (LLMs) offer the reasoning capabilities to decompose these objectives into actionable plans. However, effectively grounding this high-level reasoning in physical multi-agent execution remains an open challenge. Existing LLM-based approaches fall mainly into two categories: Single-robot methods achieve robust contact-rich manipulation but lack the coordination mechanisms required for tasks spanning multiple workspaces. Current multi-robot frameworks focus on high-level planning, often treating manipulation as an idealized primitive that fails to account for real-world execution uncertainties. To address this, we propose a novel closed-loop agentic LLM-based framework to ensure robust multi-robot manipulation. Our system consists of three specialized agents: the Planning Agent decomposes instructions into allocated sub-tasks, the Manipulation Agent for each robot executes actions via adaptive tool use, and the Verification Agent closes the loop by monitoring physical outcomes and feeding back semantic corrections. Extensive real-world experiments demonstrate that our framework achieves superior success rates, ensures robust adaptability ranging from single to cross workspace manipulation, and offers a generalizable approach for diverse manipulation tasks.

Abstract:
Multi-Robot Motion Planning (MRMP) in continuous environments, where robots must generate dynamically feasible, collision-free trajectories, is challenging due to the combinatorial growth of the joint trajectory space and the difficulty of enforcing dynamic feasibility and hard safety constraints. Recent approaches recast trajectory planning as probabilistic inference, sampling from a posterior over trajectories using diffusion models whose score functions are learned from demonstration data. While showing promising performance, these approaches are limited: they often rely on sizable demonstration datasets and struggle to rigorously enforce dynamics and hard safety constraints during sampling. To this end, we introduce Model-Based Diffusion Optimal Control (MDOC), a provably safe model-based diffusion planner that efficiently produces dynamically feasible trajectories without relying on data. Crucially, we show that MDOC's safety mechanism--combining known dynamics models with Control Barrier Function (CBF)-constrained projections--naturally scales to multi-robot planning settings through Conflict-Based Search (CBS). Across simulation experiments, this integrated method consistently outperforms representative baseline planners in sample efficiency, geometric smoothness, and success rate, while reducing computation time and producing collision-free trajectories.

Abstract:
Neural implicit mapping has emerged as a powerful paradigm for robotic navigation and scene understanding. However, real-world robotic deployment requires continual adaptation to changing environments under strict memory and computation constraints, which existing mapping systems fail to support. Most prior methods rely on replaying historical observations to preserve consistency and assume static scenes. As a result, they cannot adapt to continual learning in dynamic robotic settings. To address these challenges, we propose TACO (TemporAl Consensus Optimization), a replay-free framework for continual neural mapping. We reformulate mapping as a temporal consensus optimization problem, where we treat past model snapshots as temporal neighbors. Intuitively, our approach resembles a model consulting its own past knowledge. We update the current map by enforcing weighted consensus with historical representations. Our method allows reliable past geometry to constrain optimization while permitting unreliable or outdated regions to be revised in response to new observations. TACO achieves a balance between memory efficiency and adaptability without storing or replaying previous data. Through extensive simulated and real-world experiments, we show that TACO robustly adapts to scene changes and consistently outperforms other continual learning baselines.

Abstract:
LiDAR-camera fusion enhances 3D panoptic segmentation by leveraging camera images to complement sparse LiDAR scans, but it also introduces a critical failure mode. Under adverse conditions, degradation or failure of the camera sensor can significantly compromise the reliability of the perception system. To address this problem, we introduce UP-Fuse, a novel uncertainty-aware fusion framework in the 2D range-view that remains robust under camera sensor degradation, calibration drift, and sensor failure. Raw LiDAR data is first projected into the range-view and encoded by a LiDAR encoder, while camera features are simultaneously extracted and projected into the same shared space. At its core, UP-Fuse employs an uncertainty-guided fusion module that dynamically modulates cross-modal interaction using predicted uncertainty maps. These maps are learned by quantifying representational divergence under diverse visual degradations, ensuring that only reliable visual cues influence the fused representation. The fused range-view features are decoded by a novel hybrid 2D-3D transformer that mitigates spatial ambiguities inherent to the 2D projection and directly predicts 3D panoptic segmentation masks. Extensive experiments on Panoptic nuScenes, SemanticKITTI, and our introduced Panoptic Waymo benchmark demonstrate the efficacy and robustness of UP-Fuse, which maintains strong performance even under severe visual corruption or misalignment, making it well suited for robotic perception in safety-critical settings.

Abstract:
Reliable odometry is essential for mobile robots as they increasingly enter more challenging environments, which often contain little information to constrain point cloud registration, resulting in degraded LiDAR–Inertial Odometry (LIO) accuracy or even divergence. To address this, we present BIEVR-LIO, a novel approach designed specifically to exploit subtle variations in the available geometry for improved robustness. We propose a high-resolution map representation that stores surfaces as compact voxel-wise oriented height images. This representation can directly be used for registration without calculation of intermediate geometric primitives while still supporting efficient updates. Since informative geometry is often sparsely distributed in the environment, we further propose a map-informed point sampling strategy that focuses registration on geometrically informative regions, improving robustness in uninformative environments while reducing computational cost. Extensive evaluation across multiple sensors, platforms, and environments demonstrates state-of-the-art performance in well-constrained scenes and substantial improvements in challenging scenarios where baseline methods diverge. Additionally, we demonstrate that the fine-grained geometry captured by BIEVR-LIO can be used for downstream tasks such as elevation mapping for robot locomotion.

Abstract:
Robotic navigation in human environments requires a spatio-temporal semantic representation that can reconcile open-vocabulary perception with long-term environmental changes. While foundation models provide strong zero-shot recognition, their predictions are intermittent and view-dependent, and naively integrating them into mapping pipelines leads to identity drift and stale semantics over time. We present SuperMap, a 4D spatio-temporal mapping framework for language-guided navigation that integrates high-frequency geometric SLAM with asynchronous open-vocabulary perception. Our core contribution is a consistency-driven mapping engine that combines 3D-aware instance association/re-activation with a principled existence-and-label confidence update to maintain stable object identities and prune outdated map content under occlusions and scene changes. SuperMap produces a queryable 4D scene-graph representation that interfaces naturally with Vision-Language Models by supporting compositional queries over object semantics, relations, and history. We demonstrate SuperMap on benchmarks and real robots, including dynamic scenes with appearance/disappearance and relocation, and provide ablations and runtime analysis. We release the full system as open-source to provide the community with a deployable baseline for open-vocabulary spatio-temporal mapping.

Abstract:
While Large Language Models (LLMs) and Vision-Language Models (VLMs) demonstrate remarkable capabilities in high-level reasoning and semantic understanding, applying them directly to contact-rich manipulation remains a challenge due to their lack of explicit physical grounding and inability to perform adaptive control. To bridge this gap, we propose CoRAL (Contact-Rich Adaptive LLM-based control), a modular framework that enables zero-shot planning by decoupling high-level reasoning from low-level control. Unlike black-box policies, CoRAL utilizes Large Language Models (LLMs) not as direct controllers, but as cost designers that synthesize context-aware objective functions for a sampling-based motion planner (MPPI). To address the ambiguity of physical parameters in visual data, we introduce a neuro-symbolic adaptation loop: a Vision-Language Model provides semantic priors for environmental dynamics (e.g., mass, friction estimates), which are then explicitly refined in real-time via online system identification, while the LLM iteratively modulates the cost function structure to correct strategic errors based on interaction feedback. Furthermore, a retrieval-based memory unit allows the system to reuse successful strategies across recurrent tasks. This hierarchical architecture ensures real-time control stability by decoupling high-level semantic reasoning from reactive execution, effectively bridging the gap between slow LLM inference and dynamic contact requirements. We validate CoRAL on both simulation and real-world hardware across challenging and novel tasks, such as "flipping objects against walls" leveraging extrinsic contacts. Experiments demonstrate that CoRAL outperforms state-of-the-art VLA and foundation-model-based planner baselines by boosting success rates over 50% on average in unseen contact-rich scenarios, effectively handling sim-to-real gaps through its adaptive physical understanding.

Abstract:
We present GHOST, a framework for learning visuomotor manipulation policies that generalize beyond the training distribution. GHOST factorizes control into (i) a high-level policy that predicts the next sub-goal as a distribution over 3D end-effector poses from multi-view RGB-D observations, and (ii) a low-level goal-conditioned controller that executes embodiment-specific actions. To condition image-based policies on 3D goals, we introduce a simple spatial interface that projects predicted goals into the image plane and represents them as end-effector heatmaps. Across a suite of manipulation tasks, this hierarchical factorization consistently improves performance and robustness compared to a flat Diffusion Policy. Further, we show that this hierarchical interface also makes it easy to incorporate human demonstrations without relying on (noisy) action retargeting. As sub-goals are largely embodiment-agnostic, we train the high-level policy on human video to specify how learned skills should be applied and composed, while keeping the low-level policy trained purely on robot data. This hierarchy enables adaptation to novel objects and task variations using a small number of human demonstrations.

Abstract:
Action chunking enables Vision Language Action (VLA) models to run in real time, but naive chunked execution often exhibits discontinuities at chunk boundaries. Real-Time Chunking (RTC) alleviates this issue but is external to the policy, leading to spurious multimodal switching and trajectories that are not intrinsically smooth. We propose Legato, a training-time continuation method for action-chunked flow-based VLA policies. Specifically, Legato initializes denoising from a schedule-shaped mixture of known actions and noise, exposing the model to partial action information. Moreover, Legato reshapes the learned flow dynamics to ensure that the denoising process remains consistent between training and inference under per-step guidance. Legato further uses randomized schedule condition during training to support varying inference delays and achieve controllable smoothness. Empirically, Legato produces smoother trajectories and reduces spurious multimodal switching during execution, leading to less hesitation and shorter task completion time. Extensive real-world experiments show that Legato consistently outperforms RTC across five manipulation tasks, achieving approximately 10% improvements in both trajectory smoothness and task completion time.

Abstract:
A significant challenge for robot learning research is our ability to accurately measure and compare the performance of robot policies. Benchmarking in robotics is historically challenging due to the stochasticity, reproducibility, and time-consuming nature of real-world rollouts. This challenge is exacerbated for recent generalist policies, which have to be evaluated across a wide variety of scenes and tasks. Evaluation in simulation offers a scalable complement to real world evaluations, but the visual and physical domain gap between existing simulation benchmarks and the real world has made them an unreliable signal for policy improvement. Furthermore, building realistic and diverse simulated environments has traditionally required significant human effort and expertise. To bridge the gap, we introduce Policy Evaluation and Environment Reconstruction in Simulation (PolaRiS), a scalable real-to-sim framework for high-fidelity simulated robot evaluation. PolaRiS utilizes neural reconstruction methods to turn short video scans of real-world scenes into interactive simulation environments. Additionally, we develop a simple simulation data co-training recipe that bridges remaining real-to-sim gaps and enables zero-shot evaluation in unseen simulation environments. Through extensive paired evaluations, across 600 real-world rollouts and over 93,000 simulation rollouts, we demonstrate that PolaRiS evaluations provide a much stronger correlation to real world, non-finetuned generalist policy ranking than existing simulated benchmarks. Its simplicity also enables rapid creation of diverse simulated environments. As such, this work takes a step towards distributed and democratized evaluation for the next generation of robotic foundation models.

Abstract:
This paper develops LongNav-R1, an end-to-end multi-turn reinforcement learning (RL) framework designed to optimize Visual-Language-Action (VLA) models for long-horizon navigation. Unlike existing single-turn paradigm, LongNav-R1 reformulates the navigation decision process as a continuous multi-turn conversation between the VLA policy and the embodied environment. This multi-turn RL framework offers two distinct advantages: i) it enables the agent to reason about the causal effects of historical interactions and sequential future outcomes; and ii) it allows the model to learn directly from online interactions, fostering diverse trajectory generation and avoiding the behavioral rigidity often imposed by human demonstrations. Furthermore, we introduce Horizon-Adaptive Policy Optimization. This mechanism explicitly accounts for varying horizon lengths during advantage estimation, facilitating accurate temporal credit assignment over extended sequences. Consequently, the agent develops diverse navigation behaviors and resists collapse during long-horizon tasks. Experiments on object navigation benchmarks validate the framework's efficacy: With 4,000 rollout trajectories, LongNav-R1 boosts the Qwen3-VL-2B success rate from 64.3% to 73.0%. These results demonstrate superior sample efficiency and significantly outperform state-of-the-art methods. The model's generalizability and robustness are further validated by its zero-shot performance in long-horizon real-world navigation settings. All source code will be open-sourced upon publication.

Abstract:
Diffusion policies have recently emerged as a powerful framework for robotic manipulation. However, like other behavior cloning methods, they remain vulnerable to distributional shift, often requiring human-in-the-loop interventions to correct failures during deployment. These interactions naturally provide paired supervision in the form of the robot’s undesired actions and the human teacher’s corrective actions. Yet existing data aggregation pipelines and standard behavior cloning losses largely ignore this negative signal from undesired actions, leading to overfitting to teacher's actions and an increasing reliance on costly expert data. To address this limitation, we propose Set-Supervised Diffusion Policy (SDP), a novel learning framework that utilizes contrastive action-chunk data to train diffusion policies from human corrections. From paired positive and negative action-chunks, SDP constructs a set of desired action-chunks and designs a training pipeline that encourages the diffusion policy to align with the set. Through extensive experiments across multiple robotic manipulation tasks, we demonstrate that SDP consistently improves policy performance, with particularly strong gains in robustness to noisy data. Moreover, SDP induces high-quality aggregated datasets, enabling more efficient and reliable policy learning from human-in-the-loop corrections. Our code is available at https://set-supervised-diffusion-policy.github.io/.

Abstract:
Learning universal policies from cross-embodied data remains a fundamental challenge in robotics. Although Vision-Language-Action (VLA) models are pre-trained on large and diverse datasets, they typically rely on embodiment-specific fine-tuning to achieve strong performance in downstream tasks. This requirement severely limits their generalization capability and restricts knowledge transfer across embodiments performing similar tasks. To overcome these limitations, we focus on cross-embodied settings with shared robotic bases and heterogeneous end-effectors, and propose X-DiffVLA, a diffusion-based VLA model featuring a unified cross-embodied action head. X-DiffVLA can leverage the generative strengths of diffusion models to capture both the diversity and latent correlations in cross-embodied datasets. Specifically, we introduce Embodiment Forcing, a classifier-free guidance technique to implicitly steer action generation toward embodiment-specific functional components, capturing fine-grained structural nuances without explicit supervision. In addition, a Morphological Tree Diffusion approach is designed to strengthen behavioral correlations across diverse end-effectors, maximizing the transferability of heterogeneous demonstrations. Experimental results across RoboCasa and Isaac Gym, covering different embodiments from grippers to dexterous hands, show that X-DiffVLA achieves state-of-the-art performance, with improvements of 15.3% and 12.5%, respectively. Real-world evaluations further validate the robustness of the proposed framework and its effectiveness in scalable cross-embodied policy learning.

Abstract:
Equipping embodied agents with the ability to reason about tasks, foresee physical outcomes, and generate precise actions is essential for general-purpose manipulation. While recent Vision-Language-Action (VLA) models have leveraged pre-trained foundation models, they typically focus on either linguistic planning or visual forecasting in isolation. These methods rarely integrate both capabilities simultaneously to guide action generation, leading to suboptimal performance in complex, long-horizon manipulation tasks. To bridge this gap, we propose BagelVLA, a unified model that integrates linguistic planning, visual forecasting, and action generation within a single framework. Initialized from a pretrained unified understanding and generative model, BagelVLA is trained to interleave textual reasoning and visual prediction directly into the action execution loop. To efficiently couple these modalities, we introduce Residual Flow Guidance (RFG), which initializes from current observation and leverages single-step denoising to extract predictive visual features, guiding action generation with minimal latency. Extensive experiments demonstrate that BagelVLA outperforms existing baselines by a significant margin on multiple simulated and real-world benchmarks, particularly in tasks requiring multi-stage reasoning.

Abstract:
Discrete-time joint acceleration bounds are widely used to enforce position and velocity limits. However, under voltage-constrained electric actuators, kinematically admissible accelerations may be physically unrealizable, exposing a missing execution-level abstraction. We propose Actuator-Aware Joint Acceleration Control (AJAC), a joint-level realizability contract that grounds kinematic acceleration bounds in voltage-constrained actuator physics by operating accelerations. Hardware experiments on electric actuators and a wheel-legged quadruped show that AJAC removes unrealizable accelerations, restores consistent near-boundary execution, and reduces boundary-induced oscillations.

Abstract:
Dexterous manipulation is a cornerstone capability for robotic systems aiming to interact with the physical world in a human-like manner. Although vision-based methods have advanced rapidly, tactile sensing remains crucial for fine-grained control—particularly in unstructured or visually occluded settings. We present ViTacFormer, a representation-learning approach that couples a cross-attention encoder to fuse high-resolution vision and touch with an autoregressive tactile-prediction head that anticipates future contact signals. Building on this architecture, we devise an easy-to-challenging curriculum that steadily refines the visual-tactile latent space, boosting both accuracy and robustness. The learned cross-modal representation drives imitation learning for multi-fingered hands, enabling precise and adaptive manipulation. Across a suite of challenging real-world benchmarks, our method achieves approximately 50% higher success rates than prior state-of-the-art systems. To our knowledge, it is also the first to autonomously complete long-horizon dexterous manipulation tasks that demand highly precise control with an anthropomorphic hand—successfully executing up to 11 sequential stages and sustaining continuous operation for 2.5 minutes.

Abstract:
The pursuit of general-purpose embodied agents is currently hindered by fragmented evaluation protocols that isolate navigation skills and fixate on specific robot morphologies. This disconnect fails to reflect real-world scenarios where agents must orchestrate diverse behaviors across varying physical embodiments. To bridge this gap, we introduce OmniNavBench, a holistic benchmark designed to rigorously assess cross-skill coordination and cross-embodiment generalization. Distinguished from existing datasets, OmniNavBench introduces three paradigm shifts: (1) Compositional Complexity. We propose composite instructions that interleave sub-tasks from 6 distinct categories (i.e., PointNav, VLN, ObjectNav, SocialNav, Human Following and EQA), compelling agents to seamlessly transition between exploration, interaction, and social compliance within a single unified episode. (2) Morphological Universality and Sensor Flexibility. We present a simulation platform that breaks the reliance on single-morphology evaluation. This ecosystem empowers researchers to test generalization across different robot types, including humanoid, quadrupedal, and wheeled, while accommodating diverse algorithmic needs through a modular sensor interface and a hybrid suite of 170 environments blending synthetic assets with real-world scans. (3) Demonstrations Quality. Moving beyond mechanical shortest-path algorithms, we curate 1,769 expert trajectories via human teleoperation, capturing critical behavioral nuances, such as exploratory glance and anticipatory avoidance, essential for natural human-robot coexistence. Extensive evaluations demonstrate that current methods, despite their claimed unified design, struggle to adapt to the complex, interleaved nature of truly general-purpose navigation. This exposes a critical disparity between existing capabilities and the demands of real-world deployment, underscoring OmniNavBench as a crucial testbed for the next generation of generalist navigators. Dataset will be released.

Abstract:
Multi-objective reinforcement learning in robotic domains requires balancing complex, non-convex trade-offs between conflicting objectives. While linear scalarization methods provide stability, they are theoretically incapable of recovering solutions within non-convex regions of the Pareto front. Conversely, static non-linear scalarizations (e.g., Tchebycheff) can theoretically access these regions but often suffer from severe gradient variance and optimization instability in deep RL. In this work, we propose an Adaptive Smooth Tchebycheff framework that resolves this tension by dynamically modulating the curvature of the optimization landscape. We introduce a novel conflict-driven controller that regulates the optimization smoothness based on real-time gradient interference. This allows the agent to anneal toward precise, non-convex scalarization when objectives align, while elastically reverting to stable, smooth approximations when destructive gradient conflicts emerge. We validate our approach on a challenging robotic stealth visual search task---a proxy for monitoring of protected/fragile ecosystems---where an agent must balance search, exposure/interference minimization and exploration speed. Extensive ablations confirm that our conflict-aware adaptation enables the robust discovery of Pareto-optimal policies in non-convex regions inaccessible to linear baselines and unstable for static non-linear methods.

Abstract:
Learning-based controllers have achieved impressive performance in agile quadrotor flight but typically rely on massive training in simulation, necessitating accurate system identification for effective Sim2Real transfer. However, even with precise modeling, fixed policies remain susceptible to out-of-distribution scenarios, ranging from external aerodynamic disturbances to internal hardware degradation. To ensure safety under these evolving uncertainties, such controllers are forced to operate with conservative safety margins, inherently constraining their agility outside of controlled settings. While online adaptation offers a potential remedy, safely exploring physical limits remains a critical bottleneck due to data scarcity and safety risks. To bridge this gap, we propose a self-adaptive framework that eliminates the need for precise system identification or offline Sim2Real transfer. We introduce Adaptive Temporal Scaling (ATS) to actively explore platform physical limits, and employ online residual learning to augment a simple nominal model. Based on the learned hybrid model, we further propose Real-world Anchored Short-horizon Backpropagation Through Time (RASH-BPTT) to achieve efficient and robust in-flight policy updates. Extensive experiments demonstrate that our quadrotor reliably executes agile maneuvers near actuator saturation limits. The system evolves a conservative base policy with a peak speed of 1.9 m/s to 7.3 m/s within approximately 100 seconds of flight time. These findings underscore that real-world adaptation serves not merely to compensate for modeling errors, but as a practical mechanism for sustained performance improvement in aggressive flight regimes.

Abstract:
Autonomous robotic exploration in remote and extreme environments allows scientists to model complex transport phenomena and collective behaviors described by continuously deforming flow fields. Although these environments are naturally modeled as time-varying domains, most adaptive exploration methods assume static environments and fail to provide adequate coverage, let alone satisfy any formal guarantees. This is especially the case in oceanography where autonomous uncrewed underwater systems (UxS) have highly restrictive compute and payload requirements that necessitate path planning methods that yield robust data collection strategies in open-loop and underactuated settings. In this work, to address the aforementioned issues, we propose to formulate adaptive search as an ergodic coverage problem and investigate certifying coverage in the ergodic sense over evolving domains with flow-induced dynamics. We expand upon recent work demonstrating maximum mean discrepancy (MMD) as a functional ergodic metric, and derive a flow-adaptive formulation that explicitly accounts for domain evolution within the coverage objective. We show that this approach preserves ergodic coverage guarantees in ambient flows and enables effective exploration in under-actuated, and even open-loop planning settings by integrating environment dynamics. Experiments validate that our method generalizes to diverse spatiotemporal processes including ocean exploration, and tracking human and cattle movement. Physical experiments on aerial and legged robotic platforms validate our ability to obtain ergodic coverage in non-convex, flow-restricted environments while respecting robot dynamics.

