CoRL 2025 论文列表

Training manipulation policies for humanoid robots with diverse data enhances their robustness and generalization across tasks and platforms. However, learning solely from robot demonstrations is labor-intensive, requiring expensive tele-operated data collection,n which is difficult to scale. This paper investigates a more scalable data source, egocentric human demonstrations, to serve as cross-embodiment training data for robot learning. We mitigate the embodiment gap between humanoids and humans from both the data and modeling perspectives. We collect an egocentric task-oriented dataset that is directly aligned with humanoid manipulation demonstrations. We then train a human-humanoid behavior policy, which we term Human Action Transformer (HAT). The state-action space of HAT is unified for both humans and humanoid robots and can be differentiably retargeted to robot actions. Co-trained with smaller-scale robot data, HAT directly models humanoid robots and humans as different embodiments without additional supervision. We show that human data improves both the generalization and robustness of HAT with significantly better data collection efficiency.

In robot imitation learning, policy performance is tightly coupled with the quality and composition of the demonstration data. Yet, developing a precise understanding of how individual demonstrations contribute to downstream outcomes—such as closed-loop task success or failure—remains a persistent challenge. Inspired by the theory of influence functions, we propose CUPID. Given a set of evaluation rollouts, CUPID estimates the influence of a training demonstration on the policy’s expected return. This enables ranking and selection of demonstrations according to their impact on the policy’s closed-loop performance. We use our estimator to curate data by 1) filtering out training demonstrations that harmed the policy’s performance and 2) subselecting newly collected trajectories that will most help improve the policy. Extensive simulated and hardware experiments show that our approach consistently identifies which data drives test-time performance. For example, training with less than 33% of curated data can result in state-of-the-art diffusion policies on the simulated Robomimic benchmark, and we observe similar improvements in hardware experiments. Furthermore, our hardware experiments show that our influence-based estimator can identify robust strategies under distribution shift, isolate spurious correlations, and even enhance post-training of generalist policies.

Imitation learning has shown great promise in robotic manipulation, but the policy’s execution is often unsatisfactorily slow due to commonly tardy demonstrations collected by human operators. In this work, we present DemoSpeedup, a self-supervised method to accelerate visuomotor policy execution via entropy-guided demonstration acceleration. DemoSpeedup starts from training an arbitrary generative policy (e.g., ACT or Diffusion Policy) on normal-speed demonstrations, which serves as a per-frame action entropy estimator. The key insight is that frames with lower action entropy estimates call for more consistent policy behaviors, which often indicate the demands for higher-precision operations. In contrast, frames with higher entropy estimates correspond to more casual sections, and therefore can be more safely accelerated. Thus, we segment the original demonstrations according to the estimated entropy, and accelerate them by down-sampling at rates that increase with the entropy values. Trained with the speedup demonstrations, the resulting policies execute up to 3 times faster while maintaining the task completion performance. Interestingly, these policies could even achieve higher success rates than those trained with normal-speed demonstrations, due to the benefits of reduced decision-making horizons.

Mobile manipulation is the fundamental challenge for robotics in assisting humans with diverse tasks and environments in everyday life. Conventional mobile manipulation approaches often struggle to generalize across different tasks and environments due to the lack of large-scale training. However, recent advances in manipulation foundation models demonstrate impressive generalization capability on a wide range of fixed-base manipulation tasks, which are still limited to a fixed setting. Therefore, we devise a plug-in module named MoTo, which can be combined with any off-the-shelf manipulation foundation model to empower them with mobile manipulation ability. Specifically, we propose an interaction-aware navigation policy to generate agent docking points for generalized mobile manipulation. To enable zero-shot ability, we propose an interaction keypoints framework via vision-language models (VLM) under multi-view consistency for both target object and robotic arm following instructions, where fixed-base manipulation foundation models can be employed. We further propose motion planning objectives for the mobile base and robot arm, which minimize the distance between the two keypoints and maintain the physical feasibility of trajectories. In this way, MoTo guides the agent to move to the docking points where fixed-base manipulation can be successfully performed, and leverages VLM generation and trajectory optimization to achieve mobile manipulation in a zero-shot manner, without any requirement on mobile manipulation expert data. Extensive experimental results on OVMM and real-world demonstrate that MoTo achieves success rates of 2.68% and 16.67% higher than the state-of-the-art mobile manipulation methods, respectively, without requiring additional training data.

Learning from demonstration has proven effective in robotics for acquiring natural behaviors, such as stylistic motions and lifelike agility, particularly when explicitly defining style-oriented reward functions is challenging. Synthesizing stylistic motions for real-world tasks usually requires balancing task performance and imitation quality. Existing methods generally depend on expert demonstrations closely aligned with task objectives. However, practical demonstrations are often incomplete or unrealistic, causing current methods to boost style at the expense of task performance. To address this issue, we propose formulating the problem as a constrained Markov Decision Process (CMDP). Our approach integrates a style-imitation objective with constraints to maintain near-optimal task performance. We introduce an adaptively adjustable Lagrangian multiplier to guide the agent to imitate demonstrations selectively, capturing stylistic nuances without compromising task performance. We validate our approach across multiple robotic platforms and tasks, demonstrating both robust task performance and high-fidelity style learning. On ANYmal-D hardware we show a 14.5\% drop in mechanical energy and a more agile gait pattern, showcasing real-world benefits.