Abstract:
In-hand object reorientation requires precise estimation of the object pose to handle complex task dynamics. While RGB sensing offers rich semantic cues for pose tracking, existing solutions rely on multi-camera setups or costly ray tracing. We present a sim-to-real framework for monocular RGB in-hand reorientation that integrates 3D Gaussian Splatting (3DGS) to bridge the visual sim-to-real gap. Our key insight is performing domain randomization in the Gaussian representation space: by applying physically consistent, pre-rendering augmentations to 3D Gaussians, we generate photorealistic, randomized visual data for object pose estimation. The manipulation policy is trained using curriculum-based reinforcement learning with teacher–student distillation, enabling efficient learning of complex behaviors. Importantly, both perception and control models can be trained independently on consumer-grade hardware, eliminating the need for large compute clusters. Experiments show that the pose estimator trained with 3DGS data outperforms those trained using conventional rendering data in challenging visual environments. We validate the system on a physical multi-fingered hand equipped with an RGB camera, demonstrating robust reorientation of five diverse objects even under challenging lighting conditions. Our results highlight Gaussian splatting as a practical path for RGB-only dexterous manipulation.

Abstract:
Zero-order optimization has recently received significant attention for designing optimal trajectories and policies for robotic systems. However, most existing methods (e.g., MPPI, CEM, and CMA-ES) are local in nature, as they rely on gradient estimation. In this paper, we introduce consensus-based optimization (CBO) to robotics, which is guaranteed to converge to a global optimum under mild assumptions. We provide theoretical analysis and illustrative examples that give intuition into the fundamental differences between CBO and existing methods. To demonstrate the scalability of CBO for robotics problems, we consider three challenging trajectory optimization scenarios: (1) a long-horizon problem for a simple system, (2) a dynamic balance problem for a highly underactuated system, and (3) a high-dimensional problem with only a terminal cost. Our results show that CBO is able to achieve lower costs with respect to existing methods on all three challenging settings. This opens a new framework to study global trajectory optimization in robotics.

Abstract:
A central challenge of visual control with model-based reinforcement learning (RL) is reliable long-horizon planning: Long rollouts with learned latent dynamics exhibit branching futures and multi-modal action-value distributions. In addition, compounding model error amplified by visual occlusions make deep imagination brittle. We present ELVIS, a latent model predictive controller (MPC) designed to make long-horizon planning practical. ELVIS plans in a Dreamer-style recurrent state space model (RSSM) and replaces standard unimodal model predictive path integral (MPPI) with a Gaussian-mixture MPPI that maintains multiple coherent hypotheses over long horizons, avoiding mode averaging under branching rollouts. In parallel, ELVIS stabilizes deep imagination with a shared uncertainty-aware λ_t-return: an ensemble of latent critics defines an upper-confidence-bound (UCB) score that gates a time-varying λ_t, adaptively trading off bootstrapping versus look-ahead to limit compounding error during planning. The same return is used both to train an actor-critic prior from imagined rollouts and to score candidate trajectories inside GMM-MPPI, aligning RL objectives with the planner’s long-horizon optimization. On fourteen DeepMind Control Suite visual tasks, ELVIS establishes state-of-the-art performance compared with TD-MPC2 and DreamerV3. Finally, ELVIS transfers zero-shot to a real-world sand spraying task with severe occlusions, improving surface-quality metrics and demonstrating robustness beyond simulation.

Abstract:
Robotic systems with redundant degrees of freedom can achieve the same task outcome using multiple configurations, resulting in solution sets that form manifolds in the configuration space. Existing approaches typically exploit such redundancy locally through Jacobian-based techniques to compute individual solutions or trajectories. While effective for solution computation, these methods do not retain a representation of the geometry of the solution set itself. In this work, we adopt a representation-centric approach to estimate the geometric structure of the solution space. We consider solution manifolds induced by general task-defining maps and construct an implicit scalar field over the configuration space, whose zero-level set corresponds to the solution manifold. To this end, we generate samples in the neighborhood of the solution manifold using a Jacobian-guided exploration strategy, which efficiently captures its local and global structure. The resulting implicit representation is defined over the configuration space and naturally induces a continuous, distance field that encodes proximity to the solution manifold. Experiments on a planar three-link robot and a seven-degree-of-freedom Franka manipulator demonstrate the effectiveness of the proposed representation. Furthermore, the framework enables consistent modeling of solution spaces across families of tasks with continuous variation.

Abstract:
Follow-the-leader (FTL) motion exploits the unique morphology of continuum robots (CRs) to navigate confined spaces by having the body retrace the path of the tip. While extensively studied, existing FTL methods typically assume a fixed base or a single degree-of-freedom insertion mechanism, limiting their applicability to practical systems in which CRs are mounted on robotic manipulators with full six-degree-of-freedom pose control. This paper presents a sampling-based motion planner for FTL motion of manipulator-mounted CRs that jointly considers robot configuration and base pose. The key idea is to decouple global shape search from base pose determination by computing the base pose through a closed-form geometric construction, thereby avoiding iterative optimization during online planning. The approach supports general forward models and enables efficient planning by shifting the majority of computation offline. We establish theoretical guarantees including resolution completeness of the shape search and exact tip tracking throughout waypoint traversal and interpolation. Experiments on 120 simulated paths over 3 test classes demonstrate 0% tip error and 1.9% mean shape deviation (w.r.t. robot length) at 100% success rate. We validate the practicality of our approach on a 6-DOF tendon-driven CR mounted on a serial manipulator.

Abstract:
While gradient-based methods can efficiently optimize trajectories and controllers by exploiting physical priors and differentiable simulators, contact-rich manipulation remains challenging due to discontinuous and vanishing gradients arising from hybrid contact dynamics. Recent methods smooth the dynamics to obtain continuous gradients, but the resulting model mismatch often causes controller failures when executed on real systems. We address this trade-off by developing a framework that plans efficiently with smoothed dynamics while explicitly quantifying and compensating for the induced modeling errors. Our method provides formal guarantees of constraint satisfaction and goal reachability on the true hybrid dynamics, enabling robust gradient-based synthesis for contact-rich manipulation. Specifically, we construct smooth approximations of both system dynamics and contact geometry to obtain a well-conditioned optimization landscape, and characterize the discrepancy from the true dynamics as a set-valued deviation. This deviation is incorporated into the optimization of time-varying affine feedback policies that admit analytical predictions of true system behavior under feedback, while relying solely on informative gradients from the smoothed dynamics. We evaluate our method on several contact-rich tasks, including planar pushing, object rotation, and in-hand dexterous manipulation, achieving guaranteed constraint satisfaction with lower goal error than baselines. By bridging differentiable physics with set-valued robust control, our method is the first certifiable gradient-based policy synthesis method for contact-rich manipulation.

Abstract:
We introduce CRAFT Hand, a tendon-driven anthropomorphic hand with hybrid hard-soft compliance for contact-rich manipulation. The design is based on a simple idea: contact is not uniform across the hand. Impacts concentrate at joints, while links carry most of the load. CRAFT places soft material at joints and keeps links rigid, and uses rolling-contact joint surfaces to keep flexion on repeatable motion paths. Fifteen motors mounted off the fingers drive the hand through tendons, keeping the form factor compact and the fingers light. In structural tests, CRAFT improves strength and endurance while maintaining comparable repeatability. In teleoperation, CRAFT improves handling of fragile and low-friction items, and the hand covers 33/33 grasps in the Feix taxonomy. The full design costs under 600 and will be released open-source with vision-based teleoperation and simulation integration.

Abstract:
Accurate simulation of tendon-driven continuum robots (TDCRs) remains challenging due to their continuous deformation, complex tendon actuation, and strong nonlinearity. One valid approach is learning from real-world data. In this paper, we present a lightweight, low-cost 3D-printed TDCR hardware platform, along with a task-agnostic self-modeling pipeline for learning its kinematic behavior. We employ a point-cloud flow-matching model that learns the robot’s kinematics from randomly sampled kinematic states, capturing the relationship between tendon actuation and the resulting deformation. We evaluate our method via motion prediction experiments, comparing against prior 3D deformable object modeling approaches in both synthetic and real-world settings. The results demonstrate improved accuracy in predicting robot shapes given motor configurations, highlighting the effectiveness of the proposed self-modeling framework for continuum robots.

Abstract:
We present Whole-Body Mobile Manipulation Interface (HoMMI), a data collection and policy learning framework that learns whole-body mobile manipulation directly from robot-free human demonstrations. We augment UMI interfaces with egocentric sensing to capture the global context required for mobile manipulation, enabling portable, robot-free, and scalable data collection. However, naively incorporating egocentric sensing introduces a larger human-to-robot embodiment gap in both observation and action spaces, making policy transfer difficult. We explicitly bridge this gap with a cross-embodiment hand-eye policy design, including an embodiment agnostic visual representation; a relaxed head action representation; and a whole-body controller that realizes hand-eye trajectories through coordinated whole-body motion under robot-specific physical constraints. Together, these enable long-horizon mobile manipulation tasks requiring bimanual and whole-body coordination, navigation, and active perception. All code, data, and hardware design will be publicly available.

Abstract:
Recent imitation learning (IL) algorithms -- such as flow-matching and diffusion policies -- demonstrate remarkable performance in learning complex manipulation tasks. However, these policies often fail even when operating within their training distribution due to extreme sensitivity to initial conditions and irreducible approximation errors that lead to compounding drift. This makes it unsafe to deploy IL policies in the field where out-of-distribution scenarios are prevalent. A prerequisite for safe deployment is enabling the policy to determine whether it can execute a task the way it was learned from demonstrations. This paper presents a principled approach to identify, for a trained IL policy, a safe set from where the policy is guaranteed to succeed in completing the learned task. We propose a Lipschitz-continuous Q-value function that maps state-action pairs to a safety score based on three task-agnostic criteria: visibility, recognizability, and graspability. The zero-superlevel set of this function defines a Control Invariant Set over state-action pairs. When the nominal policy proposes an action outside this set, we leverage Nagumo's theorem to compute a recovery action via gradient ascent on the Q-function, steering the policy back to safety. To learn this Q-function, we construct a photorealistic digital twin using Gaussian Splatting that enables systematic collection of failure data without risk to physical hardware. Experiments with a Franka Emika robot demonstrate that flow-matching policies, which fail under run-time perturbations, achieve consistent task success when guided by the proposed safety watchdog.

Abstract:
Data scarcity fundamentally limits the generalization of bimanual dexterous manipulation, as real-world data collection for dexterous hands is expensive and labor-intensive. Human manipulation videos, as a direct carrier of manipulation knowledge, offer significant potential for scaling up robot learning. However, the substantial embodiment gap between human hands and robotic dexterous hands makes direct pretraining from human videos extremely challenging. To bridge this gap and unleash the potential of large-scale human manipulation video data, we propose DexImit, an automated framework that converts monocular human manipulation videos into physically plausible robot data, without any additional information. DexImit employs a four-stage generation pipeline: (1) reconstructing hand–object interactions from arbitrary viewpoints with near-metric scale; (2) performing subtask decomposition and bimanual scheduling; (3) synthesizing robot trajectories consistent with the demonstrated interactions; (4) comprehensive data augmentation for zero-shot real-world deployment. Building on these designs, DexImit can generate large-scale robot data based on human videos, either from the Internet or video generation models. DexImit is capable of handling diverse manipulation tasks, including tool use (e.g., cutting an apple), long-horizon tasks (e.g., making a beverage), and fine-grained manipulations (e.g., stacking cups).

Abstract:
Contact-rich dexterous manipulation with multi-finger hands remains an open challenge in robotics because task success depends on multi-point contacts that continuously evolve and are highly sensitive to object geometry, frictional transitions, and slip. Recently, tactile-informed manipulation policies have shown promise. However, most use tactile signals as additional observations rather than modeling contact state or how their action outputs interact with low-level controller dynamics. We present Contact-Grounded Policy (CGP), a visuotactile policy that grounds multi-point contacts by predicting coupled trajectories of actual robot state and tactile feedback, and using a learned contact-consistency mapping to convert these predictions into executable target robot states for a compliance controller. CGP consists of two components: (i) a conditional diffusion model that forecasts future robot state and tactile feedback in a compressed latent space, and (ii) a learned contact-consistency mapping that converts the predicted robot state-tactile pair into executable targets for a compliance controller, enabling it to realize the intended contacts. We evaluate CGP using a physical four-finger Allegro V5 hand with Digit360 fingertip tactile sensors, and a simulated five-finger Tesollo DG-5F hand with dense whole-hand tactile arrays. Across a range of dexterous tasks including in-hand manipulation, delicate grasping, and tool use, CGP outperforms visuomotor and visuotactile diffusion-policy baselines.

Abstract:
General-purpose robots operating in partially observable environments such as homes require memory to support long-term autonomy. They must recall different types of past information, such as where objects were placed, which subtasks have already been completed by a human partner, and when an appliance was turned on This capability requires an effective memory retrieval mechanism. However, hand-designed or heuristic-based retrieval methods often fail to generalize in different tasks. Attention-based retrieval provides a promising alternative, as both queries and keys are learned from data without making task-specific assumptions. However, directly applying an attention-based memory retrieval mechanism in imitation learning introduces two key challenges: (1) the policy may learn spurious correlations between the information retrieved from the past and predicted actions, and (2) errors accumulated over time in the memory due to prediction inaccuracies, compounded by interactions with the environment, lead to model drift and cascading failures in long-horizon control. To address these challenges, we introduce HALO, a visuomotor policy equipped with an attention-based memory retrieval mechanism for long-horizon control. To mitigate spurious correlations, HALO leverages vision-language models (VLMs) by generating task-relevant question–answer pairs from demonstration trajectories and jointly training the policy with a video question– answering objective. This supervision encourages the retrieval module to focus on information that is relevant to the task. Second, to reduce the impact of accumulated errors in memory during closed-loop control, HALO uses sparse attention that restricts retrieval to only the most relevant parts of the history. Together, these components enable more reliable long-horizon control by guiding the policy to retrieve task-relevant information from up to two minutes of past experience.

Abstract:
Action-conditioned video prediction models (often referred to as world models) have shown strong potential for robotics applications, but existing world models are often slow and struggle to capture accurate physical interactions over long horizons, limiting their use for scalable robot policy training and evaluation. We present Interactive World Simulator, a framework that builds interactive world models using a moderate-sized robot interaction dataset and enables scalable robot policy training and evaluation. In our experiments, we show that our world models 1) produce physically accurate pixel-level predictions, and 2) support stable long-horizon interactions for more than 10 minutes at 15 FPS on a single RTX 4090 GPU. Our framework uses data collected through interaction with the world models to train imitation policies, including Diffusion Policy, Action Chunking Transformer, and π-series policies. Through extensive real-world evaluation across diverse tasks involving rigid objects, deformable objects, object piles, and their interactions, we find that imitation policies trained on world-model-generated data perform comparably to policies trained on the same amount of real-world data. Additionally, we evaluate policies both within the world models and in the real world across diverse tasks, and observe a strong correlation between performance in the world models and in the real world. Together, these results establish the Interactive World Simulator as a stable and physically consistent surrogate that enables scalable robotic data generation and faithful, reproducible policy evaluation.

Abstract:
We introduce Ψ₀ ( Psi-Zero ), an open foundation model to address challenging humanoid loco-manipulation tasks. While existing approaches often attempt to address this fundamental problem by co-training on large and diverse human and humanoid data, we argue that this strategy is suboptimal due to the fundamental kinematic and motion disparities between humans and humanoid robots. Therefore, data efficiency and model performance remain unsatisfactory despite the considerable data volume. To address this challenge, Ψ₀ decouples the learning process to maximize the utility of heterogeneous data sources. Specifically, we propose a staged training paradigm with different learning objectives: First, we autoregressively pre-train a VLM backbone on large-scale egocentric human videos to acquire generalizable visual-action representations. Then, we post-train a flow-based action expert on high-quality humanoid robot data to learn precise robot joint control. Our research further identifies a critical yet often overlooked data recipe: in contrast to approaches that scale with noisy Internet clips or heterogeneous cross-embodiment robot datasets, we demonstrate that pre-training on high-quality egocentric human manipulation data followed by post-training on domain-specific real-world humanoid trajectories yields superior performance. Extensive real-world experiments demonstrate that \ours\ achieves the best performance using only about 800 hours of human video data and 30 hours of real-world robot data, outperforming baselines pre-trained on more than 10× as much data by over 40% in overall success rate across multiple real and simulation tasks. We will open-source the entire ecosystem to the community, including a data processing and training pipeline, a humanoid foundation model, and a real-time action inference engine.

Abstract:
Humanoid loco-manipulation requires executing precise manipulation tasks while maintaining dynamic stability amid base motion and impacts. Existing approaches typically formulate commands in body-centric frames, fail to inherently correct cumulative world-frame drift induced by legged locomotion. We reformulate the problem as world-frame end-effector tracking and propose HiWET, a hierarchical reinforcement learning framework that decouples global reasoning from dynamic execution. The high-level policy generates subgoals that jointly optimize end-effector accuracy and base positioning in the world frame, while the low-level policy executes these commands under stability constraints. We introduce a Kinematic Manifold Prior (KMP) that embeds the manipulation manifold into the action space via residual learning, reducing exploration dimensionality and mitigating kinematically invalid behaviors. Extensive simulation and ablation studies demonstrate that HiWET achieves precise and stable end-effector tracking in long-horizon world-frame tasks. We validate zero-shot sim-to-real transfer of the low-level policy on a physical humanoid, demonstrating stable locomotion under diverse manipulation commands. These results indicate that explicit world-frame reasoning combined with hierarchical control provides an effective and scalable solution for long-horizon humanoid loco-manipulation.

Abstract:
High-fidelity motion tracking serves as the ultimate litmus test for generalizable, human-level motor skills. However, current policies often hit a "generality barrier": as motion libraries scale in diversity, tracking fidelity inevitably collapses—especially for real-world deployment of high-dynamic motions. We identify this failure as the result of two compounding factors: the learning bottleneck in scaling multi-motion optimization and the physical executability constraints that arise in real-world actuation. To overcome these, we introduce OmniXtreme, a scalable framework that decouples general motor skill learning from sim-to-real physical skill refinement. Our approach uses a flow-matching policy with high-capacity architectures to scale representation capacity without the interference-intensive multi-motion RL optimization, followed by an actuation-aware refinement phase that ensures robust performance on physical hardware. Extensive experiments demonstrate that OmniXtreme maintains high-fidelity tracking across diverse, high-difficulty datasets. On real robots, the unified policy successfully executes multiple extreme motions, effectively breaking the long-standing fidelity–scalability trade-off in high-dynamic humanoid control.

Abstract:
Progress in robot learning, particularly physical multi-agent reinforcement learning (MARL), is currently challenged by a limited availability of accessible, standardized benchmarks. While simulation-based MARL has produced remarkable emergent behaviors from coordinated team play to complex tool use, translating these advances to physical systems remains difficult. A key barrier is infrastructural: physical platforms are often prohibitively expensive, require specialized facilities, or lack open mechanisms and designs for reproducibility. We introduce KlaskTron, an open-source, low-cost (<2500 USD), desktop-scale robotic testbed for adversarial MARL based on the dynamic tabletop game KLASK. The platform features dual CoreXY gantries with high-torque brushless motors, enabling the high-speed, precise manipulation required for competitive play. Crucially, we provide a complete ecosystem: hardware designs (CAD, BOM), a GPU-native digital twin in NVIDIA Isaac Lab for high-fidelity modeling and simulation, and a validated sim-to-real baseline using a learned neural actuator model. We demonstrate that this baseline enables zero-shot policy transfer, with trained agents exhibiting emergent adversarial behaviors. To ensure full reproducibility of our results, all project assets are released publicly: documentation at [Anonymized Link], hardware CAD/BOM at [Anonymized Link], and software/simulation at [Anonymized Link].

Abstract:
In large-scale heterogeneous robot systems (HRS), scheduling efficiency in terms of throughput and makespan relies on exploiting parallel execution, while human-issued safety and precedence instructions impose rigid temporal-logic constraints that create severe combinatorial complexity and challenge traditional optimization and metaheuristic methods. We propose Asynchronous Logic-Induced Sleep-Wake Coordination (ALIS-WC), an event-driven reinforcement learning (RL) framework that performs sleep-wake scheduling of heterogeneous robots under Linear Temporal Logic (LTL) constraints. At its core, ALIS-WC applies a two-stage filtering mechanism: it first filters out robots that are physically incapable of contributing to the current decision, and then uses temporal-logic dependencies to label the remaining robots as sleeping or eligible, so that the learned dispatch policy only coordinates among idle, constraint-compliant robots at each decision event. We first design a human-ALIS-WC interaction interface: a lightweight language front-end that maps natural-language (NL) mission descriptions into LTL safety and precedence clauses for the scheduler using a compact 8B translator that achieves high NL-to-LTL accuracy. We further design a potential-based reward-shaping term over normalized global mission progress that provides a difficulty-invariant learning signal and accelerates policy learning without changing the underlying objective. Experiments on large-scale graph-based benchmarks with dozens of heterogeneous robots, 100 tasks, and up to 10 LTL clauses show that ALIS-WC improves makespan and task completion over strong metaheuristic, search-based, and reinforcement-learning baselines while maintaining 100% satisfaction of all enforced temporal-logic constraints. Additional stress tests with up to 60 clauses indicate that ALIS-WC remains effective under much denser temporal-logic specifications.

Abstract:
We study cooperative shortest path planning for an unmanned ground vehicle (UGV) assisted by an unmanned aerial vehicle (UAV) in environments with unknown road blockages that are only discovered when a robot reaches the damaged point. This formulation generalizes the original Canadian Traveller Problem (CTP), which assumes a single ground vehicle and that the traversability status of all incident edges is revealed upon arrival at a vertex. We first analyze the case where the start and the goal are connected by k disjoint paths, and prove that the worst-case competitive ratio ρ for a single UGV is 2k-1. With UAV assistance, and under the simplifying assumption of negligible initial transit and deadheading UAV costs, the ratio improves to ρ = 2\fracv_Gv_A + v_Gk - 1, where v_G and v_A denote the UGV and UAV speed, respectively. To address general graphs and non-negligible UAV initial transit and deadheading costs, we present an optimal path partitioning strategy that assigns path prefix inspection to the UGV and path suffix inspection to the UAV, and prove the optimality of the UAV inspection strategy on general graphs. We evaluate our algorithm by performing experiments on road networks from the world's 50 most populous cities, with randomized blockages, and show that the proposed method reduces UGV travel times by up to 30%.

Abstract:
3D spatial perception is fundamental to generalizable robotic manipulation, yet obtaining reliable, high-quality 3D geometry remains challenging. Depth sensors suffer from noise and material sensitivity, while existing reconstruction models lack the precision and metric consistency required for physical interaction. We introduce Robo3R, a feed-forward, manipulation-ready 3D reconstruction model that predicts accurate, metric-scale scene geometry directly from RGB images and robot states in real time. Robo3R jointly infers scale-invariant local geometry and relative camera poses, which are unified into the scene representation in the canonical robot frame via a learned global similarity transformation. To meet the precision demands of manipulation, Robo3R employs a masked point head for sharp, fine-grained point clouds, and a keypoint-based Perspective-n-Point (PnP) formulation to refine camera extrinsics and global alignment. Trained on Robo3R-4M, a curated large-scale synthetic dataset with four million high-fidelity annotated frames, Robo3R consistently outperforms state-of-the-art reconstruction methods and depth sensors. Across downstream tasks including imitation learning, sim-to-real transfer, grasp synthesis, and collision-free motion planning, we observe consistent gains in performance, suggesting the promise of this alternative 3D sensing module for robotic manipulation.

Abstract:
Autonomous navigation in dense, dynamic clutter remains a fundamental challenge for Unmanned Aerial Vehicles (UAVs) due to the heterogeneous obstacle scales and complex motion patterns. Existing methods often rely on fragile explicit tracking or noise-sensitive implicit flow estimation, both of which struggle with irregular geometries and real-time constraints. We propose D-Nav, a novel end-to-end reinforcement learning framework that directly maps raw LiDAR observations to control actions through a saliency-driven dual-resolution spatio-temporal representation. At the global level, D-Nav constructs a spatio-temporal spherical depth representation that encodes scene structure and motion trends directly from sequential LiDAR measurements. Building on this global context, a saliency-based refinement mechanism is designed to identify dynamically critical regions and extract fine-grained geometric and motion cues at the local level. This formulation enables the policy to reason about both large-scale dynamic context and small, irregular, fast-moving obstacles in complex dynamic environments. In addition, D-Nav introduces a new mission-aware waypoint-goal condition mechanism that explicitly guides the policy to balance collision avoidance with the need to traverse task-critical regions. Extensive simulations demonstrate a significant improvement in navigation success, increasing from 0.37 to 0.56 in complex dynamic environments. Real-world experiments further validate the robustness and real-time capability of the proposed system across diverse dynamic scenarios. The code is available at https://anonymous.4open.science/r/D-NAV-2EF6 .

Abstract:
Traversability analysis in off-road settings remains a fundamental challenge for mobile robots. Key difficulties include constructing an accurate, expressive local map from multi-modal sensor data and using the map to design traversability rules that yield desirable navigation behavior. Importantly, this system must be resilient to the limited sensing regime brought about by complex environments and high speeds. In this paper, we present TravSUITE, a traversability system suitable for high- speed navigation in off-road environments. TravSUITE consists of two major components: 1) a VFM-based voxel mapper that builds a rich geometric-semantic local map from streams of on-board sensor data, and 2) a unified neural network that jointly pre- dicts traversability-relevant quantities in bird’s eye view (BEV), including geometry, semantics, speed and cost. Our training strategy is entirely annotation-free and self-supervised, leveraging tasks such as map inpainting and inverse reinforcement learning (IRL) to learn both map representations and traversability. We also perform a thorough ablation study and comparison to state- of-the-art approaches, and the results indicate that cost learning and auxiliary inpainting each contribute significantly to planning quality, and their combination is critical for achieving state-of- the-art performance in path planning. We also design a simple risk adaptation mechanism to leverage our method’s uncertainty estimates at deploy-time, and demonstrate that a combination of inpainting and risk estimation can result in 80% fewer navigation errors and 5% faster autonomous traversal speeds in real-world hardware experiments.

Abstract:
Vision-language-action (VLA) models can enable broad open world generalization, but require large and diverse datasets. It is appealing to consider whether some of this data can come from human videos, which cover diverse real-world situations and are easy to obtain. However, it is difficult to train VLAs with human videos alone, and establishing a mapping between humans and robots requires manual engineering and presents a major research challenge. Drawing inspiration from advances in large language models, where the ability to learn from diverse supervision emerges with scale, we ask whether a similar phenomenon holds for VLAs that incorporate human video data. We introduce a simple co-training recipe, and find that human-to-robot transfer emerges once the VLA is pre-trained on sufficient scenes, tasks, and embodiments. Our analysis suggests that this emergent capability arises because diverse pretraining produces embodiment-agnostic representations for human and robot data. We validate these findings through a series of experiments probing human to robot skill transfer and find that with sufficiently diverse robot pre-training our method can nearly double the performance on generalization settings seen only in human data.

Abstract:
Pretrained vision-language models (VLMs) can make semantic and visual inferences across diverse settings, providing valuable common-sense priors for robotic control. However, effectively grounding this knowledge in robot behaviors remains an open challenge. Prior methods often employ a hierarchical approach where VLMs reason over high-level commands to be executed by separate low-level policies, e.g., vision-language-action models (VLAs). The interface between VLMs and VLAs is usually natural language task instructions, which fundamentally limits how much VLM reasoning can steer low-level behavior. We thus introduce Steerable Policies: VLAs trained on rich synthetic commands at various levels of abstraction, like subtasks, motions, and grounded pixel coordinates. By improving low-level controllability, Steerable Policies can unlock pretrained knowledge in VLMs, enabling improved task generalization. We demonstrate this benefit by controlling our Steerable Policies with both a learned high-level embodied reasoner and an off-the-shelf VLM prompted to reason over command abstractions via in-context learning. Across extensive real-world manipulation experiments, these two novel methods outperform prior embodied reasoning VLAs and VLM-based hierarchical baselines, including on challenging generalization and long-horizon tasks.

Abstract:
Prevailing Vision-Language-Action Models (VLAs) for robotic manipulation are built upon vision-language backbones pretrained on large-scale, but disconnected static web data. As a result, despite improved semantic generalization, the policy must implicitly infer complex physical dynamics and temporal dependencies solely from robot trajectories. This reliance creates an unsustainable data burden, necessitating continuous, large-scale expert data collection to compensate for the lack of innate physical understanding. We contend that while vision-language pretraining effectively captures semantic priors, it remains blind to physical causality. A more effective paradigm leverages video to jointly capture semantics and visual dynamics during pretraining, thereby isolating the remaining task of low-level control. To this end, we introduce mimic-video, a novel Video-Action Model (VAM) that pairs a pretrained Internet-scale video model with a flow matching-based action decoder conditioned on its latent representations. The decoder serves as an Inverse Dynamics Model (IDM), generating low-level robot actions from the latent representation of video-space action plans. Our extensive evaluation shows that our approach achieves state-of-the-art performance on simulated and real-world robotic manipulation tasks, improving sample efficiency by 10x and convergence speed by 2x compared to traditional VLA architectures.

Abstract:
Fine-grained and contact-rich manipulation remain challenging for robots, largely due to the underutilization of tactile feedback. To address this, we introduce TouchGuide, a novel cross-policy visuo-tactile fusion paradigm that fuses modalities within a low-dimensional action space. Specifically, TouchGuide operates in two stages to guide a pre-trained diffusion or flow-matching visuomotor policy at inference time. First, the policy produces a coarse, visually-plausible action using only visual inputs during early sampling. Second, a task-specific Contact Physical Model (CPM) provides touch guidance to steer and refine the action, ensuring it aligns with realistic physical contact conditions. Trained through contrastive learning on limited expert demonstrations, the CPM provides a tactile-informed feasibility score to steer the sampling process toward refined actions that satisfy physical contact constraints. Furthermore, to facilitate TouchGuide training with high-quality and cost-effective data, we introduce TacUMI, a data collection system. TacUMI achieves a favorable trade-off between precision and affordability; by leveraging rigid fingertips, it obtains direct tactile feedback, thereby enabling the collection of reliable tactile data. Extensive experiments on five challenging contact-rich tasks, such as shoe lacing and chip handover, show that TouchGuide consistently and significantly outperforms state-of-the-art visuo-tactile policies.

Abstract:
Robots deployed in the real world must learn from their experience and improve over time. This requires a mechanism of practicing and learning from feedback. In this paper, we propose a generator--verifier framework for generalist robot policies for autonomous policy improvement. We use a pre-trained policy as a "generator'' and pair it with a gradient-free "visual verifier" that evaluates and selects actions at inference time. This framework enables inference-time steering that improves real-world performance without additional training. Across SIMPLER simulation and real-world DROID setups, we show that inference-time verification consistently improves policy performance over na\"ive execution, and that these gains hold across different choices of verifiers, including VLM-based and heuristic ones. Beyond inference-time steering, we demonstrate that verified rollouts provide effective supervision for offline policy improvement: policies fine-tuned on autonomously verified data achieve steep performance gains, with performance continuing to improve as more verified demonstrations are collected. Notably, we find that post-training with verified rollouts matches the efficiency of human expert demonstrations, while requiring no human interventions. Our results highlight test-time verification as a practical and scalable mechanism for improving robotic policies during autonomous deployment.

Abstract:
Vision-Language-Action (VLA) models aim for general robot learning by aligning action as a modality within powerful Vision-Language Models (VLM). Existing VLAs rely on end-to-end supervision to implicitly enable the action decoding process to learn task-relevant features. However, without explicit guidance, these models often overfit to spurious correlations, such as visual shortcuts or environmental noise, limiting their generalization. In this paper, we introduce GuidedVLA, a framework designed to manually guide the action generation to focus on task-relevant factors. Our core insight is to treat the action decoder not as a monolithic learner, but as an assembly of functional components. Individual attention heads are supervised by manually defined auxiliary signals to capture distinct factors. As an initial study, we instantiate this paradigm with three specialized heads: object grounding, spatial geometry, and temporal skill logic. Across simulation and real-robot experiments, GuidedVLA improves success rates in both in-domain and out-of-domain settings compared to strong VLA baselines. Finally, we show that the quality of these specialized factors correlates positively with task performance and that our mechanism yields decoupled, high-quality features. Our results suggest that explicitly guiding the learning of decision making for action decoder is a promising direction for building more robust and general VLA models.

Abstract:
Deploying robots at scale demands robustness to the long tail of everyday situations. The countless variations in scene layout, object geometry, and task specifications that characterize real environments are vast and underrepresented in existing robot benchmarks. Measuring this level of generalization requires infrastructure at a scale and diversity that physical evaluation alone cannot provide. We introduce MolmoSpaces, a fully open ecosystem to support large-scale benchmarking of robot policies. MolmoSpaces consists of over 230k diverse indoor environments, ranging from handcrafted household scenes to procedurally generated multiroom houses, populated with 130k richly annotated object assets, including 48k manipulable objects with 42M stable grasps. Crucially, these environments are simulator-agnostic, supporting popular options such as MuJoCo, Isaac, and ManiSkill. The ecosystem supports the full spectrum of embodied tasks: static and mobile manipulation, navigation, and multiroom long-horizon tasks requiring coordinated perception, planning, and interaction across entire indoor environments. We also design MolmoSpaces-bench, a benchmark suite of 8 tasks in which robots interact with our diverse scenes and richly annotated objects. Our experiments show MolmoSpaces-bench exhibits strong sim-to-real correlation (R = 0.96, ρ = 0.98), confirm newer and stronger policies outperform earlier versions in our benchmarks, and identify key sensitivities to prompt phrasing, initial joint positions, and camera occlusion. Through MolmoSpaces and its open-source assets and tooling, we provide a foundation for scalable data generation, policy training, and benchmark creation for robot learning research.

Authors: Yufei Jia, Heng Zhang, Ziheng Zhang, Lei Han, Junzhe Wu, Mingrui Yu, Zifan Wang, Dixuan Jiang, Zheng Li, Chenyu Cao, Zhuoyuan Yu, Xun Yang, Haizhou Ge, Yuchi Zhang, Jiayuan Zhang, Zhenbiao Huang, Tianle Liu, Shenyu Chen, Jiacheng Wang, Bin Xie, Xuran Yao, Xiwa Deng, Guangyu Wang, Jinzhi Zhang, Lei Hao, Zhixing Chen, Yuxiang Chen, Anqi Wang, Hongyun Tian, Yiyi Yan, Zhanxiang Cao, Yizhou Jiang, Hanyang Shao, Yue Li, Lu Shi, Bokui Chen, Wei Sui, Hanqing Cui, Yusen Qin, Tiancai Wang, Ruqi Huang, Guyue Zhou

Abstract:
Embodied AI research is undergoing a shift toward vision-centric perceptual paradigms. While massively parallel simulators have catalyzed breakthroughs in proprioception-based locomotion, their potential remains largely untapped for vision-centric tasks due to the prohibitive computational overhead of large-scale photorealistic rendering. Furthermore, the creation of simulation-ready 3D assets heavily relies on labor-intensive manual modeling, while the significant sim-to-real physical gap hinders the transfer of contact-rich manipulation policies. To address these bottlenecks, we propose gs_playground, a multi-modal simulation framework designed to accelerate end-to-end perceptual learning. We develop a novel high-performance parallel physics engine, specifically designed to integrate with a batch 3D Gaussian Splatting (3DGS) rendering pipeline to ensure high-fidelity synchronization. Our system achieves a breakthrough throughput of \mathbf10^4 FPS at \mathbf640 × 480 resolution, significantly lowering the barrier for large-scale visual RL. Additionally, we introduce an automated Real2Sim workflow that reconstructs photorealistic, physically consistent, and memory-efficient environments, streamlining the generation of complex simulation-ready scenes. Extensive experiments on locomotion, navigation, and manipulation demonstrate that gs_playground effectively bridges the perceptual and physical gaps across diverse embodied tasks. We will open-source the full-stack framework to empower the research community.

Abstract:
Diverse applications for robotics, such as industry and agriculture, require robots to operate across various embodiments, changing visual conditions, and complex planning. Vision–Language Models (VLMs) offer a promising foundation for general-purpose and interpretable robotic reasoning. Aligning VLMs with diverse robot applications requires a modular understanding of the individual decision components that underlie robotic behavior. Capturing such structure is challenging for conventional robot benchmarks that are primarily based on teleoperated, end-to-end datasets. We propose Robot Question Answering (RQA), a modular evaluation framework and RoboVista, a benchmark curated from real robotic systems, research papers, and expert annotations. RoboVista contains 474 VQAs with human annotated reasoning and covers 39 unique task types in agricultural, industrial, domestic, surgical robotics, autonomous driving, and open robot datasets. Experiments on RoboVista show that state-of-the-art VLMs exhibit substantial gaps. Physical robot experiments suggest strong correlation between RoboVista performance and real-world task execution.

Abstract:
Sampling-based controllers, such as Model Predictive Path Integral (MPPI) methods, offer substantial flexibility but often suffer from high variance and low sample efficiency. To address these challenges, we introduce a hybrid variance-reduced MPPI framework that integrates a prior model into the sampling process. Our key insight is to decompose the objective function into a known approximate model and a residual term. Since the residual captures only the discrepancy between the model and the objective, it typically exhibits a smaller magnitude and lower variance than the original objective. Although this principle applies to general modeling choices, we demonstrate that adopting a quadratic approximation enables the derivation of a closed-form, model-guided prior that effectively concentrates samples in informative regions. Crucially, the framework is agnostic to the source of geometric information, allowing the quadratic model to be constructed from exact derivatives, structural approximations (e.g., Gauss- or Quasi-Newton), or gradient-free randomized smoothing. We validate the approach on standard optimization benchmarks, a nonlinear, underactuated cart-pole control task, and a contact-rich manipulation problem with non-smooth dynamics. Across these domains, we achieve faster convergence and superior performance in low-sample regimes compared to standard MPPI. These results suggest that the method can make sample-based control strategies more practical in scenarios where obtaining samples is expensive or limited.

Abstract:
High-precision heavy-duty grading is a common step in earthworks, traditionally carried out manually by skilled operators. Removing a significant amount of material while achieving a high-precision surface requires substantial machine-specific experience. Different hydraulic architectures react differently to operator inputs and soil interaction forces, which makes generalizable controllers challenging. In this paper, we present an autonomous controller that achieves high-precision grading at expert-operator speed on Load Sensing and Negative Flow Control machines alike. We split our controller into two parts: (1) a hydraulic-aware low-level loop that is hydraulic architecture-specific and (2) a path-tracking layer that coordinates joint motions and responses. Through a calibration process, our technique is applicable to load-sensing and negative-flow-control machinery. To showcase its versatility, we benchmark our approach on two excavators with different hydraulics and compare it against a commercial state-of-the-art solution. Our technique (RMSE 1.8cm) outperforms the commercial solution (RMSE 4.7cm) in precision by a factor of x2.6 and improves machine usage by leveraging the maximum function pressure, as opposed to commercial solutions that stall prematurely.

Abstract:
Animatronic faces are a central component of socially interactive robots, enabling rich nonverbal communication through facial articulation. However, state-of-the-art animatronic faces are typically tailored systems: each new facial geometry requires extensive manual mechanical redesign, making large-scale personalization prohibitively slow and costly. In this work, we pursue automated and scalable mechanical face synthesis, aiming to rapidly generate a physically realizable facial mechanism for any given face. We introduce a parametric, linkage-driven mechanical face template whose topology and actuator layout are explicitly parameterized to support systematic scaling and retargeting across diverse facial morphologies. Building on this template, we propose a hierarchical automatic design algorithm that takes a single 2D portrait as input, reconstructs a target 3D face, and synthesizes a collision-free, manufacturable internal mechanism. The algorithm combines anatomy-guided feasible motion volumes, AU-derived trajectory-based expressiveness objectives, and a collision-driven outer-loop refinement strategy. Beyond hardware synthesis, we argue that future mechanical faces deployed at scale must engage in bidirectional, multi-turn conversation, rather than functioning solely as speaking or listening heads. To this end, we develop a dual-identity conversational facial motion synthesis framework that jointly models speaking and listening behaviors from audio, producing temporally coherent 3D facial motion suitable for physical execution. We validate our system through extensive experiments, including (i) quantitative evaluation of automatic mechanism synthesis across diverse facial geometries, (ii) comparisons against manual mechanical design, (iii) benchmarks on conversational facial motion synthesis and real-time deployment, and (iv) perceptual user studies. The entire hardware design, code, and datasets will be released.

Abstract:
Robots must learn from both what people do and what they say, but either modality alone is often incomplete: physical corrections are grounded but ambiguous in intent, while language expresses high-level goals but lacks physical grounding. We introduce QuickLAP: Quick Language–Action Preference learning, a Bayesian framework that fuses physical and language feedback to infer reward functions in real time. Our key insight is to treat language as a probabilistic observation over the user’s latent preferences, clarifying which reward features matter and how physical corrections should be interpreted. QuickLAP uses Language Models (LMs) to extract reward feature attention masks and preference shifts from free-form utterances that are combined with physical feedback in a closed-form update rule. This enables fast, real-time, and robust reward learning that handles ambiguous feedback. In a robotic manipulation and a semi-autonomous driving simulator, QuickLAP reduces reward learning error by over 70% compared to physical-only and heuristic multimodal baselines. User studies further validate our approach: participants found QuickLAP significantly more understandable and collaborative, and preferred its learned behavior over baselines. Code is available at [redacted]

Abstract:
Physical caregiving robots need to assist different users with different tasks in diverse environments, and they come in many embodiments. While substantial progress has been made on individual caregiving tasks, most existing systems remain tightly coupled to specific environments and robot embodiments, and often do not explicitly model or constrain interaction around people, despite humans being special agents in the environment. This motivates a focus on adapting to context that emerges from the joint interaction between the environment and the robot’s embodiment. We propose E^2-CARE, a framework that enables context-aware adaptation by representing primitive caregiving skills as interaction templates whose execution is reshaped online. E^2-CARE represents the environment, the robot, and the human within a unified 3D dynamic scene graph that models these interaction contexts explicitly, and synthesizes task-specific constraints to govern how each skill is executed. By enforcing these constraints at runtime, the same skill templates can be reused zero-shot and safely across diverse environments and robot embodiments. We evaluate E^2-CARE across four activities of daily living in hundreds of simulated household environments, including assistive home settings, and across diverse robot embodiments, and validate it through user studies on two caregiving tasks with two robots in various real-world environments. Results demonstrate consistent and successful adaptation across these environments and embodiments.

Abstract:
Towards the aim of generalized robotic manipulation, spatial generalization is the most fundamental capability that requires the policy to work robustly under different spatial distribution of objects, environment and agent itself. To achieve this, substantial human demonstrations need to be collected to cover different spatial configurations for training a generalized visuomotor policy via imitation learning. Prior works explore a promising direction that leverages data generation to acquire abundant spatially diverse data from minimal source demonstrations. However, most approaches face significant sim-to-real gap and are often limited to constrained settings, such as fixed-base scenarios and predefined camera viewpoints. In this paper, we propose a real-to-real 3D data generation framework (R2RGen) that directly augments the pointcloud observation-action pairs to generate real-world data. R2RGen is simulator- and rendering-free, thus being efficient and plug-and-play. Specifically, we propose a unified three-stage framework, which (1) pre-processes source demonstrations under different camera setups in a shared 3D space with scene / trajectory parsing; (2) augments objects and robot's position with a group-wise backtracking strategy; (3) aligns the distribution of generated data with real-world 3D sensor using camera-aware post-processing. Empirically, R2RGen substantially enhances data efficiency on extensive experiments and demonstrates strong potential for scaling and application on mobile manipulation.

Abstract:
To meet the demands of increasingly diverse dexterous hand hardware, it is crucial to develop a policy that enables zero-shot cross-embodiment grasping without redundant re-learning. Cross-embodiment alignment is challenging due to heterogeneous hand kinematics and physical constraints. Existing approaches typically predict intermediate motion targets and retarget them to each embodiment, which may introduce errors and violate embodiment-specific limits, hindering transfer across diverse hands. To overcome these limitations, we propose DexGrasp-Zero, a policy that learns universal grasping skills from diverse embodiments, enabling zero-shot transfer to unseen hands. We first introduce a morphology-aligned graph representation that maps each hand's kinematic keypoints to anatomically grounded nodes and equips each node with tri-axial orthogonal motion primitives, enabling structural and semantic alignment across different morphologies. Relying on this graph-based representation, we design a Morphology-Aligned Graph Convolutional Network (MAGCN) to encode the graph for policy learning. MAGCN incorporates a Physical Property Injection mechanism that fuses hand-specific physical constraints into the graph features, enabling adaptive compensation for varying link lengths and actuation limits for precise and stable grasping. Our extensive simulation evaluations on the YCB dataset demonstrate that our policy, jointly trained on four heterogeneous hands (Allegro, Shadow, Schunk, Ability) , achieves an 85% zero-shot success rate on unseen hardware (LEAP, Inspire), outperforming the state-of-the-art method by 59.5%. Real-world experiments further evaluate our policy on three robot platforms (LEAP, Inspire, Revo2) , achieving an 82% average success rate on unseen objects.

Abstract:
Dynamic manipulation of deformable objects is challenging for humans and robots because they have infinite degrees of freedom and exhibit underactuated dynamics. We introduce a Task-Level Iterative Learning Control method for dynamic manipulation of deformable objects. We demonstrate this method on a non-planar rope manipulation task called the flying knot. Using a single human demonstration and a simplified rope model, the method learns directly on hardware without reliance on large amounts of demonstration data or massive amounts of simulation. At each iteration, the algorithm constructs a local inverse model of the robot and rope by solving a quadratic program to propagate task-space errors into action updates. We evaluate performance across 7 different kinds of ropes, including chain, latex surgical tubing, and braided and twisted ropes, ranging in thicknesses of 7-25mm and densities of 0.013-0.5 kg/m. Learning achieves a 100% success rate within 10 trials on all ropes. Furthermore, the method can successfully transfer between most rope types in approximately 2-5 trials.