The ability to flexibly leverage limbs for loco-manipulation is essential for enabling autonomous robots to operate in unstructured environments. Yet, prior work on loco-manipulation is often constrained to specific tasks or predetermined limb configurations. In this work, we present einforcement Learning for Interlimb Coordination (ReLIC), an approach that enables versatile loco-manipulation through flexible interlimb coordination. The key to our approach is an adaptive controller that seamlessly bridges the execution of manipulation motions and the generation of stable gaits based on task demands. Through the interplay between two controller modules, ReLIC dynamically assigns each limb for manipulation or locomotion and robustly coordinates them to achieve the task success. Using efficient reinforcement learning in simulation, ReLIC learns to perform stable gaits in accordance with the manipulation goals in the real world. To solve diverse and complex tasks, we further propose to interface the learned controller with different types of task specifications, including target trajectories, contact points, and natural language instructions. Evaluated on 12 real-world tasks that require diverse and complex coordination patterns, ReLIC demonstrates its versatility and robustness by achieving a success rate of 78.9% on average.

Building capable household and industrial robots requires mastering the control of versatile, high-degree-of-freedom (DoF) systems such as mobile manipulators. While reinforcement learning (RL) holds promise for autonomously acquiring robot control policies, scaling it to high-DoF embodiments remains challenging. Direct RL in the real world demands both safe exploration and high sample efficiency, which are difficult to achieve in practice. Sim-to-real RL, on the other hand, is often brittle due to the reality gap. This paper introduces SLAC, a method that renders real-world RL feasible for complex embodiments by leveraging a low-fidelity simulator to pretrain a task-agnostic latent action space. SLAC trains this latent action space via a customized unsupervised skill discovery method designed to promote temporal abstraction, disentanglement, and safety, thereby facilitating efficient downstream learning. Once a latent action space is learned, SLAC uses it as the action interface for a novel off-policy RL algorithm to autonomously learn downstream tasks through real-world interactions. We evaluate SLAC against existing methods on a suite of bimanual mobile manipulation tasks, where it achieves state-of-the-art performance. Notably, SLAC learns contact-rich whole-body tasks in under an hour of real-world interactions, without relying on any demonstrations or hand-crafted behavior priors.

We present GraspMolmo, a generalizable open-vocabulary task-oriented grasping (TOG) model. GraspMolmo predicts semantically appropriate, stable grasps conditioned on a natural language instruction and a single RGB-D frame. For instance, given "pour me some tea", GraspMolmo selects a grasp on a teapot handle rather than its body. Unlike prior TOG methods, which are limited by small datasets, simplistic language, and uncluttered scenes, GraspMolmo learns from a large-scale synthetic dataset of 379k samples featuring cluttered environments and diverse, realistic task descriptions. We fine-tune the Molmo visual-language model on this data, enabling GraspMolmo to generalize to novel open-vocabulary instructions and objects. In challenging real-world evaluations, GraspMolmo achieves state-of-the-art results, with a 70% prediction success on complex tasks, compared to the 35% achieved by the next best alternative. GraspMolmo also successfully demonstrates the ability to predict semantically correct bimanual grasps zero-shot. We release our synthetic dataset, code, model, and benchmarks to accelerate research in task-semantic robotic manipulation.

In many dynamic robotic tasks, such as striking pucks into a goal outside the reachable workspace, the robot must first identify the relevant physical properties of the object for successful task execution, as it is unable to recover from failure or retry without human intervention. To address this challenge, we propose a task-informed exploration approach, based on reinforcement learning, that trains an exploration policy using rewards automatically generated from the sensitivity of a privileged task policy to errors in estimated properties. We also introduce an uncertainty-based mechanism to determine when to transition from exploration to task execution, ensuring sufficient property estimation accuracy with minimal exploration time. Our method achieves a 90% success rate on the striking task with an average exploration time under 1.2 seconds—significantly outperforming baselines that achieve at most 40% success or require inefficient querying and retraining in a simulator at test time. Additionally, we demonstrate that our task-informed rewards capture the relative importance of physical properties in both the striking task and the classical CartPole example. Finally, we validate our approach by demonstrating its ability to identify object properties and adjust task execution in a physical setup using the KUKA iiwa robot arm.

The ability to robustly grasp a variety of objects is essential for dexterous robots. In this paper, we present a framework for zero-shot dynamic dexterous grasping using single-view visual inputs, designed to be resilient to various disturbances. Our approach utilizes a hand-centric object shape representation based on dynamic distance vectors between finger joints and object surfaces. This representation captures the local shape around potential contact regions rather than focusing on detailed global object geometry, thereby enhancing generalization to shape variations and uncertainties. To address perception limitations, we integrate a privileged teacher policy with a mixed curriculum learning approach, allowing the student policy to effectively distill grasping capabilities and explore for adaptation to disturbances. Trained in simulation, our method achieves success rates of 97.0% across 247,786 simulated objects and 94.6% across 512 real objects, demonstrating remarkable generalization. Quantitative and qualitative results validate the robustness of our policy against various disturbances.