Abstract:
General-purpose robot reward models are typically trained to predict absolute task progress from expert demonstrations, providing only local, frame-level supervision. While effective for expert demonstrations, this paradigm scales poorly to large scale real-world robotics datasets where failed and suboptimal trajectories are abundant, and assigning dense progress labels is ambiguous or ill-defined. We introduce Robometer, a scalable reward modeling framework that combines intra-trajectory progress supervision with inter-trajectory preference supervision. Robometer is trained with a dual objective: a frame-level progress loss that anchors reward magnitude on expert data, and a trajectory-comparison preference loss that imposes global ordering constraints across trajectories of the same task, enabling effective learning from suboptimal and failed trajectories. To support this formulation at scale, we curate RBM-1M, a large-scale reward-learning dataset comprising over one million trajectories spanning diverse robot embodiments and tasks, including substantial suboptimal and failure data. Across benchmarks and real-world evaluations, Robometer learns more generalizable reward functions than prior methods and improves robot learning performance across a diverse set of downstream applications.

Abstract:
The prevalent paradigm in robot learning attempts to generalize across environments, embodiments, and tasks with language prompts at runtime. A fundamental tension limits this approach: language is often too abstract to guide the concrete physical understanding required for robust manipulation. In this work, we introduce Contact-Anchored Policies (CAP), which replace language conditioning with points of physical contact in space. Simultaneously, we structure CAP as a library of modular utility models rather than a monolithic generalist policy. This factorization allows us to implement a real-to-sim iteration cycle: we build EgoGym, a lightweight simulation benchmark, to rapidly identify failure modes and refine our models and datasets prior to real-world deployment. We show that by conditioning on contact and iterating via simulation, CAP generalizes to novel environments and embodiments out of the box on three fundamental manipulation skills while using only 23 hours of demonstration data, and outperforms large, state-of-the-art VLAs in zero-shot evaluations by 56%. All model checkpoints, codebase, hardware, simulation, and datasets will be open-sourced.

Abstract:
Policy steering is an emerging way to adapt robot behaviors at deployment-time: a learned verifier analyzes low-level action samples proposed by a pre-trained policy (e.g., diffusion policy) and selects only those aligned with the task. While Vision-Language Models (VLMs) are promising general-purpose verifiers due to their reasoning capabilities, existing frameworks often assume these models are well-calibrated. In practice, the overconfident judgment from VLM can degrade the steering performance under both high-level semantic uncertainty in task specifications and low-level action uncertainty or incapability of the pre-trained policy. We propose uncertainty-aware policy steering (UPS), a framework that jointly reasons about semantic task uncertainty and low-level action feasibility, and selects an uncertainty resolution strategy: execute a high-confidence action, clarify task ambiguity via natural language queries, or ask for action interventions to correct the low-level policy when it is deemed incapable at the task. We leverage conformal prediction to calibrate the composition of the VLM and the pre-trained base policy, providing statistical assurances that the verifier selects the correct strategy. After collecting interventions during deployment, we employ residual learning to improve the capability of the pre-trained policy, enabling the system to learn continually but with minimal expensive human feedback. We demonstrate our framework through experiments in simulation and on hardware, showing that UPS can disentangle confident, ambiguous, and incapable scenarios and minimizes expensive user interventions compared to uncalibrated baselines and prior human- or robot-gated continual learning approaches.

Abstract:
We present EigenSafe, an operator-theoretic framework for safety assessment of learning-enabled stochastic systems. In many robotic applications, the dynamics are inherently stochastic due to factors such as sensing noise and environmental disturbances, and it is challenging for conventional methods such as Hamilton-Jacobi reachability and control barrier functions to provide a well-calibrated safety critic that is tied to the actual safety probability. We derive a linear operator that governs the dynamic programming principle for safety probability, and find that its dominant eigenpair provides critical safety information for both individual state-action pairs and the overall closed-loop system. The proposed framework learns this dominant eigenpair, which can be used to either inform or constrain policy updates. We demonstrate that the learned eigenpair effectively facilitates safe reinforcement learning. Further, we validate its applicability in enhancing the safety of learned policies from imitation learning through robot manipulation experiments using a UR3 robotic arm in a food preparation task.

Abstract:
Language-conditioned manipulation policies typically process instructions and observations through shared network parameters. This task-state entanglement provides a pathway for observation leakage -- networks learn scene-to-action shortcuts that bypass language grounding entirely. DISC eliminates this failure structurally. Rather than conditioning a universal policy on language, DISC uses a hypernetwork to generate the entire parameter set of a task-specific visuomotor policy from the instruction alone. The generated policy never directly accesses language; therefore, its task-awareness must come from the language. Consequently, observation leakage has no pathway to emerge. On the other hand, generating coherent high-dimensional policy weights is itself a challenging problem. We address it with a two-stage hypernetwork whose refinement stage embeds the structure of gradient-based optimization as a feed-forward inductive bias, producing globally consistent parameters without actual gradient computation. Trained entirely from scratch on standard data budgets, DISC outperforms all entangled baselines on LIBERO-90 and Meta-World, with advantages that widen on complex, long-horizon tasks -- and surpasses the large-scale pretrained π₀ despite using no external pretraining data. On a real-world benchmark where all tasks share identical visual context, DISC substantially outperforms entangled alternatives, directly confirming that generated parameters, not visual shortcuts, drive behavior. The hypernetwork further learns a semantically structured parameter manifold that enables few-shot adaptation from minimal demonstrations and robust generalization across paraphrased instructions.

Abstract:
The ability to manipulate tools significantly expands the set of tasks a robot can perform. Yet, tool manipulation represents a challenging class of dexterity, requiring grasping thin objects, in-hand object rotations, and forceful interactions. Since collecting teleoperation data for these behaviors is challenging, sim-to-real reinforcement learning (RL) is a promising alternative. However, prior approaches typically require substantial engineering effort to model objects and tune reward functions for each task. In this work, we propose SimToolReal, taking a step towards generalizing sim-to-real RL policies for tool manipulation. Instead of focusing on a single object and task, we procedurally generate a large variety of tool-like object primitives in simulation and train a single RL policy with the universal goal of manipulating each object to random goal poses. This approach enables SimToolReal to perform general dexterous tool manipulation at test-time without any object or task-specific training. We demonstrate that SimToolReal outperforms prior retargeting and fixed-grasp methods by 37% while matching the performance of specialist RL policies trained on specific target objects and tasks. Finally, we show that SimToolReal generalizes across a diverse set of everyday tools, achieving strong zero-shot performance over 120 real-world rollouts spanning 24 tasks, 12 object instances, and 6 tool categories.

Abstract:
In this paper, we address the problem of tactile sim-to-real policy transfer for contact-rich tasks. Existing methods primarily focus on vision-based sensors and emphasize image rendering quality while providing overly simplistic models of force and shear. Consequently, these models exhibit a large sim-to-real gap for many dexterous tasks. Here, we present HydroShear, a non-holonomic hydroelastic tactile simulator that advances the state-of-the-art by modeling: a) stick-slip transitions, b) path-dependent force and shear build up, and c) full SE(3) object-sensor interactions. HydroShear extends hydroelastic contact models using Signed Distance Functions (SDFs) to track the displacements of the on-surface points of an indenter during physical interaction with the sensor membrane. Our approach generates physics-based, computationally efficient force fields from arbitrary watertight geometries while remaining agnostic to the underlying physics engine. In experiments with GelSight Minis, HydroShear more faithfully reproduces real tactile shear compared to existing methods. This fidelity enables zero-shot sim-to-real transfer of reinforcement learning policies across four tasks: peg insertion, bin packing, book shelving for insertion, and drawer pulling for fine gripper control under slip. Our method achieves a 93% average success rate, outperforming policies trained on tactile images (34%) and alternative shear simulation methods (58%–61%).

Abstract:
Reinforcement learning has shown strong promise for quadrupedal agile locomotion, even with proprioception-only sensing. In practice, however, sim-to-real gap and reward overfitting in complex terrains can produce policies that fail to transfer, while physical validation remains risky and inefficient. To address these challenges, we introduce a unified framework encompassing a Mixture-of-Experts (MoE) locomotion policy for robust multi-terrain representation with RoboGauge, a predictive assessment suite that quantifies sim-to-real transferability. The MoE policy employs a gated set of specialist experts to decompose latent terrain and command modeling, achieving superior deployment robustness and generalization via proprioception alone. RoboGauge further provides multi-dimensional proprioception-based metrics via sim-to-sim tests over terrains, difficulty levels, and domain randomizations, enabling reliable MoE policy selection without extensive physical trials. Experiments on a Unitree Go2 demonstrate robust locomotion on unseen challenging terrains, including snow, sand, stairs, slopes, and 30 cm obstacles. In dedicated high-speed tests, the robot reaches 4 m/s and exhibits an emergent narrow-width gait associated with improved stability at high velocity.

Abstract:
Despite recent efforts to collect multi-task or multiembodiment datasets, to design efficient recipes for training Vision-Language-Action models (VLAs), and to showcase these models on selected robot platforms, generalist robot capabilities and cross-embodiment transfer remain largely elusive ideals. This cross-embodiment robot learning paradigm remains limited by fragmented data-collection infrastructure, the lack of standardization on versatile data formats, and the significant engineering effort involved in reproducing hardware setups and organizing multiple control stacks for quickly deploying models on diverse robot platforms. As a result, most robot code tends to be highly specific to the exact robot setup that the user decided on, which adds major overhead when attempting to reuse, recycle, or share artifacts between users. To bridge this gap, we present Robot I/O (RIO), an open-source Python-based framework that provides flexible, lightweight components for robot control, teleoperation, data formatting, sensor configuration, and policy deployment across diverse hardware platforms and morphologies. RIO provides abstractions that enable users to make any choice (robots, sensors, teleoperation interfaces, middlewares, data formats, policies) and to switch between them, with minimal reconfiguration effort. We validate RIO on VLA deployment workflows across three morphologies (single-arm, bimanual, humanoid) and four robot hardware platforms with varying grippers and cameras. We showcase policy rollouts by collecting teleoperated data to fine-tune state-of-the-art VLAs, including π0.5 and GR00T, on household tasks such as pick-andplace, folding, and bowl scrubbing. By open sourcing all our efforts, we hope the wider robotics community can accelerate their pace of robot learning on real-world robot hardware.

Abstract:
Precise magnetic field modeling is fundamental to the closed-loop control of electromagnetic navigation systems (eMNS) and the analytical Multipole Expansion Model (MPEM) is the current standard. However, the MPEM relies on strict physical assumptions regarding source symmetry and isolation, and requires optimization-based calibration that is highly sensitive to initialization. These constraints limit its applicability to systems with complex or irregular coil geometries. This work introduces an alternative modeling paradigm based on multi-layer perceptrons that learns nonlinear magnetic mappings while strictly preserving the linear dependence on currents. As a result, the field models enable fast, closed-form minimum-norm inversion with evaluation times of approximately 1 ms, which is critical for high-bandwidth magnetic control. For model training and evaluation we use large-scale, high-density datasets collected from the research-grade OctoMag and clinical-grade Navion systems. Our results demonstrate that data-driven models achieve predictive fidelity equivalent to the MPEM while maintaining comparable data efficiency. Furthermore, we demonstrate that straightforward design choices effectively eliminate spurious workspace ill-conditioning frequently reported in MPEM-based calibration. To facilitate future research, we release the complete codebase and datasets open source.

Abstract:
Achieving stable, sustained grasping with soft robotic hands remains a fundamental challenge. Compliance enables safe and adaptive contact, yet the intrinsic viscoelasticity of soft polymers leads to stress relaxation and a continuous decay of grasping force during holding. Inspired by human grasping, which combines phase-dependent stiffness regulation with continuous sensing and feedback, this paper presents an integrated structure--perception--learning framework. We develop a variable-stiffness soft gripper that uses onboard vision and infrared thermography to track deformation and the temperature field in real time, preserving continuous tracking of the interaction state. To mitigate relaxation-induced force decay, we propose a temperature-coupled viscoelastic force representation, together with a physics-informed learning model, to reconstruct the force trend and provide explicit compensation during holding. Experiments show that, in a 280s force-controlled grasp-and-hold task, the proposed method maintains the desired force with a mean absolute error of 0.066N, outperforming fixed-aperture and instantaneous-only baselines by 80% and 95%, respectively. Overall, the results support a mechanism--AI co-design view: mechanisms shape feasible interactions, while learning compensates remaining uncertainty in viscoelastic dynamics, together enabling stable, sustained grasping.

Abstract:
We consider an open-world planning setting in which service robots must operate in unknown environments with incomplete knowledge of objects and actions. Traditional closed-world approaches with pre-programmed knowledge bases fail when robots encounter unexpected situations and tasks, posing a fundamental challenge for autonomous knowledge expansion in human environments. In this work, we propose an open-world planning framework that enables robots to automatically generate, verify, and update hypotheses about their abstract world models. Our key insight is to explicitly maintain uncertainty-aware knowledge expansion and integrate hypothesis verification into goal-reaching planning. The framework leverages foundation models to generate initial hypotheses over states and transitions, and applies automated planning to produce action sequences that jointly address hypothesis verification and task execution. Through iterative execution and refinement, the robot expands its knowledge by incorporating verification feedback from the foundation models when hypotheses prove incorrect. Extensive experiments in simulated and real-world environments demonstrate that our framework enables autonomous knowledge expansion and effective operation in open-world settings. These results indicate that integrating uncertainty-aware model expansion from robot foundation models with planning advances the practical deployment of household service robots.

Abstract:
This work presents a novel and efficient non-linear programming framework that tightly integrates hierarchical decision-making with inverse kinematic planning and control. Decision-making plays a central role in many aspects of robotics, from sparse inverse kinematic control with a minimal number of joints, to inverse kinematic planning while simultaneously selecting a discrete end-effector location from multiple candidates. Current approaches often rely on heavy computations using mixed-integer non-linear programming, separate decision-making from inverse kinematics (some times approximated by reachability methods), or employ efficient but less accurate \ell₁-norm formulations of linear sparse programming, without addressing the underlying non-linear problem formulations. In contrast, the proposed sparse hierarchical non-linear programming solver is efficient, versatile, and accurate by exploiting sparse hierarchical structure and leveraging the rarely used \ell₀-norm in robotics. The solver efficiently addresses complex non-linear hierarchical decision-making problems, such as inverse kinematic planning with simultaneous prioritized selection of end-effector locations from a large set of candidates, or inverse kinematic control with simultaneous selection of bi-manual grasp locations on a randomly rotated box.

Abstract:
This paper considers a Horizon-Optimal Control problem that seeks a dynamically feasible trajectory while minimizing the planning horizon, which is a fundamental problem in robotics with numerous applications. While many famous optimal control methods, such as LQR, iLQR/DDP, are well studied and deployed on various robots, they often have a fixed planning horizon, and their horizon-optimal counterparts are still undiscovered. The best result in the literature solves the horizon-optimal LQR problem by shifting the horizon and reusing the value functions computed by the Riccati recursion, which leads to an efficient algorithm. However, this approach is limited to LQR with time-invariant dynamics and costs only. This paper finds that the Riccati recursion can be reformulated into a form of Linear Fractional Transformation (LFT), which enjoys the structure that enables efficient computational reuse even for non-stationary dynamics and costs. Based on this insight, we develop a new efficient algorithm to solve Horizon-Optimal Time-Varying LQR problem to optimality, and further fuse it with DDP to handle general non-quadratic costs and nonlinear dynamics. Results show, our approach always finds the same optimal solution as a naive brute force baseline method, while running up to 40 times faster. For nonlinear dynamics, our method always finds better solutions than approximation using time-invariant LQR.

Abstract:
Contact often occurs without macroscopic surface deformation, such as during interaction with liquids, semi-liquids, or ultra-soft materials. However, most existing tactile sensors rely on deformation to infer contact, making such light-contact interactions difficult to perceive robustly. To address this, we present LightTact, a visual-tactile fingertip sensor that makes contact directly visible via a deformation-independent principle. LightTact features an ambient-blocking optical configuration that suppresses both external light and internal illumination at non-contact regions, while transmitting only the scattered light generated at true contacts. As a result, LightTact produces high-contrast raw images in which non-contact pixels remain near-black (mean gray value < 3) and contact pixels preserve the natural appearance of the contacting surface. Built on this, LightTact achieves accurate pixel-level contact segmentation that is robust to material properties, contact force, surface appearance, and environmental lighting. We further demonstrate that LightTact unlocks new robotic manipulation behaviors that require detection of extremely light contact, including water spreading, facial-cream dipping, and soft thin-film interaction. In addition, we show that LightTact’s spatially aligned visual-tactile images can be directly interpreted by vision-language models.

Abstract:
Robots operating in unstructured environments must satisfy dynamic constraints that can change across tasks and even within a single execution. While diffusion policies can learn multimodal behaviors from demonstrations, adapting a trained policy at runtime to newly encountered or evolving constraints remains an open challenge. We propose RAG-Diff, a runtime adaptation framework for a frozen transformer diffusion policy that leverages retrieval-augmented memory. RAG-Diff maintains PrefMem, a memory bank that stores vision-language embeddings together with (i) state-action snippets and (ii) constraint annotations. At test time, RAG-Diff queries PrefMem to retrieve the nearest entry and uses it to steer sampling in two complementary ways. First, I-Atten (in-place attention recomputation) inserts the retrieved snippet as additional cross-attention memory tokens and performs a classifier-free-guidance-style update, biasing denoising toward preference-consistent motion. Second, a predictive guidance mechanism incorporates the retrieved constraint parameters during diffusion sampling to discourage violations. To demonstrate the effectiveness of RAG-Diff, we choose physical robot caregiving as a domain with personalized and time-varying constraints. We first benchmark on an adapted PushT environment in simulation with contact-force limits and region-to-avoid constraints. We then evaluated our method on a suite of physical caregiving tasks spanning diverse preference types: (i) interaction and affordance preferences in bed bathing, (ii) ROM-based assistance-level preferences in medicine delivery, (iii) semantic preferences in shelf cleaning, and (iv) trajectory preferences in feeding, in both RCareWorld simulation and with a real robot. We further conducted real-world user studies on the bed-bathing task. Results show that RAG-Diff improves both task success and constraint satisfaction compared to a range of baselines, including unguided diffusion and other guidance- or sampling-based variants.

Abstract:
Humanoid robots exhibit significant potential for executing complex whole-body interaction tasks in unstructured environments. While recent advancements in Human-Object Interaction (HOI) have been substantial, prevailing methodologies predominantly address the manipulation of fully actuated objects, where the target is rigidly coupled to the robot's end-effector and its state is strictly constrained by the robot's kinematics. This paradigm neglects the pervasive class of underactuated objects characterized by independent dynamics and non-holonomic constraints, which pose significant control challenges due to complex coupling forces and frequent visual occlusions. To bridge this gap, we propose HAIC, a unified framework designed to enable robust interaction across a spectrum of object dynamics without reliance on external state estimation. Central to our approach is a novel dynamics predictor that infers high-order object states, specifically velocity and acceleration, solely from proprioceptive history. These predictions are explicitly projected onto static geometric priors to construct a spatially grounded representation of dynamic occupancy, allowing the policy to internalize collision boundaries and contact affordances in visual blind spots. We employ an asymmetric fine-tuning strategy where the world model continuously adapts to the student policy’s exploration, ensuring robust state estimation under distribution shifts. We evaluate our framework on a Unitree G1 humanoid robot. Empirical results demonstrate that HAIC achieves high success rates in agile object interactions, including skateboarding, cart pushing, and cart pulling under various weight load conditions, by proactively compensating for inertial physical perturbations, while HAIC simultaneously masters multi-object interaction involving long-horizon tasks and carrying a box across composed terrain by predicting the dynamics of multiple objects.

Abstract:
While recent advances in humanoid locomotion have achieved stable walking on varied terrains, capturing the agility and adaptivity of highly dynamic human motions remains an open challenge. In particular, agile parkour in complex environments demands not only low-level robustness, but also human-like motion expressiveness, long-horizon skill composition, and perception-driven decision-making. In this paper, we present Perceptive Humanoid Parkour (PHP), a modular framework that enables humanoid robots to autonomously perform long-horizon, vision-based parkour across challenging obstacle courses. Our approach first leverages motion matching, formulated as nearest-neighbor search in a feature space, to compose retargeted atomic human skills into long-horizon kinematic trajectories. This framework enables the flexible composition and smooth transition of complex skill chains while preserving the elegance and fluidity of dynamic human motions. Next, we train motion-tracking reinforcement learning (RL) expert policies for these composed motions, and distill them into a single depth-based, multi-skill student policy, using a combination of DAgger and RL. Crucially, the combination of perception and skill composition enables autonomous, context-aware decision-making: using only onboard depth sensing and a discrete 2D velocity command, the robot selects and executes whether to step over, climb onto, vault or roll off obstacles of varying geometries and heights. We validate our framework with extensive real-world experiments on a Unitree G1 humanoid robot, demonstrating highly dynamic parkour skills such as climbing tall obstacles up to 1.25 m (96% robot height), as well as long-horizon multi-obstacle traversal with closed-loop adaptation to real-time obstacle perturbations.

Abstract:
While recent advances have demonstrated strong performance in individual humanoid skills such as upright locomotion, fall recovery and whole-body coordination, learning a single policy that masters all these skills remains challenging due to the diverse dynamics and conflicting control objectives involved. To address this, we introduce X-Loco, a framework for training a vision-based generalist humanoid locomotion policy. X-Loco trains multiple oracle specialist policies and adopts a synergetic policy distillation with a case-adaptive specialist selection mechanism, which dynamically leverages multiple specialist policies to guide a vision-based student policy. This design enables the student to acquire a broad spectrum of locomotion skills, ranging from fall recovery to terrain traversal and whole-body coordination skills. To the best of our knowledge, X-Loco is the first framework to demonstrate vision-based humanoid locomotion that jointly integrates upright locomotion, whole-body coordination and fall recovery, while operating solely under velocity commands without relying on reference motions. Experimental results show that X-Loco achieves superior performance, demonstrated by tasks such as fall recovery and terrain traversal. Ablation studies further highlight that our framework effectively leverages specialist expertise and enhances learning efficiency.

Abstract:
This paper introduces Dr-BA, a first-of-its-kind radar bundle adjustment (BA) framework that operates directly on 2D spinning radar intensity images. Unlike camera or lidar sensors, radar is largely unaffected by precipitation, making it a critical modality for autonomous systems that require all-weather robustness. Existing state estimation approaches using spinning radar typically extract sparse point clouds from range-azimuth-intensity measurements and apply point cloud alignment techniques to estimate vehicle motion, scene structure, or to localize within an existing map. In contrast, Dr-BA uses the full radar returns from multiple scans to jointly estimate dense maps and sensor poses. By formulating the problem as a separable optimization, we derive an efficient and general solution that decouples pose estimation from mapping. In addition to solving the BA problem, this formulation naturally extends to direct radar-only localization (DRL) within a previously built map. Dr-BA achieves state-of-the-art radar-based BA and cross-session localization performance, demonstrated on more than 200 km of on-road data across five distinct routes.

Abstract:
Dexterous grasping is fundamental to robotics, yet data-driven grasp prediction heavily relies on large, diverse datasets that are costly to generate and typically limited to a narrow set of gripper morphologies. Analytical grasp synthesis can be used to scale data collection, but necessary simplifying assumptions often yield physically infeasible grasps that need to be filtered in high-fidelity simulators, significantly reducing the total number of grasps and their diversity. We propose a scalable generate-and-refine pipeline for synthesizing large-scale, diverse, and physically feasible grasps. Instead of using high-fidelity simulators solely for verification and filtering, we leverage them as an optimization stage that continuously improves grasp quality without discarding precomputed candidates. More specifically, we initialize an evolutionary search with a seed set of analytically generated, potentially suboptimal grasps. We then refine these proposals directly in a high-fidelity simulator (Isaac Sim) using an asynchronous, gradient-free evolutionary algorithm, improving stability while maintaining diversity. In addition, this refinement stage can be guided toward human preferences and/or domain-specific quality metrics without requiring a differentiable objective. We further distill the refined grasp distribution into a diffusion model for robust real-world deployment, and highlight the role of diversity for both effective training and during deployment. Experiments on a newly introduced Handles dataset and a DexGraspNet subset demonstrate that our approach achieves over 120 distinct stable grasps per object (a 1.7-6x improvement over unrefined analytical methods) while outperforming diffusion-based alternatives by 46-60% in unique grasp coverage.

Abstract:
Whole-arm manipulation involves direct contact with the environment while the robot completes a task by distributing contact across multiple links as contacts form, slide, and break. This setting breaks common implicit assumptions in many learning-based manipulation pipelines: arm configuration tightly couples motion and contact forces, contact state is partially observed under occlusion, and purely learned rollouts can become physically inconsistent under distribution shift because many multi-link contact configurations are sparsely represented in the data. To address this, we propose TACTIC (Tactile and Vision Conditioned Contact-Centric Control), a receding-horizon controller for whole-arm manipulation. TACTIC uses a contact-centric hybrid predictive model that combines RGB-D, distributed tactile sensing, and a compact 2D proximity representation. The model couples a learned, action-conditioned latent dynamics model with analytical kinematics through contact Jacobians, enabling rollouts of future contact configurations and interaction forces. TACTIC integrates these rollouts into a sampling-based MPC planner with contact-aware action sampling: contact Jacobian-based projections steer sampled action sequences toward force-modulating directions, and objectives defined over predicted proximity and interaction forces trade task progress against whole-arm force regulation. We evaluate TACTIC in simulation against state-of-the-art model-based and model-free methods, and perform ablations that isolate the contribution of each design choice. Across experiments, TACTIC consistently outperforms other methods. We further demonstrate real-world performance on a robot with distributed tactile sensing across three whole-arm manipulation tasks that require multi-contact trajectories: turning over and repositioning a manikin, and goal-reaching in a 3D dynamic maze.

Abstract:
Reliable robotic manipulation requires control policies that can accurately represent and adapt to uncertainty arising from contact-rich interactions. Modern data-driven methods mitigate uncertainty through large-scale training and computation, and degrade significantly in performance with limited number of training samples. By contrast, classical model-based controllers are computationally efficient and reliable, but their limited ability to represent task-relevant uncertainty can hinder performance in contact-rich interactions. In this work, we propose to expand the capabilities of model-based manipulation control through more flexible uncertainty modeling that retains performance while exactly adapting to uncertainty. Our approach casts the manipulation problem as a distributionally robust control optimization and proposes a novel deterministic formulation based on Stein variational inference that preserves performance while explicitly modeling task-sensitive parameter uncertainty. As a result, the derived controllers are more aware of task sensitivities to uncertainty, yielding high reliability without compromising performance. Experimental results demonstrate up to 3× improved robustness across a range of contact-rich manipulation tasks under broad parametric uncertainty, outperforming existing model-based control methods.

Abstract:
Generalist robot manipulation policies are becoming increasingly capable, but are limited in evaluation to a small number of hardware rollouts. This strong resource constraint in real-world testing necessitates both more informative performance measures and reliable and efficient evaluation procedures to properly assess model capabilities and benchmark progress in the field. This work presents a novel framework for robot policy comparison that is sample-efficient, statistically rigorous, and applicable to a broad set of evaluation metrics used in practice. Based on safe, anytime-valid inference (SAVI), our test procedure is sequential, allowing the evaluator to stop early when sufficient statistical evidence has accumulated to reach a decision at a pre-specified level of confidence. Unlike previous work developed for binary success, our unified approach addresses a wide range of informative metrics: from discrete partial credit task progress to continuous measures of episodic reward or trajectory smoothness, spanning both parametric and nonparametric comparison problems. Through extensive validation on simulated and real-world evaluation data, we demonstrate up to 70% reduction in evaluation burden compared to standard batch methods and up to 50% reduction compared to state-of-the-art sequential procedures designed for binary outcomes, with no loss of statistical rigor. Notably, our empirical results show that competing policies can be separated more quickly when using fine-grained task progress than binary success metrics.

Abstract:
Recent progress in vision–language–action (VLA) models has demonstrated strong potential for dual-arm manipulation, enabling complex behaviors and generalization to unseen environments. However, mainstream bimanual VLA formulations largely overlook the critical challenge of combinatorial diversity. Different pairings of single-arm behaviors can induce qualitatively distinct task behaviors, yet existing models do not explicitly account for this structure. We argue that effective bimanual VLAs should support skill reuse—the ability to recombine previously learned single-arm skills across novel left–right pairings—thereby avoiding the need to separately learn every possible combination. Current VLA designs entangle skills across arms, preventing such recomposition and limiting scalability. To address this limitation, we propose SkillVLA, a framework explicitly designed to enable skill reuse in dual-arm manipulation. Extensive experiments demonstrate that SkillVLA substantially improves skill composition, increasing overall success rate from 0% to 51%, and achieves strong performance on cooperative and long-horizon tasks.

Abstract:
Pretrained on large-scale and diverse datasets, VLA models demonstrate strong generalization and adaptability as general-purpose robotic policies. However, Supervised Fine-Tuning (SFT), which serves as the primary mechanism for adapting VLAs to downstream domains, requires substantial amounts of task-specific data and is prone to catastrophic forgetting. To address these limitations, we propose LifeLong-RFT, a simple yet effective Reinforcement Fine-Tuning (RFT) strategy for VLA models independent of online environmental feedback and pre-trained reward models. By integrating chunking-level on-policy reinforcement learning with the proposed Multi-Dimensional Process Reward (MDPR) mechanism, LifeLong-RFT quantifies the heterogeneous contributions of intermediate action chunks across three dimensions to facilitate policy optimization. Specifically, (1) the Quantized Action Consistency Reward (QACR) ensures accurate action prediction within the discrete action space; (2) the Continuous Trajectory Alignment Reward (CTAR) aligns decoded continuous action chunks with reference trajectories to ensure precise control; (3) the Format Compliance Reward (FCR) guarantees the structural validity of outputs. Comprehensive experiments across SimplerEnv, LIBERO, and real-world tasks demonstrate that LifeLong-RFT exhibits strong performance in multi-task learning. Furthermore, for continual learning on the LIBERO benchmark, our method achieves a 22% gain in average success rate over SFT, while effectively adapting to new tasks using only 20% of the training data. Overall, our method provides a promising post-training paradigm for VLAs.

Abstract:
While robotic manipulation capabilities have advanced rapidly, physical safety remains a major barrier to deploying household robots: task success is insufficient if the robot damages itself or its surroundings. Simulation offers a harm-free alternative to costly and dangerous real-world training and evaluation, yet existing simulators lack general mechanisms to detect, quantify, and represent damage. To address this gap, we introduce OOPSIEVERSE, a unified simulation framework and benchmark for damage-aware household manipulation. At a theoretical level, OOPSIEVERSE augments an existing Markov Decision Problem with additional damage-related observations, rewards and/or termination conditions based on user preferences. OOPSIEVERSE provides damage as an explicit, physically-grounded, and task-agnostic signal by converting sources such as contact forces, temperature changes, and liquid interactions into corresponding mechanical, thermal or fluid damage. OOPSIEVERSE comprises two core elements: (1) DAMAGESIM, a simulator-agnostic framework for detecting and quantifying damage during navigation and manipulation, and (2) a suite of household tasks designed to evaluate common damage modes and distinguish between task completion and safe execution. We demonstrate the generality of our framework by instantiating DAMAGESIM in two simulators with different physics backends, OmniGibson (Nvidia Omniverse) and RoboCasa (MuJoCo). We further showcase the utility of OOPSIEVERSE across multiple use cases, including (1) guiding safer demonstration collection via real- time damage feedback, (2) learning safer manipulation policies through damage-conditioned imitation learning and reinforcement learning, (3) benchmarking the safety of state-of-the-art Vision Language Action policies, and (4) improving real-world safety of sim-to-real transferred policies. Together, our results highlight the potential of OOPSIEVERSE as an open-source foundation for systematic, scalable research on safe robot manipulation. For code and additional information, please refer to https:\\oopsiverse-anon.github.io.

Abstract:
Real-world tasks involve nuanced combinations of goal and safety specifications, which often directly compete. In high dimensions, the challenge is exacerbated: formal automata become cumbersome, and the combination of sparse rewards tends to require laborious tuning. In this work, we consider the structure of the Bellman Value as a means to naturally organize the problem for improved automatic performance without introducing additional abstractions. Namely, we prove the Bellman Value for a complex task defined in temporal logic can be decomposed into a graph of Bellman Values , where the graph is connected by a set of well-studied Bellman equations (BEs): the Reach-Avoid BE, the Avoid BE, and a novel type, the Reach-Avoid-Loop BE. From this perspective, we design a specialized PPO variant, Value-Decomposition PPO ( VDPPO ) that uses a single learned representation by embedding the decomposed Value graph. We conduct a variety of simulated and real multi-objective experiments, including delivery and herding, to test our method on diverse high-dimensional systems involving heterogeneous teams and complex agents. Ultimately, we find this approach greatly improves performance over existing baselines, balancing safety and liveness automatically.

Abstract:
Conventional human-in-the-loop approaches typically involve users only when a robot encounters failure or uncertainty, treating humans primarily as tools to improve robot performance. However, in many human-centered robotics settings, interaction should support user engagement, keeping users meaningfully involved in decision-making rather than limiting them to failure-driven interventions. For many users, this cannot be achieved through limited, failure-driven interaction alone; they wish to remain involved in the robot's decision-making to sustain engagement throughout the task. This is particularly compelling in physical caregiving, where mobility limitations can reduce users' ability to intervene or modulate the robot's behavior in the moment. As a result, interaction policies that engage users only upon failure may further reduce engagement by relegating users to passive observers for long stretches of the task. For example, a user with mobility limitations may experience reduced engagement when being continuously and passively fed by a robot. At the same time, overly frequent interaction can be tiring and increase the user's workload. To address this trade-off, we propose Engagement-aware MPC (E-MPC), a user-engagement-aware method that plans interaction to maintain engagement while respecting a workload constraint. E-MPC leverages a user interaction dynamics model that captures how user engagement evolves as a function of both the frequency and type of interaction. Rather than requesting input only when difficulties arise during task execution, the robot proactively considers the user's preferred level of engagement throughout the task, balancing autonomy and interaction while ensuring task success. We evaluate E-MPC in simulation with several ablations and baseline comparisons. Baselines optimize for task success alone or jointly for user workload and task success. Results demonstrate the effectiveness of our approach across diverse user personas. In addition, we conduct a real-world user study with participants with emulated mobility limitations on a robot-assisted bite acquisition system, showing that E-MPC improves user experience while maintaining task success.

Abstract:
Leveraging pre-trained 2D image representations in behavior cloning policies has achieved great success and has become a standard approach for robotic manipulation. However, such representations fail to capture the 3D spatial information about objects and scenes that is essential for precise manipulation. In this work, we introduce Contrastive Learning for 3D Multi-View Action-Conditioned Robotic Manipulation Pretraining (CLAMP), a novel 3D pre-training framework that utilizes point clouds and robot actions. From the merged point cloud computed from RGB-D images and camera extrinsics, we re-render multi-view four-channel image observations with depth and 3D coordinates, including dynamic wrist views, to provide clearer views of target objects for high-precision manipulation tasks. The pre-trained encoders learn to associate the 3D geometric and positional information of objects with robot action patterns via contrastive learning on large-scale simulated robot trajectories. During encoder pre-training, we pre-train a Diffusion Policy to initialize the policy weights for fine-tuning, which is essential for improving fine-tuning sample efficiency and performance. After pre-training, we fine-tune the policy on a limited amount of task demonstrations using the learned image and action representations. We demonstrate that this pre-training and fine-tuning design substantially improves learning efficiency and policy performance on unseen tasks. Furthermore, we show that CLAMP outperforms state-of-the-art baselines across six simulated tasks and five real-world tasks.

Abstract:
We present LoTIS, a model for visual navigation that provides robot-agnostic image-space guidance by localizing a reference RGB trajectory in the robot’s current view, without requiring camera calibration, poses, or robot-specific training. Instead of predicting actions tied to specific robots, we predict the image-space coordinates of the reference trajectory as they would appear in the robot’s current view. This creates robot-agnostic visual guidance that easily integrates with local planning. Consequently, our model's predictions provide guidance zero-shot across diverse embodiments. By decoupling perception from action and learning to localize trajectory points rather than imitate behavioral priors, we enable a cross-trajectory training strategy that learns robust invariance to viewpoint and camera changes. We outperform state-of-the-art methods by 20-50 percentage points in success rate on forward navigation, and paired with a local planner we achieve 94-98% success rate across diverse sim and real environments. Furthermore, we achieve over 5× improvements on challenging tasks where baselines fail, such as backward traversal. The system is straightforward to use: we show how even a video from a handheld phone camera directly enables different robots to navigate to any point on the trajectory.

Abstract:
Designing learnable information-theoretic objectives for robot exploration remains challenging. Such objectives aim to guide exploration toward data that reduces uncertainty in model parameters, yet it is often unclear what information the collected data can actually reveal. Although reinforcement learning (RL) can optimize a given objective, constructing objectives that reflect parametric learnability is difficult in high-dimensional robotic systems. Many parameter directions are weakly observable or unidentifiable, and even when identifiable directions are selected, omitted directions can still influence exploration and distort information measures. To address this challenge, we propose Quasi-Optimal Experimental Design (QOED), an adaptive information objective grounded in optimal experimental design. QOED (i) performs eigenspace analysis of the Fisher information matrix to identify an observable subspace and select identifiable parameter directions, and (ii) modifies the exploration objective to emphasize these directions while suppressing nuisance effects from unidentifiable parameters. Under bounded nuisance influence and limited coupling between critical and nuisance directions, QOED provides a constant-factor approximation to the ideal information objective that explores all parameters. We evaluate QOED on simulated and real-world navigation and manipulation tasks, where identifiable-direction selection and nuisance suppression yield performance improvements of 35.23% and 21.98%, respectively. When integrated as an exploration objective in model-based policy optimization, QOED further improves policy performance over established RL baselines.

Abstract:
Position controllers have become the dominant interface for executing learned manipulation policies. Yet a critical design decision remains understudied: how should we choose controller gains for policy learning? The conventional wisdom is to select gains based on desired task compliance or stiffness. However, this logic breaks down when controllers are paired with state-conditioned policies: effective stiffness emerges from the interplay between learned reactions and control dynamics, not from gains alone. We argue that gain selection should instead be guided by learnability: how amenable different gain settings are to the learning algorithm in use. In this work, we systematically investigate how position controller gains affect three core components of modern robot learning pipelines: offline imitation learning, reinforcement learning from scratch, and sim-to-real transfer. Through extensive experiments across multiple tasks and robot embodiments, we find that: (1) offline imitation learning benefits from compliant and overdamped gain regimes, (2) reinforcement learning can succeed across all gain regimes given compatible hyperparameter tuning, and (3) sim-to-real transfer is harmed by stiff and overdamped gain regimes. These findings reveal that optimal gain selection depends not on the desired task behavior, but on the learning paradigm employed.

Abstract:
Despite strong multi-task pretraining, existing policies often exhibit poor task steerability. For example, a robot may fail to respond to a new instruction "put the bowl in the sink" when moving towards the oven, executing "close the oven", even though it can complete both tasks when executed separately. We propose ReSteer, a framework to quantify and improve task steerability in multitask robot policies. We conduct an exhaustive evaluation of state-of-the-art policies, revealing a common lack of steerability. We find that steerability is associated with limited overlap among training task trajectory distributions, and introduce a proxy metric to measure this overlap from policy behavior. Building on this insight, ReSteer improves steerability via three components: (i) a steerability estimator that identifies low-steerability states without full-rollout evaluation, (ii) a steerable data generator that synthesizes motion segments from these states, and (iii) a self-refinement pipeline that improves policy steerability using the generated data. In simulation on LIBERO, ReSteer improves steerability by 11% over 18k rollouts. In real-world experiments, we show that improved steerability is critical for interactive use, enabling users to instruct robots to perform any task at any time. We hope this work motivates further study on quantifying steerability and data collection strategies for large robot policies.

Abstract:
Scaling robot learning to long-horizon tasks remains a formidable challenge. While end-to-end policies often lack the structural priors needed for effective long-term reasoning, traditional neuro-symbolic methods rely heavily on hand-crafted symbolic priors. To address the issue, we introduce ENAP (Emergent Neural Automaton Policy), a framework that allows bi-level neuro-symbolic policy adaptively emerges from demonstrations. Specifically, we first employ adaptive clustering and an extension of the L algorithm to infer a Mealy state machine from visuomotor data, which serves as an interpretable high-level planner capturing latent task modes. Then, this discrete structure guides a low-level reactive residual network to learn precise continuous control via behavior cloning. By explicitly modeling the task policy with discrete transitions and continuous residuals, ENAP achieves high sample efficiency and interpretability without requiring task-specific labels. Extensive experiments on complex manipulation and long-horizon tasks demonstrate that ENAP outperforms state-of-the-art end-to-end VLA policies by up to 27% in low-data regimes, while offering a structured representation of robotic intent

Abstract:
Extrinsic dexterity leverages environmental contact to overcome the limitations of prehensile manipulation. However, achieving such dexterity in cluttered scenes remains challenging and underexplored, as it requires selectively exploiting contact among multiple interacting objects with inherently coupled dynamics. Existing approaches lack explicit modeling of such complex dynamics and therefore fall short in non-prehensile manipulation in cluttered environments, which in turn limits their practical applicability in real-world environments. In this paper, we introduce a Dynamics-Aware Policy Learning (DAPL) framework that can facilitate policy learning with a learned representation of contact-induced object dynamics in cluttered environments. This representation is learned through explicit world modeling and used to condition reinforcement learning, enabling extrinsic dexterity to emerge without hand-crafted contact heuristics or complex reward shaping. We evaluate our approach in both simulation and the real world. Our method outperforms prehensile manipulation, human teleoperation, and prior representation-based policy by over 25% in success rate on unseen simulated cluttered scenes with varying densities. Real-world success reaches around 50% across 10 cluttered scenes, while a practical grocery deployment further demonstrates robust sim-to-real transfer and applicability.

Abstract:
Policy evaluation is a fundamental component of the development and deployment pipeline for robotic policies. In modern manipulation systems, this problem is particularly challenging: rewards are often sparse, task progression of evaluation rollouts are often non-monotonic as the policies exhibit recovery behaviors, and evaluation rollouts are necessarily of finite length. This finite length introduces truncation bias, breaking the infinite-horizon assumptions underlying standard methods relying on Bellman equations/principle of optimality. In this work, we propose a framework for offline policy evaluation from sparse rewards based on a liveness-based Bellman operator. Our formulation interprets policy evaluation as a task-completion problem and yields a conservative fixed-point value function that is robust to finite-horizon truncation. We analyze the theoretical properties of the proposed operator, including contraction guarantees, and show how it encodes task progression while mitigating truncation bias. We evaluate our method on two simulated manipulation tasks using both a Vision–Language–Action model and a diffusion policy. Empirical results demonstrate that our approach more accurately reflects task progress and substantially reduces truncation bias, outperforming classical baselines such as TD(0) and Monte Carlo policy evaluation.

Abstract:
Tactile imaging seeks to reconstruct the internal structure of soft objects through touch sensing, with applications in medical diagnosis and robotic manipulation. Recent self-supervised learning approaches have shown promising results, but rely on global, unstructured representations and robot-controlled sensing, limiting generalization and practical use. We propose Local Encoder for Spatial Sensing (LESS), an object-centric tactile representation that exploits the local nature of touch. The tactile scene is modeled as a grid of recurrent encoders with local receptive fields, whose states are fused to reconstruct 2D or 3D images of internal structure. This compositional design enables strong generalization: models trained on single-inclusion phantoms accurately image objects with multiple inclusions and varying sizes. The local structure further supports spatial uncertainty estimation. In addition, we enable hand-held tactile imaging via external pose tracking and human-like palpation data, and extend tactile imaging to full 3D reconstruction.

Abstract:
Planning under uncertainty for real-world robotics tasks, such as autonomous driving, requires reasoning in enormous high-dimensional belief spaces, rendering the problem computationally intensive. While parallelization offers scalability, existing hybrid CPU-GPU solvers face critical bottlenecks due to host-device synchronization latency and branch divergence on SIMT architectures, limiting their utility for real-time planning and hindering real-robot deployment. We present Vec-QMDP, a CPU-native parallel planner that aligns POMDP search with modern CPUs' SIMD architecture, achieving 227×--1073× speedup over state-of-the-art serial planners. Vec-QMDP adopts a Data-Oriented Design (DOD), refactoring scattered, pointer-based data structures into contiguous, cache-efficient memory layouts. We further introduce a hierarchical parallelism scheme: distributing sub-trees across independent CPU cores and SIMD lanes, enabling fully vectorized tree expansion and collision checking. Efficiency is maximized with the help of UCB load balancing across trees and a vectorized STR-tree for coarse-level collision checking. Evaluated on large-scale autonomous driving benchmarks, Vec-QMDP achieves state-of-the-art planning performance with millisecond-level latency, establishing CPUs as a high-performance computing platform for large-scale planning under uncertainty.

Abstract:
Robotic systems that interact with the physical world must reason about kinematic and dynamic constraints imposed by their own embodiment, their environment, and the task at hand. We introduce KinDER, a benchmark for Kinematic and Dynamic Embodied Reasoning that targets physical reasoning challenges arising in robot learning and planning. KinDER comprises 25 procedurally generated environments, a Gymnasium-compatible Python library with parameterized skills and demonstrations, and a standardized evaluation suite with 8 implemented baselines spanning task and motion planning, imitation learning, reinforcement learning, and foundation-model-based approaches. The environments are designed to isolate five core physical reasoning challenges: basic spatial relations, nonprehensile multi-object manipulation, tool use, combinatorial geometric constraints, and dynamic constraints, disentangled from perception, language understanding, and application-specific complexity. Empirical evaluation shows that existing methods struggle to solve many of the environments, indicating substantial gaps in current approaches to physical reasoning. We additionally include real-to-sim-to-real experiments on a mobile manipulator to assess the correspondence between simulation and real-world physical interaction. KinDER is fully open-sourced and intended to enable systematic comparison across diverse paradigms for advancing physical reasoning in robotics.

Abstract:
To achieve human-like skin tactile perception with super-resolution, the method of introducing a soft layer on sensing array has attracted increasing attention. Due to the limitations of sensing units principle, most existing tactile sensors can only sense normal force. However, multi-dimensional force information is important for robot manipulation. To address this, we propose a tactile sensing unit based on a tiny monolithic tri-cantilever structure that decouples three-dimensional force. A reconstruction algorithm combined both model-based and learning-based approaches is then proposed to detect the three-dimensional force applied to the sensing unit. These units are arranged in an array and covered with a soft silicone layer which induces traction-coupling effects. By leveraging deep learning, our tactile sensor can estimate the magnitude and position of external three-dimensional force with super-resolution. Experiments have shown that our tactile sensor achieves a Mean Absolute Error (MAE) of 0.19\,N for three-dimensional force estimation and 0.49\,mm for contact localization. Notably, this corresponds to a 26-fold improvement in spatial resolution, surpassing the state-of-the-art literature. Then the benefits and potential applications of our proposed sensors are validated in several tasks, including the teleoperative transfer of a test tube into a rack and stable robotic grasping under external interference. These demonstrate the practicality of our design and provide new solutions for tactile sensors.

Abstract:
Many useful robot tasks require attending to the history of past observations. For example, finding an item in a room requires remembering which places have already been searched. However, the best-performing robot policies typically condition only on the current observation, limiting their applicability to such tasks. Naively conditioning on past observations often fails due to spurious correlations: policies latch onto incidental features of training histories that do not generalize to out-of-distribution trajectories upon deployment. In this paper, we analyze why policies latch onto these spurious correlations. We find that this problem arises because of limited coverage over the space of possible histories during training, which grows exponentially with horizon. Existing regularization techniques provide inconsistent benefits across tasks, as they do not fundamentally address this coverage problem. Motivated by these findings, we propose Big Picture Policies (BPP), an approach that conditions on a minimal set of meaningful keyframes detected by a vision-language model. By projecting diverse rollouts onto a compact representation of task-relevant events, BPP substantially reduces distribution shift between training and deployment, without sacrificing expressivity. We evaluate BPP on four challenging real-world manipulation tasks and three simulation tasks, all requiring history conditioning. BPP achieves 70% higher success rates than the best comparison on real-world evaluations.

Abstract:
Efficient exploration remains a bottleneck in reinforcement learning (RL), particularly for long-horizon, high-dimensional tasks. While recent methods leverage pre-trained policies for guidance, they are often constrained by the base policy's original behavior distribution. We introduce Timestep Modulated Reinforcement Learning (TMRL), a framework that enables agents to explore dynamically beyond these boundaries. TMRL leverages a forward diffusion process to inject noise into the context of a pre-trained policy, effectively aliasing nearby states to facilitate shared exploration modes. By training an RL policy to modulate the diffusion timestep at deployment, the agent can adaptively control conditioning strength, balancing marginal and conditional behaviors. Experimental results in navigation and robotic manipulation demonstrate that TMRL significantly outperforms existing baselines, proving that timestep modulation is a robust mechanism for adapting action sequences to novel tasks.

Abstract:
Embodied World Models (EWMs) have emerged as a scalable and risk-free paradigm for evaluating Vision-Language-Action (VLA) systems. However, their reliability as evaluation benchmarks is often limited by the representation gap between low-dimensional actions and high-dimensional video synthesis. This gap leads to a lack of geometric correspondence, manifesting as accumulated trajectory drift and inconsistent object-robot interactions in long-horizon rollouts. To bridge this gap, we propose ViPSim, a framework that achieves consistent long-horizon generation through the synergistic collaboration of Visual and Parameter Spaces. We define the Visual Space as a domain of explicit spatial priors, integrating pixel-aligned projections of actions, camera perspectives, depth-informed scene geometry, and robot morphological masks to provide dense structural constraints. Concurrently, the Parameter Space is defined as a domain of numerical drivers that injects raw action sequences and camera matrices to provide precise motion guidance. By unifying these two spaces, ViPSim ensures that the generated states are simultaneously constrained by geometric boundaries and driven by precise numerical commands. Extensive experiments demonstrate that ViPSim is backbone-agnostic and significantly enhances trajectory consistency. Notably, our approach exhibits emergent capabilities in generating complex interactions with deformable objects (e.g., cloth folding) and maintains robust performance in out-of-distribution and cross-embodiment scenarios, providing a high-fidelity foundation for the automated evaluation of embodied intelligence.

Abstract:
Simulation-to-real transfer remains a central challenge in robotics, as mismatches between simulated and real-world dynamics often lead to failures. While reinforcement learning offers a principled mechanism for adaptation, existing sim-to-real finetuning methods struggle with exploration and long-horizon credit assignment in the low-data regimes typical of real-world robotics. We introduce \textttSimulation Distillation (\textttSimDist), a sim-to-real framework that distills structural priors from a simulator into a latent world model and enables rapid real-world adaptation via online planning and supervised dynamics finetuning. By transferring reward and value models directly from simulation, \textttSimDist provides dense planning signals from raw perception without requiring value learning during deployment. As a result, real-world adaptation reduces to short-horizon system identification, avoiding long-horizon credit assignment and enabling fast, stable improvement. Across precise manipulation and quadruped locomotion tasks, \textttSimDist substantially outperforms prior methods in data efficiency, stability, and final performance.

Abstract:
Real-time whole-body teleoperation is a critical method for humanoid robots to perform complex tasks in unstructured environments. However, developing a unified controller that robustly supports diverse human motions remains a significant challenge. Existing methods typically distill multiple expert policies into a single general policy, which often inevitably leads to performance degradation, particularly on highly dynamic motions. This paper presents TeleGate, a unified whole-body teleoperation framework for humanoid robots that achieves high-precision tracking across various motions while avoiding the performance loss inherent in knowledge distillation. Our key idea is to preserve the full capability of domain-specific expert policies by training a lightweight gating network, which dynamically activates experts in real-time based on proprioceptive states and reference trajectories. Furthermore, to compensate for the absence of future reference trajectories in real-time teleoperation, we introduce a VAE-based motion prior module that extracts implicit future motion intent from historical observations, enabling anticipatory control for motions requiring prediction such as jumping and standing up. We conducted empirical evaluations in simulation and also deployed our technique on the Unitree G1 humanoid robot. Using only 2.5 hours of motion capture data for training, our TeleGate achieves high-precision real-time teleoperation across diverse dynamic motions (e.g., running, fall recovery, and jumping), significantly outperforming the baseline methods in both tracking accuracy and success rate.

Abstract:
Learning agile humanoid behaviors from human motion offers a powerful route to natural, coordinated control, but existing approaches face a persistent trade-off: reference-tracking policies are often brittle outside the demonstration dataset, while purely task-driven Reinforcement Learning (RL) can achieve adaptability at the cost of motion quality. We introduce a unified multi-task RL framework that bridges this gap by treating reference motion as a prior for behavioral shaping rather than a deployment-time constraint. A single goal-conditioned policy is trained jointly on two tasks that share the same observation and action spaces, but differ in their initialization schemes, command spaces, and reward structures: (i) a reference-guided imitation task in which reference trajectories define dense imitation rewards but are not provided as policy inputs, and (ii) a goal-conditioned generalization task in which goals are sampled independently of any reference and where rewards reflect only task success. By co-optimizing these objectives within a shared formulation, the policy acquires structured, human-like motor skills from dense reference supervision while learning to adapt these skills to novel goals and initial conditions. This is achieved without adversarial objectives, explicit trajectory tracking, phase variables, or reference-dependent inference. We evaluate the method in a challenging box-based parkour playground that demands diverse athletic behaviors (e.g., jumping and climbing), and show that the learned controller transfers beyond the reference distribution while preserving motion naturalness. Finally, we demonstrate long-horizon behavior generation by composing multiple learned skills, illustrating the flexibility of the learned polices in complex scenarios.

Abstract:
Achieving robust vision-based humanoid locomotion remains challenging due to two fundamental issues: the sim-toreal gap introduces significant perception noise that degrades performance on fine-grained tasks, and training a unified policy across diverse terrains is hindered by conflicting learning objectives. To address these challenges, we present an end-to-end framework for vision-driven humanoid locomotion. For robust sim-to-real transfer, we develop a high-fidelity depth sensor simulation that captures stereo matching artifacts and calibration uncertainties inherent in real-world sensing. We further propose a vision-aware behavior distillation approach that combines latent space alignment with noise-invariant auxiliary tasks, enabling effective knowledge transfer from privileged height maps to noisy depth observations. For versatile terrain adaptation, we introduce terrain-specific reward shaping integrated with multi-critic and multi-discriminator learning, where dedicated networks capture the distinct dynamics and motion priors of each terrain type. We validate our approach on two humanoid platforms equipped with different stereo depth cameras. The resulting policy demonstrates robust performance across diverse environments, seamlessly handling extreme challenges such as high platforms and wide gaps, as well as fine-grained tasks including bidirectional long-term staircase traversal.

Abstract:
Multi-robot teams are increasingly gaining attention due to their ability to scale up in terms of task workloads and complexities. However, existing approaches struggle with three key limitations: the inability of unimodal policies to capture multi-modal joint strategies, the rigidity of fixed policies against dynamic execution-time requirements, and the difficulty of resolving conflicting objectives during deployment. To address these challenges, we present Collaborative Flow Policy Guidance (CFPG), a novel framework that enables the modification of existing collaborative multi-robot policies by composing objectives during execution time. First, we introduce Multi-Agent Flow Policy Optimization (MAFPO) to learn robust, multi-modal collaborative policies in a fully on-policy manner without offline datasets. Second, we enable training-free adaptation to new execution-time objectives by leveraging flow matching guidance to steer actions toward user-specified goals. Third, during guidance we employ a hierarchical gradient projection mechanism to resolve conflicts among the nominal objective and execution objectives. We theoretically analyze CFPG and demonstrate that it achieves superior performance and robustness across multiple simulation environments and real robots and show that CFPG surpasses state-of-the-art methods.

Abstract:
Linear Temporal Logic (LTL) is a formal language that can be used to specify robot behaviors and goal states. We extend LTL to enable the specification of complex cooperative behaviors in a large multi-agent system (MAS). The extended language, TeamingLTL, models features commonly present in MAS such as partial observability and cooperation between agents. This enables a user to specify behaviors or goals while deferring the realization of those states to a separate, automated task planning and allocation system. We show how to specify the behavior of a MAS using TeamingLTL and use this formulation in program verification and the creation of correct-by-design controllers.

Abstract:
Knowledge distillation (KD) has the potential to accelerate multi-agent reinforcement learning (MARL) by employing a centralized teacher for decentralized students. However, centralized teachers in MARL often fail because decentralized student exploration induces out-of-distribution (OOD) state distributions the teacher was never trained on, compounded by partial observability, which creates observation mismatches between teacher and students at execution time. We propose HINT (Hierarchical INteractive Teacher-based transfer), a novel KD framework for MARL in a centralized training, decentralized execution setup. By leveraging hierarchical RL, HINT provides a scalable, high-performing teacher. Pseudo off-policy RL treats student trajectories as additional training data for the teacher, allowing it to adapt its policy to student-induced state distributions. Performance-based filtering removes teacher guidance that depends on centralized observations unavailable to decentralized students, retaining only outcome-relevant signals. Across FireCommander and MARINE, HINT consistently outperforms state-of-the-art online MARL baselines, improving task success rates by 60%–165%.

Abstract:
The challenge of generating reliable local plans has long hindered practical applications in highly cluttered and dynamic environments. Key fundamental bottlenecks include acquiring large-scale expert demonstrations across diverse scenes and improving learning efficiency with limited data. This paper proposes SanD-Planner, a sample-efficient diffusion-based local planner that conducts depth image-based imitation learning within the clamped B-spline space. By operating within this compact space, the proposed algorithm inherently yields smooth outputs with bounded prediction errors over local supports, naturally aligning with receding-horizon execution. Integration of an ESDF-based safety checker with explicit clearance and time-to-completion metrics further reduces the training burden associated with value-function learning for feasibility assessment. Experiments show that training with 500 episodes (merely 0.25% of the demonstration scale used by the baseline), SanD-Planner achieves state-of-the-art performance on the evaluated open benchmark, attaining success rates of 90.1% in simulated cluttered environments and 72.0% in indoor simulations. The performance is further proven by demonstrating zero-shot transferability to realistic experimentation in both 2D and 3D scenes. The dataset and pre-trained models will also be open-sourced.

Abstract:
The pursuit of general-purpose robotics has yielded impressive foundation models, yet simulation-based benchmarking remains a bottleneck due to rapid performance saturation and a lack of true generalization testing. Existing benchmarks often exhibit significant domain overlap between training and evaluation, trivializing success rates and obscuring insights into robustness. We introduce RoboLab, a simulation benchmarking framework designed to address these challenges. Concretely, our framework is designed to answer two questions: (1) to what extent can we understand the performance of a real-world policy by analyzing its behavior in simulation, and (2) which external factors most strongly affect that behavior under controlled perturbations. First, RoboLab enables human-authored and LLM-enabled generation of scenes and tasks in a robot- and policy-agnostic manner within a physically realistic and photorealistic simulation. With this, we propose the Robot Question and Answering (RQA) benchmark, consisting of∼80 tasks categorized into four task axes: visual, procedural, relational, and complexity. Second, we introduce a systematic analysis of real-world policies that quantify both their performance and the sensitivity of their behavior to controlled perturbations, indicating that high-fidelity simulation can serve as a proxy for analyzing performance and its dependence on external factors. Evaluation with RoboLab exposes significant performance gap in current state-of-the-art models. By providing granular metrics and a scalable toolset, RoboLab offers a scalable framework for evaluating the true generalization capabilities of task-generalist robotic policies.

Abstract:
Simulation-based reinforcement learning (RL) is central for robotic control when expert demonstrations are unavailable. However, scaling RL to high-dimensional robots remains challenging. On-policy methods such as PPO are reliable but require large amounts of simulation because they discard past data. Off-policy methods can reuse experience and are more sample-efficient, but they often become unstable in high-dimensional control due to critic errors that are amplified during bootstrapped updates. We introduce FlashSAC, a fast and stable off-policy RL algorithm for high-dimensional robotic control. FlashSAC improves training stability in two ways: (1) it explicitly bounds weight, feature, and gradient norms to limit critic error amplification, and (2) it increases data coverage through large-scale parallel simulation, a high-capacity replay buffer, and strong exploration. These design choices preserve the sample efficiency of off-policy learning while improving training stability. Across 50+ state-based and vision-based tasks in 10 simulators, FlashSAC consistently surpasses PPO and strong off-policy baselines in both final performance and wall-clock efficiency, with larger gains on higher-dimensional tasks. In sim-to-real humanoid walking, FlashSAC reduces training time from hours to minutes while maintaining stable real-world deployment. Our results show that stabilizing off-policy learning enables scalable sim-to-real RL for high-dimensional robotic systems.

Abstract:
Multi-step motion prediction for continuum robots is difficult, especially under actuation distribution shift, where error accumulation can distort the predicted steady response and destabilize rollouts. This paper introduces a hybrid equilibrium-anchored plus residual-learning framework for a tendon-driven 3D continuum arm that makes steady behavior explicit. An equilibrium prior is learned from inexpensive static equilibrium data and used in a contractive update that continuously pulls predictions toward the equilibrium estimate, improving rollout stability. A lightweight feature-lifted residual model, linear in parameters, learns the remaining one-step mismatch from dynamic trajectory data, recovering transient dynamics. The approach is validated on 200-step rollouts under actuation that is stronger and faster than in training. The Hybrid reduces backbone position RMSE by 26% and tip position RMSE by 27%, producing consistent accuracy gains over prior-only and residual-only predictors while remaining stable across all tested trajectories. The same proposed model also improves robustness on standard nonlinear benchmarks against a combined Koopman baseline under matched evaluation protocols.

Abstract:
In various surgical procedures, regions of interest (ROIs) such as organs or lesions are often occluded by overlying tissues, requiring surgeons to achieve adequate exposure for precise intervention. However, the irregular geometry, nonlinear biomechanical properties of overlying tissues, and limited intraoperative visibility of the ROI pose significant challenges to the autonomous execution of tissue retraction. To address this, we formulate a realistic model of the tissue retraction task and propose a learning-based adaptive control framework for achieving ROI exposure. The method optimizes control inputs online by monitoring changes in the visual boundary of the tissue, while leveraging a deep deformation estimation model trained on simulation data to identify the optimal grasping point and ensure the convergence and safety of the adaptive controller. Through simulations and real-world experiments on different deformable materials, it has been demonstrated that this framework exhibits zero-shot adaptation to similar tasks and can complete the autonomous retraction process, from initial grasp selection to full ROI exposure. Therefore, it has the potential to be applied in actual surgical assistance scenarios.

Abstract:
Semidefinite programming (SDP) provides a principled framework for convex relaxations of nonconvex geometric constraints in motion planning, yet existing solvers are too computationally expensive for real-time control, particularly on resource-constrained embedded systems. To address this gap, we introduce TinySDP, the first semidefinite programming solver designed for embedded systems, enabling real-time model-predictive control (MPC) with formal safety guarantees on microcontrollers for problems with nonconvex obstacle constraints. Our approach integrates positive-semidefinite cone projections into a cached-Riccati-based ADMM solver, leveraging computational structure for embedded tractability. We pair this solver with an a posteriori rank-1 certificate that converts relaxed solutions into explicit geometric guarantees at each timestep. On challenging benchmarks, e.g., cul-de-sac and dynamic obstacle avoidance scenarios that induce failures in local methods, TinySDP achieves collision-free navigation with up to 73% shorter paths than state-of-the-art baselines. We validate our approach on a Crazyflie quadrotor, demonstrating that certifiable semidefinite constraints can be enforced at real-time rates for agile embedded robotics.

Abstract:
Many Inertial Navigation Systems (INS) use Global Navigation Satellite System (GNSS) position as the primary measurement to drive filter performance and bound error growth. However, commercial-grade GNSS receivers introduce unknown measurement delays ranging from 50 ms to 300 ms depending on sensor quality and operating mode. Such time delays can significantly degrade INS performance unless they are explicitly compensated for. Existing algorithms commonly estimate this delay offline, run the filter concurrently with GNSS measurements using buffered Inertial Measurement Unit (IMU) data, and predict the current state by forward-integrating buffered inertial measurements via IMU preintegration. The state-of-the-art online method is an Extended Kalman Filter (EKF) that explicitly models the time delay as a state parameter, which defines the preintegration duration. This paper introduces a novel geometric framework for modeling time-delayed INS, in which Galilean symmetry is leveraged to provide a joint representation of space and time for consistent state estimation. An Equivariant Filter (EqF) is derived for the coupled estimation of navigation states and time delay. Validation is performed on two fixed-wing Uncrewed Aerial Vehicles (UAV) with GNSS time lags of 90 ms and 120 ms. The test flights last two to three minutes. Simulations further investigate delays up to 500 ms and provide a statistical comparison against the state-of-the-art EKF. Results show that the EqF preserves accuracy and consistency, while the EKF lacks consistency and its performance degrades significantly with increasing measurement delays.

Abstract:
Hazardous environments are characterized by uncertain, rapidly changing conditions, where autonomous robots are responsible for making safety-critical navigation decisions to ensure a safer path. When navigation relies solely on reactive perception, a robot may encounter dangerous situations before it has sufficient time to respond effectively. In real-world settings, however, hazards such as smoke, heat, flooding, or structural instability can be observed directly by the environment itself through cameras and sensors. This creates an opportunity to anticipate risk in advance, before a robot enters unsafe regions. In this work, we focus on enabling proactive and risk-aware robot navigation through environment-centric perception. We present a unified multimodal spatiotemporal model that estimates navigation risk by fusing environmental visual and sensory data through early integration, producing instantaneous probabilistic risk estimates. To avoid reacting to short-lived disturbances and sensor noise, these estimates are incorporated into a Bayesian filtering process that treats risk as an underlying latent state and maintains a stable belief over evolving environmental hazards. The resulting risk belief is integrated into the robot’s navigation stack as a cost term, guiding the planner toward safer routes and allowing the robot to avoid hazardous areas before reaching them. We evaluate the proposed approach on a physical multi-hazard testbed that includes smoke, heat, water presence, vibration, and structural obstacles. Experimental results show reduced hazard exposure and safer navigation behavior compared to risk-unaware baselines, highlighting the benefits of environment-centric risk inference for autonomous robot navigation.

Abstract:
Point clouds are a fundamental representation for robotic perception tasks such as localization, mapping, and object pose estimation. However, LiDAR-acquired point clouds are inherently sparse and non-uniform, providing incomplete observations of the underlying geometry. Such sparsity and non-uniformity hinder reliable geometric reasoning, leading to degraded performance in downstream perception tasks. To mitigate these issues, prior work has attempted to compensate for the sparsity and non-uniformity of point clouds by estimating point cloud geometry. However, in the absence of an explicit model of point cloud geometry, existing approaches have predominantly relied on either hand-crafted statistics of local point distributions or end-to-end supervised deep learning, which often suffer from limited scalability or require large amounts of accurately labeled training data. To address these challenges, we explicitly model and estimate point cloud geometry under a principled mathematical formulation. Theoretically, we represent the point cloud geometry as a statistical manifold induced by a family of Gaussian distributions that captures the local geometry of each point. Building on this formulation, we design a probabilistic model that predicts per-point local geometry in the form of a Gaussian distribution. Practically, we introduce a deep neural network to instantiate the estimation of these Gaussian distributions, and term the resulting estimator as Point-to-Ellipsoid (POLI). By consistently estimating point-wise local geometry across diverse point clouds, POLI learns a mapping between point cloud observations and the statistical manifolds that represent their underlying geometry. Importantly, this mapping is learned in a self-supervised manner, removing the reliance on labeled data while maintaining strong geometric inductive biases. The resulting representation integrates seamlessly into existing robotic perception pipelines without requiring architectural modifications. Extensive experiments demonstrate that the proposed theory and practice enable accurate and robust estimation of point cloud geometry and consistently improve performance across a wide range of robotic perception tasks.

Abstract:
Dexterous manipulation is physics-intensive and highly sensitive to modeling errors and perception noise, making sim-to-real transfer prohibitively challenging. Domain randomization (DR) is commonly used to improve the robustness of learned policies for such tasks, but conventional DR randomizes one instance per episode, offering very limited exposure to the variability of real-world dynamics. To this end, we propose Domain-Randomized Instance Set (DRIS), which represents and propagates a set of randomized instances simultaneously, providing richer approximation of uncertain dynamics and enabling policies to learn actions that account for multiple possible outcomes. Supported by theoretical analysis, we show that DRIS yields more robust policies and alleviates the need for real-world fine-tuning, even with a modest number of instances (e.g., 10). We demonstrate this on a challenging reactive catching task. Unlike traditional catching setups that use end-effectors designed to mechanically stabilize the object (e.g., curved or enclosing surfaces), our system uses a flat plate that offers no passive stabilization, making the task highly sensitive to noise and requiring rapid reactive motions. The learned policies exhibit strong robustness to uncertainties and achieve reliable zero-shot sim-to-real transfer.

Abstract:
Differentiable simulators have advanced policy learning and model-based control across diverse robotic tasks. To date, actuator dynamics remain underexplored and are a major source of sim-to-real error, especially on low-cost platforms where the linear current–torque model τ = K_tI breaks down under commanded-target tracking due to friction, hysteresis, backlash, and thermal effects. Beyond forward dynamics, accurate actuator models also support force perception, which is crucial for jointly modeling force and position control in manipulation tasks. We present NeuralActuator, a neural actuator model that jointly predicts (i) torque prediction to capture the full nonlinear and time-varying current–torque relationship on low-cost servos (ii) external contact forces as well as force detection gates for sensorless force perception (iii) motor conditions indicating their operating regime. We introduce a twin-arm teleoperation system that collects motor states alongside ground-truth forces from interactions and known external forces, contributing a dataset named Neural Actuation Dataset (NAD). NeuralActuator is trained through differentiable simulation using only pose trajectories as supervision, eliminating the need for torque sensors. A Transformer-based architecture captures temporal dependencies while enabling efficient real-time inference. We validate NeuralActuator on a low-cost 5-DoF platform and show that it enables accurate dynamics modeling, sensorless force estimation, motor condition estimation, and improved behavior cloning control when used as a pretrained module. Our system and datasets will be released.

Abstract:
Contact-implicit trajectory optimization (CITO) has attracted growing attention as a unified framework for planning and control in contact-rich robotic tasks. Recent approaches have demonstrated promising results in manipulation and locomotion without requiring a prescribed contact-mode schedule. It is well known that the underlying mathematical programs with complementarity constraints (MPCCs) remain numerically ill-conditioned, and systematic, scalable solution strategies for CITO remain an active area of research. More efficient and principled solvers that can handle contact constraints are therefore essential to broaden the applicability of CITO. In this work, we develop an augmented-Lagrangian approach to CITO for solving general MPCCs with convergence guarantees. The method can be interpreted as identifying the implicit active contact set on the fly during the trajectory optimization (TO) iterations; we call this approach IMPACT (IMPlicit ACtive-set Trajectory optimization). We provide an efficient C++ implementation tailored to trajectory-optimization workloads and evaluate it on the open-source CITO and contact-implicit model predictive control (CI-MPC) benchmarks. On CITO, IMPACT achieves 2.9x-70x speedups over strong baselines (geometric mean 13.8x). On CI-MPC, we show improved control quality for contact-rich trajectories on dexterous manipulation tasks in simulation. Finally, we demonstrate the proposed method on real robotic hardware on a T-shaped object pushing task.

Abstract:
A signed distance field (SDF) is a widely used geometric representation for robust collision detection between complex geometries, which is crucial for contact-rich simulations. While numerous works have studied SDF representations and SDF-based collision detection, achieving both memory efficiency and high accuracy while maintaining scalability remains a challenge. In this paper, we propose a novel SDF representation, Tetra-Encoded SDF (TE-SDF), which combines the adaptive spatial discretization of a tetrahedral mesh with exact-distance evaluation localized to each tetrahedron by encoding a compact set of candidate surface faces per tetrahedron. We demonstrate the effectiveness of TE-SDF in contact-rich simulation by implementing a fully GPU-accelerated collision detector based on TE-SDF and integrating it into a GPU-accelerated simulation framework. Our results show that TE-SDF enables memory-efficient, accurate, and scalable collision detection, expanding the domain of robotic simulation scenarios that can be handled in practice.

Abstract:
Estimating simulation-ready scenes from real-world observations is crucial for downstream planning and policy learning tasks. Regretfully, existing methods struggle in cluttered environments, often exhibiting prohibitive computational cost, poor robustness, and restricted generality when scaling to multiple interacting objects. We propose a unified optimization-based formulation for real-to-sim scene estimation that jointly recovers the shapes and poses of multiple rigid objects under physical constraints. Our method is built on two key technical innovations. First, we leverage the recently introduced shape-differentiable contact model, whose global differentiability permits joint optimization over object geometry and pose while modeling inter-object contacts. Second, we exploit the structured sparsity of the augmented Lagrangian Hessian to derive an efficient linear system solver whose computational cost scales favorably with scene complexity. Building on this formulation, we develop an end-to-end real-to-sim scene estimation pipeline that integrates learning-based object initialization, physics-constrained joint shape-pose optimization, and differentiable texture refinement. Experiments on cluttered scenes with up to 5 objects and 22 convex hulls demonstrate that our approach robustly reconstructs physically valid, simulation-ready object shapes and poses.

Abstract:
Robust state estimation for highly dynamic motion of legged robots remains challenging, especially in dynamic, contact-rich scenarios. Traditional approaches often rely on binary contact states that fail to capture the nuances of partial contact or directional slippage. This paper presents CoCo-InEKF, a differentiable Invariant Extended Kalman Filter that utilizes continuous contact velocity covariances instead of binary contact states. These learned covariances allow the method to dynamically modulate contact confidence, representing states ranging from firm contact to directional slippage or no contact. To predict these covariances for a set of predefined contact candidate points, we employ a lightweight neural network trained end-to-end using a state-error loss. This approach eliminates the need for heuristic ground-truth contact labels. In addition, we propose an automated contact candidate selection procedure and demonstrate that our method is insensitive to their exact placement. Experiments on a bipedal robot demonstrate superior linear velocity estimation compared to baseline methods, enabling robust execution of challenging motions, including dancing and complex ground interactions.

Abstract:
Manipulating deformable and fragile objects remains a fundamental challenge in robotics due to complex contact dynamics and strict requirements on object integrity. Existing approaches typically optimize either end-effector design or control strategies in isolation, limiting achievable performance. In this work, we present the first co-design framework that jointly optimizes end-effector morphology and manipulation control for deformable and fragile object manipulation. We introduce (1) a latent diffeomorphic shape parameterization enabling expressive yet tractable end-effector geometry optimization, (2) a stress-aware bi-level co-design pipeline coupling morphology and control optimization, and (3) a privileged-to-pointcloud policy distillation scheme for zero-shot real-world deployment. We evaluate our approach on challenging food manipulation tasks, including grasping and pushing jelly and scooping fillets. Simulation and real-world experiments demonstrate the effectiveness of the proposed method.

Abstract:
We introduce Phantom Twist, a type of single-propeller UAV designed to achieve low visibility through high-speed spinning and the exploitation of motion blur. We develop a two-stage automated design pipeline that optimizes the placement of functional components including batteries, control PCB, motor-propeller assembly, and counterweights. The pipeline minimizes visibility as measured by a human-aligned perceptual metric (LPIPS) while strictly satisfying inertial and aerodynamic constraints required for stable flight. We validate this approach through fabrication and flight testing of multiple prototypes. These tests confirm that our pipeline produces stable, controllable designs and that the optimized UAV exhibits significantly reduced visual perceptibility compared to conventional quadcopters.

Abstract:
We introduce SoftAct, a framework for teaching soft robot hands to perform human-like manipulation skills by explicitly reasoning about contact forces. Leveraging immersive virtual reality, our system captures rich human demonstrations, including hand kinematics, object motion, dense contact patches, and detailed contact force information. Unlike conventional approaches that retarget human joint trajectories, SoftAct employs a two-stage, force-aware retargeting algorithm. The first stage attributes demonstrated contact forces to individual human fingers and allocates robot fingers proportionally, establishing a force-balanced mapping between human and robot hands. The second stage performs online retargeting by combining baseline end-effector pose tracking with geodesic-weighted contact refinements, using contact geometry and force magnitude to adjust robot fingertip targets in real time. This formulation enables soft robotic hands to reproduce the functional intent of human demonstrations while naturally accommodating extreme embodi- ment mismatch and nonlinear compliance. We evaluate SoftAct on a suite of contact-rich manipulation tasks using a custom non-anthropomorphic pneumatic soft robot hand. SoftAct’s controller reduces fingertip trajectory tracking RMSE by up to 55% and reduces tracking variance by up to 69% compared to kinematic and learning-based baselines. At the policy level, SoftAct achieves consistently higher success in zero-shot real world deployment and in simulation. These results demonstrate that explicitly modeling contact geometry and force distribution is essential for effective skill transfer to soft robotic hands, and cannot be recovered through kinematic imitation alone.

Abstract:
Recent robot foundation models largely rely on large-scale behavior cloning, which imitates expert actions but discards transferable dynamics knowledge embedded in heterogeneous embodied data. While the Unified World Model (UWM) formulation has the potential to leverage such diverse data, existing instantiations struggle to scale to foundation-level due to coarse data usage and fragmented datasets. We introduce LDA-1B , a robot foundation model that scales through universal embodied data ingestion by jointly learning dynamics, policy, and visual forecasting, assigning distinct roles to data of varying quality. To support this regime at scale, we assemble and standardize EI-30k , an embodied interaction dataset comprising over 30k hours of human and robot trajectories in a unified format. Scalable dynamics learning over such heterogeneous data is enabled by operating in a structured DINO latent space, which avoids redundant pixel-space appearance modeling. Complementing this representation, LDA-1B employs a mixed-frequency multi-modal diffusion transformer to handle asynchronous vision and action streams, enabling stable training at the 1B-parameter scale. Experiments in simulation and the real world show LDA-1B outperforms prior methods (e.g., π_0.5) by up to 21% , 48% , and 23% on contact-rich, dexterous, and long-horizon tasks, respectively. Notably, LDA-1B enables data-efficient fine-tuning, gaining 10% by leveraging 30% low-quality trajectories typically harmful and discarded. The code and data will be publicly released to benefit the community.

Abstract:
This paper presents the MOBIUS platform, a bipedal robot capable of walking, crawling, climbing, and rolling. MOBIUS features four limbs, two 6-DoF arms with two-finger grippers for manipulation and climbing, and two 4-DoF legs for locomotion--enabling smooth transitions across diverse terrains without reconfiguration. A hybrid control architecture combines reinforcement learning for locomotion and admittance control enhanced for safety by a Reference Governor and auto-tuning toward compliant contact interactions during manipulation. A high-level MIQCP planner autonomously selects locomotion modes to balance stability and energy efficiency. Hardware experiments demonstrate robust gait transitions, dynamic climbing, and full-body load support via pinch grasp. Overall, MOBIUS demonstrates the importance of tight integration between morphology, high-level planning, and control to enable mobile loco-manipulation and grasping, substantially expanding its interaction capabilities, workspace, and traversability.

Abstract:
Enabling humanoid robots to operate in complex, dynamic environments remains a critical challenge, fundamentally limited by the ability to navigate robustly, safely, and accurately. While reinforcement learning with velocity-commanded policies has achieved remarkable robustness in humanoid locomotion, this approach lacks explicit control of the foothold placement, leading to unsafe behavior, such as stepping onto human feet, or imprecise navigation, hindering the following manipulation task. Conversely, explicit foothold-tracking policies offer a promising alternative by directly being commanded with target foot poses. However, existing approaches are often limited by unrealistic state assumptions, compromising real-world deployment, or they are part of staged pipelines, making them tied to specific downstream tasks. In this work, we introduce a novel, lightweight framework for training general-purpose 3D foothold-tracking policies. By dynamically providing footstep support through a goal sampler, this method enables the learned policy to be agnostic to specific terrains. Our new target representation effectively mitigates challenges arising in the real world, such as noisy and inaccurate pose estimation and foot contact estimation. Designed for direct real-world transfer, our policy acts as a standalone low-level controller that can be seamlessly paired with various high-level foothold generators. We demonstrate the effectiveness of our framework through extensive experiments in simulation and in the real world. By coupling our policy with different upstream planners, we achieve natural and accurate locomotion in challenging settings, paving the way for loco-manipulation tasks in complex environments.

Abstract:
Humanoid and legged robots interact with the environment through intermittent contacts, making accurate motion estimation fundamentally dependent on reasoning about contact dynamics. However, standard sensing pipelines—whether based on onboard proprioception with Extended Kalman Filters (EKFs) or external motion capture systems—recover only kinematics, while contact forces, contact timing, and inertial parameters remain unobserved. As a result, purely kinematic reconstructions often violate rigid-body dynamics, particularly during contact-rich motions. To enable accurate motion estimation from onboard kinematics in real-world deployment, we propose PRIME (Physically-consistent Robotic Inertial and Motion Estimation), a Maximum A Posteriori formulation that refines measured kinematics and actuator commands into a dynamically consistent trajectory while jointly estimating frictional contact forces and physically consistent inertial parameters. Our approach incorporates differentiable contact dynamics with smoothed complementarity constraints and an Anitescu-style friction model, yielding a smooth optimization problem that remains stable across contact transitions. We evaluate PRIME on contact-rich locomotion with quadrupedal robots and the Unitree G1 humanoid, demonstrating improved trajectory consistency and accurate inertial parameter identification. Beyond improving model-based estimation and control with calibrated inertial parameters, PRIME produces force- and contact-annotated motion reconstructions from real robots in deployment, which can be used to provide high-quality data for downstream learning applications, including large-scale behavior modeling and robot foundation models.

Abstract:
Fast and reliable initialization is critical for monocular visual–inertial navigation systems (VINS), as it establishes the starting conditions for subsequent state estimation. Despite steady progress, most existing methods heavily rely on visual feature correspondences and require 3-4 seconds of sensory data for successful initialization, which limits their applicability and efficiency. With the advent of feed-forward 3D models that can directly predict point clouds from images, we revisit the visual–inertial initialization problem from a concise perspective. In this work, we propose a feature-free initialization framework that leverages up-to-scale point clouds predicted by a feed-forward 3D model, thereby obviating the need for visual feature tracking and estimation. This design substantially reduces system complexity and improves the reliability of initialization. Experiments on public datasets demonstrate that the proposed feature-free initialization method achieves the highest success rate, exceeding 90%, and reduces the data duration required for successful initialization by over 50% compared to state-of-the-art methods, typically to under 1.2 s. We further validate our method on a self-collected dataset covering various indoor and outdoor scenarios, demonstrating robust performance, particularly in visually degraded environments where existing methods often fail. The code and dataset will be released to facilitate further research by the community.

Abstract:
Recent advances in vision-language-action (VLA) models have motivated the extension of their capabilities to embodied settings, where reinforcement learning (RL) offers a principled way to optimize task success through interaction. However, existing methods remain fragmented, lacking both a unified platform for fair comparison across architectures and algorithms and an efficient system design for scalable training. To address these challenges, we introduce RLux-VLA, a unified and efficient framework for scalable RL training of VLA models. RLux-VLA achieves unification by providing a unified interface that standardizes the integration of diverse VLA architectures, multiple RL algorithms, and heterogeneous simulators, enabling extensibility. To ensure efficiency, the system adopts a flexible resource allocation architecture for rendering, inference, and training workloads in RL pipelines. In particular, for GPU-parallelized simulators, RLux-VLA introduces a hybrid fine-grained pipeline allocation strategy, yielding a 1.61x–1.88x training speedup. Using this unified system, models trained with RLux-VLA demonstrate consistent performance improvements of approximately 20–85% across multiple simulation benchmarks, including LIBERO, ManiSkill, and RoboTwin. Furthermore, we distill a set of training practices for effective RL-based VLA training. We position RLux-VLA as a foundational system to enable efficient, unified, and reproducible research in embodied intelligence.

Authors: Ryan Punamiya, Simar Kareer, Zeyi Liu, Joshua Citron, Ri-Zhao Qiu, Xiongyi Cai, Alexey Gavryushin, Jiaqi Chen, Davide Liconti, Lawrence Y. Zhu, Patcharapong Aphiwetsa, Baoyu Li, Aniketh Cheluva, Pranav Kuppili, Yangcen Liu, Dhruv Patel, Aidan Gao, Ryan Co, Hye-Young Chung, Renee Zbizika, Jinyun Liu, Xiaomeng Xu, Haoyu Xiong, Geng Chen, Sebastiano Oliani, Wenkai Xuan, Chenyu Yang, Xi Wang, James Fort, Richard Newcombe, Josh Gao, Jason Chong, Garrett Matsuda, Aseem Doriwala, Robert K. Katzschmann, Marc Pollefeys, Xiaolong Wang, Shuran Song, Judy Hoffman, Danfei Xu

Abstract:
Robot learning increasingly depends on large and diverse data, yet robot data collection remains expensive and difficult to scale. Egocentric human data offer a promising alternative by capturing rich manipulation behavior across everyday environments. However, existing human datasets are often limited in scope, difficult to extend, and fragmented across institutions. We introduce EgoVerse, a collaborative platform for human data–driven robot learning that unifies data collection, processing, and access under a shared framework, enabling contributions from individual researchers, academic labs, and industry partners. The current release includes 1,362 hours (80k episodes) of human demonstrations spanning 1,965 tasks, 240 scenes, and 2,087 unique demonstrators, with standardized formats, manipulation-relevant annotations, and tooling for downstream learning. Beyond the dataset, we conduct a large-scale study of human-to-robot transfer with experiments replicated across multiple labs, tasks, and robot embodiments under shared protocols. We find that policy performance generally improves with increased human data, but that effective scaling depends on alignment between human data and robot learning objectives. Together, the dataset, platform, and study establish a foundation for reproducible progress in human data–driven robot learning.

Abstract:
Despite the demand for robots in high-value clinical tasks like bathing, contemporary systems still lack the safety and reliability required for complex, sustained physical interaction with humans. A key challenge hindering the development of such systems is that collecting, understanding, and effectively transferring highly dynamic, contact-rich human bathing demonstrations is difficult, even with modern motion and tactile sensing equipment. We present a straightforward, but effective framework for doing so with high fidelity by utilizing contact regions as a key processing primitive. We use our framework to build a dataset of bathing demonstrations performed by trained clinicians on human subjects. We then use this dataset to design and control an arm-mounted dexterous soft hand to perform bathing tasks on a mannequin using open- and closed-loop strategies. Our dataset is the first to provide high quality synchronized motion, shape, contact, and force during sustained, contact-rich human-human interaction, and our transfer strategies demonstrate effective use of these data across multiple levels of the robotics stack. All relevant materials will be publicly released to enable further advancements in physical human-robot interaction (pHRI) research.

Abstract:
We propose a sampling-based framework for finite-horizon trajectory and policy optimization under differentiable dynamics by casting controller design as inference. Specifically, we minimize a KL-regularized expected trajectory cost, which yields an optimal "Boltzmann-tilted" distribution over controller parameters that concentrates on low-cost solutions as temperature decreases. To sample efficiently from this sharp, potentially multimodal target, we introduce tempered sequential Monte Carlo (TSMC): an annealing scheme that adaptively reweights and resamples particles along a tempering path from a prior to the target distribution, while using Hamiltonian Monte Carlo rejuvenation to maintain diversity and exploit exact gradients obtained by differentiating through trajectory rollouts. For policy optimization, we extend TSMC via (i) a deterministic empirical approximation of the initial-state distribution and (ii) an extended-space construction that treats rollout randomness as auxiliary variables. Experiments across trajectory- and policy-optimization benchmarks show that TSMC is broadly applicable and compares favorably to state-of-the-art baselines.

Abstract:
Our work focuses on the social reasoning capabilities of foundational models for real-world human–robot interactions. We introduce the Social Human Robot Embodied Conversation (SHREC) Dataset, a large-scale benchmark of 400 real-world human-robot interaction videos and over 10K annotations, capturing robot social errors, competencies, underlying rationales, and corrections. Unlike prior datasets focused on human–human interactions, the SHREC Dataset uniquely highlights the challenges faced by real-world embodied social AI agents, where robots lack innate social abilities such as emotion understanding, intention tracking, and conversational mechanics. Moreover, current foundational models struggle to recognize these deficits, which manifest as subtle, socially situated failures. To evaluate AI models’ capacity for social reasoning, we define eight benchmark tasks targeting critical areas such as (1) detection of social errors and competencies, (2) identification of underlying social attributes, (3) comprehension of interaction flow, and (4) providing rationale and alternative correct actions. Experiments with state-of-the-art foundational models, alongside human evaluations, reveal substantial performance gaps—underscoring the difficulty and providing directions in developing socially intelligent AI.

Abstract:
Learning reward functions from demonstrations assumes that demonstrations provide adequate supervision over all features—or task-relevant aspects of behavior. In practice, demonstrations are often imperfect: humans may under-emphasize certain features due to cognitive load or physical difficulty, or the training regime may fail to sufficiently cover all relevant situations. In either case, important features may be underspecified, leading to ambiguity in the learned reward function and misaligned behavior at deployment. We propose a framework that detects such underspecified features and actively solicits targeted corrective demonstrations. Our key insight is that demonstrations implicitly reveal which features are well specified: features that are consistently optimized show little variation across demonstrations, while features that are underspecified vary widely. We leverage this statistical signal to infer which features may have been insufficiently demonstrated. The robot then explains which features it is uncertain about in natural language and queries for demonstrations that explicitly address the identified gaps. We evaluate our approach in a simulated tabletop manipulation domain and in a user study with a real Franka Panda robot. Targeted, explanation-guided queries significantly improve reward recovery compared to random querying and passive data collection, reducing ambiguity that would otherwise persist in learning from imperfect demonstrations.

Abstract:
Powered wheelchairs provide essential mobility and independence for individuals with motor impairments, yet the skills required for driving and prevalence of safety risks, such as collisions and drop-offs, limit access for users with high levels of disability. Robotic driver assistance systems have the potential to mitigate these risks, unlocking access to powered wheelchairs for a broader population of users. This paper presents REACT, a shared control algorithm for powered wheelchairs built atop the LUCI driver assistance system, that performs minimal, deterministic control arbitration and action selection under safety constraints, and is robust to user, environment, and task variability. Additionally, we introduce a novel evaluation protocol designed to capture longitudinal, task-dependent, and user-specific effects of shared control. We implement REACT on a LUCI-equipped commercial powered wheelchair and conduct case studies using our proposed evaluation method. We find that the effects of REACT assistance vary substantially across participant impairment types, control interfaces, tasks, and timescales, and in many evaluated scenarios, lead to improved driving control input behavior. Users also consistently report positive perceptions of using the system. Our findings underscore the importance of analyzing not only task performance but also user–assistance interaction when evaluating shared control for powered wheelchairs, and additionally motivate longitudinal, multi-task assessment frameworks for assistive mobility systems.

Abstract:
Dexterous manipulation policies today largely assume fixed hand designs, severely restricting their generalization to new embodiments with varied kinematic and structural layouts. To overcome this limitation, we introduce a parameterized canonical representation that unifies a broad spectrum of dexterous hand architectures. It comprises a unified parameter space and a canonical URDF format, offering three key advantages. 1) The parameter space captures essential morphological and kinematic variations for effective conditioning in learning algorithms. 2) A structured latent manifold can be learned over our space, where interpolations between embodiments yield smooth and physically meaningful morphology transitions. 3) The canonical URDF standardizes the action space while preserving dynamic and functional properties of the original URDFs, enabling efficient and reliable cross-embodiment policy learning. We validate these advantages through extensive analysis and experiments, including grasp policy replay, variational autoencoder (VAE) latent encoding, and cross-embodiment zero-shot transfer. Specifically, we train a VAE on the unified representation to obtain a compact, semantically rich latent embedding, and develop a grasping policy conditioned on the canonical representation that generalizes across dexterous hands. We demonstrate, through simulation and real-world tasks on unseen morphologies (e.g., 81.9% zero-shot success rate on 3-finger leap hand), that our framework unifies both the representational and action spaces of structurally diverse hands, providing a scalable foundation for cross-hand learning towards universal dexterous manipulation. Anonymous project website: https://ohra-anonymous.github.io/.

Abstract:
Position tracking based on bearing measurements and Ultra-wideband (UWB) ranging is widely used in robotic navigation tasks. However, due to variations in the number of robots, anchor configurations, UWB tag layouts, and the presence or absence of anonymous visual observations, existing methods typically rely on specially designed solvers or networks tailored to a particular localization problem. In this work, we introduce AnyAmber, a generalist neural network for versatile anonymous bearing and range based position tracking. To model diverse and dynamic geometric constraints, a heterogeneous EGAT network is employed for unified localization and uncertainty estimation, cascaded with a differentiable hierarchical PGO for improved accuracy. Additionally, we incorporate an Embedded-GRU module for adaptive UWB bias correction and a temporal graph-based matching network for soft assignments between robots and anonymous bearings. By adopting a unified problem formulation, our model is jointly pretrained on a large-scale multi-task dataset encompassing diverse simulated and real-world environments. In the experiments, with only a single trajectory fine-tuning in a target test scenario, the model achieves superior few-shot localization performance than existing methods.

Abstract:
This paper studies the real-time motion planning problem for steerable-wheeled mobile robots (SWMRs). Despite significant progress in SWMR control, most existing approaches design controllers for specific task scenarios and actuator limitations, and are further restricted to four-wheel rectangular layouts, resulting in limited versatility. Motivated by these issues, we formulate SWMR motion planning as an optimization problem that simultaneously maximizes motion performance while enforcing actuator feasibility. The proposed formulation supports flexible combinations of task objectives, wheel placements, and actuator constraints without modifying the underlying framework. Formulation alone does not guarantee real-time solvability; we therefore design a tailored fast algorithm that iteratively contracts the non-convex feasible set toward stationary points. Although contraction-based iterations are uncommon, their anytime feasibility and computational efficiency make them particularly suitable for real-time robotic applications. We then rigorously establish the descent, convergence, and feasibility properties of the proposed algorithm. Simulation results on a trajectory tracking task demonstrate an order-of-magnitude reduction in computation time and great improvement in constraint satisfaction compared to baseline methods.

Abstract:
Tactile and proximity sensing is fundamental for achieving autonomous robotic manipulation and safe human-robot interaction. However, traditional dual-mode sensors often face challenges such as environmental interference and the perception gap between far-field vision and near-field contact. This study presents a versatile sensing system based on Electrical Capacitance Tomography (ECT) principles, providing a unified framework for non-contact proximity perception, pre-touch orientation estimation and material recognition. We implement two distinct sensor configurations: a large-area array (10 cm × 10\text cm) for high-dynamic safety feedback and a compact module integrated into a robotic gripper (2 cm × 9\text cm). Instead of computationally expensive tomographic reconstruction, we propose CapacitiveServo-Net, a physics-informed deep learning architecture that extracts spatial dielectric features directly from mutual capacitance perturbations. This model facilitates a unified pre-touch servoing framework by mapping high-dimensional capacitive transients to geometric primitives (distance and orientation) and material properties. Experimental results on a 7-DOF manipulator demonstrate that our system achieves high-precision, non-contact proximity tracking and real-time pose alignment. Furthermore, the system demonstrates concurrent material classification and pre-touch adaptive grasp refinement during the approach phase, offering a robust, unified solution for proactive perception and manipulation in occluded or degraded visual environments.

Abstract:
Robotic grippers face substantial challenges in grasping and manipulating thin objects. Most existing grippers rely on highly precise approach and grasp motions, which limits robustness and reduces applicability. This paper explores thin-object grasping using books as a representative example. Here, we propose a novel solution that integrates an active surface with underactuated compliance to achieve stable grasping of thin objects without complex control. First, an underactuated gripper with an active surface is designed. The active-surface thumb performs in-hand repositioning of the target book without requiring adjustments of the robot arm or the other fingers, while the underactuated fingers establish compliant contact conditions with the environment, and the reconfigurable structure enables reliable grasping of books under different configurations. Second, we establish a kinematic model of the gripper, and determine the initial grasp postures for two representative scenarios (books lying flat on a tabletop and books vertically packed in a shelf). Third, by analyzing the physical model of a book lying on a table and its interaction with the gripper and the environment, we systematically optimize the structural parameters and grasping strategy. Finally, extensive experiments validate the effectiveness of the proposed gripper and strategy. The results demonstrate strong robustness and adaptability when grasping thin objects placed flat (including books, paper, fabric, and mouse pad), as well as a high success rate when grasping vertically packed books. Moreover, the proposed gripper can reliably completes long sequential “grasp-place” tasks.

Abstract:
Achieving safe manipulation-oriented navigation for humanoid robots is fundamentally challenged by two factors: locomotion-induced perceptual distortion (causing semantic-geometry distortion) and changes within the environment (causing map-reality mismatches). Existing static scene graphs often fail under these conditions, leading to interaction failures. To address this, we introduce the Multi-modal Interaction Field (MIF), a hierarchical framework that transforms the robot from a passive map-user into an active knowledge-evolver. MIF constructs three synergistic fields: (i) a denoised Appearance Field utilizing confidence-gated 3D Gaussian Splatting to suppress gait oscillation noise; (ii) a hierarchical Spatial Field for semantic reasoning; and (iii) Geometry Field, leveraging a Flow Matching based generative model to reconstruct high-fidelity meshes for rigorous Interaction Pose Safety (IPS) verification against the target object. Crucially, we propose a closed-loop Interaction and Adaptation Mechanism to adapt to environmental changes. By monitoring a multi-modal discrepancy score \mathcalD, the system autonomously distinguishes between sensor noise and genuine environmental changes (e.g., relocated objects), triggering a local evolution loop to rectify obsolete memory. Real-world experiments on a Unitree-G1 humanoid demonstrate that MIF significantly outperforms static baselines (HOV-SG), improving the success rate in dynamic relocation scenarios from 12% to 94%, while reducing semantic memory footprint by 91.4% via feature distillation.

Abstract:
We introduce asRoBallet, to the best of our knowledge, the first successful deployment of reinforcement learning (RL) on a humanoid ballbot hardware. Historically, ballbots have served as a canonical benchmark for underactuated and nonholonomic control, which are characterized by a reality gap in complex friction models for wheel-sphere-ground interactions. While current literature demonstrates successful handling of 3D balancing with LQR and MPC, transitioning to actual hardware for a humanoid ballbot using RL is currently hindered by critical gaps in contact modeling, actuator latency & jitter, and safe hardware exploration. This study proposes a high-fidelity MuJoCo simulation that explicitly models the discrete roller mechanics of ETH-type omni-wheels, thereby capturing parasitic vibrations and contact discontinuities that are previously ignored. We also developed a Friction-Aware Reinforcement Learning framework that achieves zero-shot Sim2Real transfer by mastering the coupled rolling, lateral, and torsional friction channels at the wheel-sphere and sphere-ground interfaces. We designed asRoBallet through subtractive reconfiguration, repurposing key components from an overconstrained quadruped and integrating them into a newly designed structural frame to achieve a robust research platform at low cost. We also developed a generalized iOS ecosystem that transforms consumer electronics into a low-latency interface, enabling a single operator to orchestrate expressive humanoid maneuvers via intuitive natural motion.

Abstract:
Robot guide dogs offer navigation assistance that will greatly expand the independent mobility of the visually impaired, but their effective use requires subtle human-robot coordination that is difficult for users to learn from generic verbal instructions. To tackle the challenge, we present CANINE, an automated coaching system that trains users for interactive navigation with a robot guide dog, through personalized, adaptive verbal feedback. CANINE decomposes a complex coordination task into sub-skills and operates at two levels. At the high level, it decides what to train by tracking the learner’s proficiency across sub-skills using knowledge tracing and prioritizing training in the weakest areas. At the low level, CANINE decides how to train each sub-skill by observing each human practice episode, using foundation models to infer the underlying causes of errors, and generating targeted verbal corrections adaptively. A controlled study with blindfolded participants demonstrates that CANINE significantly improves both learning efficiency and final performance, compared to generic verbal instructions.We further validate CANINE through (i) a retention study showing lasting skill improvement after two weeks and (ii) a case study with a visually impaired user. Both studies align with the controlled study’s findings, while revealing additional design considerations for real-world deployment.

Abstract:
Visuomotor manipulation policies trained via large-scale behavior cloning have achieved strong semantic scene understanding, yet often fail to reliably execute correct low-level actions under distribution shifts. For example, even in a simple pick-up task with identical scene layouts, camera viewpoints, and illumination, performance can degrade substantially when the object is placed at unseen locations. We argue that this gap arises from insufficient action understanding, namely the inability to interpret the robot’s base-frame action coordinate system in image space. To address this issue, we introduce AxisGuide, a lightweight guidance method that bridges semantic scene understanding and action-coordinate interpretation. Using camera parameters and end-effector poses, AxisGuide renders the robot base-frame axes in each camera view and augments RGB observations with a small set of cue channels that explicitly visualize the meaning of \(+x/+y/+z\) motions in image space. Extensive evaluations in both the LIBERO simulation and real-world environments demonstrate that AxisGuide yields substantial performance gains and improved generalization, highlighting the effectiveness of explicit action-coordinate cues for learning reliable and transferable generalist visuomotor policies.

Abstract:
Contact-rich manipulation demands human-like integration of perception and force feedback: vision should guide task progress, while high-frequency interaction control must stabilize contact under uncertainty. Existing learning-based policies often entangle these roles in a monolithic network, trading off global generalization against stable local refinement, while control-centric approaches typically assume a known task structure or learn only controller parameters rather than the structure itself. In this paper, we formalize a physically grounded interaction frame, an instantaneous local basis that decouples force regulation from motion execution, and propose a method to recover it from demonstrations. Based on this, we address both issues by proposing Force Policy, a global-local vision-force policy in which a global policy guides free-space actions using vision, and upon contact, a high-frequency local policy with force feedback estimates the interaction frame and executes hybrid force-position control for stable interaction. Real-world experiments across diverse contact-rich tasks show consistent gains over strong baselines, with more robust contact establishment, more accurate force regulation, and reliable generalization to novel objects with varied geometries and physical properties, ultimately improving both contact stability and execution quality.

Abstract:
Neural network (NN) dynamics models and control policies achieve strong performance in robotics, but providing sound guarantees under uncertainty is difficult, especially when the NNs are components within the closed-loop system. Existing reachability tools offer formal over-approximations, yet are often non-differentiable, overly conservative, and too slow to integrate into modern learning and real-time planning pipelines. To address this, we present a parallelizable, differentiable reachability analysis tool in JAX that unifies continuous- and discrete-time systems and supports both analytical and NN-based dynamics and controllers. Our reachability tool uses Taylor-model flowpipe construction and CROWN-style linear bound relaxation and propagation, yielding a GPU-batched reachability primitive that can be differentiated and used in downstream objectives. Building on this primitive, we design (i) a certified training method that encourages the learning of reachability-friendly dynamics models and controllers, and (ii) a reachability-informed sampling-based MPC scheme that incorporates certified reachable sets during action selection and enables gradient-based refinement. Experiments on non-prehensile object manipulation and quadrotor control tasks show competitive performance to baseline planners while providing tight, certified reachability guarantees under uncertainty.

Abstract:
Large language models (LLMs) have recently emerged as a promising tool for automating robot design from high-level specifications, yet they remain ineffective for robots operating under complex physical interactions. This limitation stems from the gap between language-based reasoning and the physical consequences of embodiment, often resulting in designs with low physical validity. In this work, we propose a multi-layered framework, AID-SR, that establishes a closed loop by translating simulator-observed physical states into structured feedback for the LLM designer. Combined with semantic critique, human feedback, and iterative refinement, the framework promotes the generation of physically feasible and functionally meaningful robot designs. We evaluate our approach on tendon-driven continuum robots across a benchmark of 14 tasks spanning reaching, grasping, locomotion, and manipulation. The proposed framework achieve 96.2% rate for passing the simulation feasibility check and by apply a common reinforcement learning training, 26.7% robots can successfully fulfill the corresponding task. We then fabricate three designed robots of AID-SR that successfully complete the task in real-world. These extensive experiments across simulation and real-world environments demonstrate and break the wall of utilizing the LLMs for automated design of continuum robots.

Abstract:
Online policy learning directly in the physical world is a promising yet challenging direction for embodied intelligence. Unlike simulation, real-world systems cannot be arbitrarily accelerated, cheaply reset, or massively replicated, which makes scalable data collection, heterogeneous deployment, and long-horizon effective training difficult. These challenges suggest that real-world policy learning is not only an algorithmic issue but fundamentally a systems problem. We present USER, a U nified and extensible S yst E m for R eal-world online policy learning. USER treats physical robots as first-class hardware resources alongside GPUs through a unified hardware abstraction layer, enabling automatic discovery, management, and scheduling of heterogeneous robots. To address cloud–edge communication, USER introduces an adaptive communication plane with tunneling-based networking, distributed data channels for traffic localization, and streaming-multiprocessor-aware weight synchronization to regulate GPU-side overhead. On top of this infrastructure, USER organizes learning as a fully asynchronous framework with a persistent, cache-aware buffer, enabling efficient long-horizon experiments with robust crash recovery and reuse of historical data. In addition, USER provides extensible abstractions for rewards, algorithms, and policies, supporting online imitation or reinforcement learning of CNN/MLP, generative policies, and large vision–language–action (VLA) models within a unified pipeline. Results in both simulation and the real world show that USER enables multi-robot coordination, heterogeneous manipulators, edge–cloud collaboration with large models, and long-running asynchronous training, offering a unified and extensible systems foundation for real-world online policy learning.

Abstract:
Robust and accurate perception of humans in their 3D scene context is essential for integrating robots into everyday environments. Existing approaches, however, often fail to predict plausible and accurate human motion estimates that are consistent with the surrounding scene, especially in the presence of heavy occlusions or partial visibility. This can limit both safety and efficiency for robotic operations. We introduce HumanFlow, a latent diffusion model that unifies human motion tracking and forecasting, conditioned on the 3D scene context. We show that our human motion model produces smooth and accurate predictions under challenging conditions, including heavy occlusions, and outperforms state-of-the-art methods in tracking accuracy while being significantly more efficient. Furthermore, we show how HumanFlow's latent space can be tightly coupled with control by conditioning a flow-matching-based, approximate MPC policy on these representations. We validate our policy in simulation with real human trajectories for MAV social navigation, demonstrating superior navigation performance and remaining collision-free, even under partial observability of the human. The code will be made publicly available upon acceptance.

Abstract:
Robots that physically interact with humans must decide not only how and when to help, but also when not to help. In physical caregiving and collaborative manipulation, robots can over-assist by misestimating user capability or defaulting to helping when users can act independently. Physical functionality is highly individual, only partially observable, difficult to specify a priori, and assistance policies are often not grounded in user-specific ability, making calibrated intervention challenging. We address this by actively inferring human joint-space reachability from sparse interaction. Our framework represents reachability using a compositional parametric model where a box constraint is deformed by local Gaussian primitives. We learn a latent space that decodes to these parameters and structure it using biomechanical anchors from musculoskeletal simulation and clinical anchors from retrieval-augmented reasoning over rehabilitation literature. The robot maintains a belief over this manifold and actively selects calibration queries to infer user-specific functionality. We evaluate through computational experiments and real-robot studies with participants wearing resistance bands. Our method achieves ≈0.50 IoU within 20 queries. In sandwich making, reachability-aware assistance significantly improves user perception of physical engagement (χ^2(3) = 18.29, p < .001) without increasing workload. In Action Research Arm Test-inspired manipulation, we demonstrate online adaptation capacity.

Abstract:
Detecting and segmenting novel object instances in open-world environments is a fundamental problem in robotic perception. Given only a small set of template images, a robot must locate and segment a specific object instance in a cluttered, previously unseen scene. Existing proposal-based approaches are highly sensitive to proposal quality and often fail under occlusion and background clutter. We propose L2G-Det, a local-to-global instance detection framework that bypasses explicit object proposals by leveraging dense patch-level matching between templates and the query image. Locally matched patches generate candidate points, which are refined through a candidate selection module to suppress false positives. The filtered points are then used to prompt an augmented Segment Anything Model (SAM) with instance-specific object tokens, enabling reliable reconstruction of complete instance masks. Experiments demonstrate improved performance over proposal-based methods in challenging open-world settings.

Abstract:
We study nonlinear output-feedback control from high-resolution RGB images and provide robust constraint satisfaction guarantees despite partial observability, sensor noise, and nonlinear dynamics. To enable scalability while retaining guarantees, we propose: (i) a learned low-dimensional observation map from pretrained visual features with state-dependent error bounds, and (ii) a causal affine time-varying output-feedback policy optimized via System Level Synthesis (SLS). We efficiently solve the resulting nonconvex program via sequential convex programming. On two simulated visuomotor tasks (a 4D car and a 10D quadrotor) with \ge 512 × 512 pixels and a humanoid task with partial observability, our method enables safe, information-gathering behavior that reduces uncertainty while maintaining zero observed constraint violations across trials. We also validate our method on hardware, safely controlling a ground vehicle from onboard images. Together, these results show that learned visual abstractions coupled with SLS make certified visuomotor output-feedback practical at scale.

Abstract:
Recent advances in robotic manipulation remain hindered by the inevitability of task failures, particularly in dynamic and unstructured environments. To handle such failure, existing frameworks typically follow a stepwise detect–reason–recover pipeline, which often incurs high latency and limited robustness due to delayed reasoning and reactive planning. Inspired by the human capability to anticipate and proactively plan for potential failures, we introduce AgentChord, an agentic system that models a manipulation task as a directed, recovery-augmented graph. Prior to execution, this graph is enriched with anticipatory recovery branches that specify context-aware corrective behaviors, enabling immediate and targeted responses when failures occur. During execution, AgentChord operates through a choreography of specialized agents, covering a composer for task structuring, an arranger for execution compilation, and a conductor for recovery orchestration. AgentChord coordinates these agents via low-latency monitors that detect deviations and trigger pre-compiled recoveries without re-planning. Empirical studies on diverse long-horizon bimanual manipulation tasks demonstrate that AgentChord substantially improves success rates and execution efficiency, advancing the reliability and autonomy of real-world robotic systems.

Abstract:
We present GPU-SLS, a GPU-parallelized framework for provably safe, robust nonlinear model predictive control (MPC) that scales to high-dimensional uncertain robotic systems and long planning horizons. Our method jointly optimizes an inequality-constrained, dynamically-feasible nominal trajectory, a tracking controller, and a closed-loop reachable set under disturbance, all in real time. To efficiently compute nominal trajectories, we develop a sequential quadratic programming procedure with a novel GPU-accelerated quadratic program (QP) solver that uses parallel associative scans and adaptive caching within an alternating direction method of multipliers (ADMM) framework. The same GPU QP backend is used to optimize robust tracking controllers and closed-loop reachable sets via system level synthesis (SLS), enabling reachability-constrained control in both fixed- and receding-horizon settings. We achieve substantial performance gains, reducing nominal trajectory solve times by 97.7% relative to state-of-the-art CPU solvers and 71.8% compared to GPU solvers, while accelerating SLS-based control and reachability by 237×. Despite large problem scales, our method achieves 100% empirical safety, unlike high-dimensional learning-based reachability baselines. We validate our approach on complex nonlinear systems, including whole-body quadrupeds (61D) and humanoids (75D), synthesizing robust control policies online on the GPU in 34 milliseconds on average and scaling to problems with 2 × 10^5 decision variables and 8× 10^4 constraints.

Abstract:
Large behavior models (LBMs) have shown strong dexterous manipulation capabilities by extending imitation learning to large-scale training on extensive multi-task robot data, yet their generalization remains limited by the insufficient coverage of available robot data. To expand this coverage without costly additional data collection, recent work increasingly relies on co-training: jointly learning from target robot data and heterogeneous data modalities. However, how different co-training data modalities and training strategies affect policy performance remains poorly understood. We present a large-scale empirical study examining five co-training data modalities---standard vision-language data, dense language annotations for robot trajectories, cross-embodiment robot data, human videos, and discrete robot action tokens—across single- and multi-phase training strategies. Our study leverages 4,000 hours of robot and human manipulation data and 50M vision–language samples to train vision-language-action (VLA) policies. We evaluate 89 policies over 58,000 simulation rollouts and 2,835 real-world rollouts. Our results show that co-training with various forms of vision-language and cross-embodiment robot data substantially improves generalization to distribution shifts, unseen tasks, and language following, while discrete action token variants yield no statistically significant benefits. Furthermore, combining effective modalities produces cumulative gains and enables rapid adaptation to unseen long-horizon dexterous tasks via fine-tuning. Together, these results provide a systematic understanding of co-training and practical guidance for building scalable generalist robot policies.

Abstract:
Behavior Cloning (BC) has emerged as a highly effective paradigm for robot learning. However, BC lacks a self-guided mechanism for online improvement after demonstrations have been collected. Existing offline-to-online learning methods often cause policies to replace previously learned good actions due to a distribution mismatch between offline data and online learning. In this work, we propose Q2RL, Q-Estimation and Q-Gating from BC for Reinforcement Learning, an algorithm for efficient offline-to-online learning. Our method consists of two parts: (1) Q-Estimation extracts a Q function from a BC policy using a few interaction steps with the environment, followed by online RL with (2) Q-Gating, which switches between BC and RL policy actions based on their respective Q values to collect samples for RL policy training. Across manipulation tasks from D4RL and robomimic benchmarks, Q2RL outperforms SOTA offline-to-online learning baselines on success rate and time to convergence. Q2RL is efficient enough to be applied in an on-robot RL setting, learning robust policies for contact-rich and high precision manipulation tasks such as pipe assembly and kitting, in 1-2 hours of online interaction, achieving success rates of up to 100% and up to 3.75x improvement against the original BC policy.