Due to the powerful vision-language reasoning and generalization abilities, multimodal large language models (MLLMs) have garnered significant attention in the field of end-to-end (E2E) autonomous driving. However, their application to closed-loop systems remains underexplored, and current MLLM-based methods have not shown clear superiority to mainstream E2E imitation learning approaches. In this work, we propose ReasonPlan, a novel MLLM fine-tuning framework designed for closed-loop driving through holistic reasoning with a self-supervised Next Scene Prediction task and supervised Decision Chain-of-Thought process. This dual mechanism encourages the model to align visual representations with actionable driving context, while promoting interpretable and causally grounded decision making. We curate a planning-oriented decision reasoning dataset, namely PDR, comprising 210k diverse and high-quality samples. Our method outperforms the mainstream E2E imitation learning method by a large margin of 19% L2 and 16.1 driving score on Bench2Drive benchmark. Furthermore, ReasonPlan demonstrates strong zero-shot generalization on unseen DOS benchmark, highlighting its adaptability in handling zero-shot corner cases.

Real-world manipulation datasets for robotic arms remain scarce due to the high costs, rigid hardware dependencies, and complex setup procedures associated with existing data collection methods. We introduce, a redesigned Universal Manipulation Interface (UMI) that addresses these challenges, enabling low-cost, scalable, and rapid deployment across heterogeneous platforms. FastUMI achieves this through: (i) hardware decoupling via extensive mechanical reengineering, which removes dependence on specialized robotic components while preserving a consistent visual perspective; (ii) replacement of complex visual–inertial odometry with a commercial off-the-shelf tracker, simplifying the software stack without compromising pose estimation accuracy; and (iii) the provision of an integrated ecosystem that streamlines data acquisition, automates quality control, and ensures compatibility with both standard and enhanced imitation-learning pipelines. To facilitate further research, we release an open-access dataset comprising over 15,000 real-world demonstrations spanning 24 tasks constituting one of the most extensive UMI-like resources to date. Empirical evaluations show that FastUMI supports rapid deployment, reduces operational overhead, and delivers robust performance across diverse manipulation scenarios, advancing scalable data-driven robotic learning.

Enabling robots to perform diverse tasks across varied environments is a central challenge in robot learning. While vision-language-action (VLA) models have shown promise for generalizable robot skills, realizing their full potential requires addressing limitations in action representation and efficient training. Current VLA models often focus on scaling the vision-language model (VLM) component, while the action space representation remains a critical bottleneck. This paper introduces DexVLA, a novel framework designed to enhance the efficiency and generalization capabilities of VLAs for complex, long-horizon tasks across diverse robot embodiments. DexVLA features a novel diffusion-based action expert, scaled to one billion parameters, designed for cross-embodiment learning. A novel embodiment curriculum learning strategy facilitates efficient training: (1) pre-training the diffusion expert on cross-embodiment data, (2) aligning the VLA model to specific embodiments, and (3) post-training for rapid adaptation to new tasks. We conduct comprehensive experiments across multiple embodiments, including single-arm, bimanual, and dexterous hand, demonstrating DexVLA's adaptability to challenging tasks without task-specific adaptation, its ability to learn dexterous skills on novel embodiments with limited data, and its capacity to complete complex, long-horizon tasks using only direct language prompting, such as laundry folding. In all settings, our method demonstrates superior performance compared to state-of-the-art models like OpenVLA and $\pi_{0}$.

The computational burden of model predictive control (MPC) limits its application on real-time systems, such as robots, and often requires the use of short prediction horizons. This not only affects the control performance, but also increases the difficulty of designing MPC cost functions that reflect the desired long-term objective. This paper proposes ZipMPC, a method that imitates a long-horizon MPC behaviour by learning a compressed and context-dependent cost function for a short-horizon MPC. It improves performance over alternative methods, such as approximate explicit MPC and automatic cost parameter tuning, in particular in terms of i) optimizing the long-term objective; ii) maintaining computational costs comparable to a short-horizon MPC; iii) ensuring constraint satisfaction; and iv) generalizing control behaviour to environments not observed during training. For this purpose, ZipMPC leverages the concept of differentiable MPC with neural networks to propagate gradients of the imitation loss through the MPC optimization. We validate our proposed method in simulation and real-world experiments on autonomous racing. ZipMPC consistently completes laps faster than selected baselines, achieving lap times close to the long-horizon MPC baseline. In challenging scenarios where the short-horizon MPC baseline fails to complete a lap, ZipMPC is able to do so. In particular, these performance gains are also observed on tracks unseen during training.

Partially Observable Markov Decision Processes (POMDPs) model decision making under uncertainty. While there are many approaches to approximately solving POMDPs, we aim to address the problem of learning such models. In particular, we are interested in a subclass of POMDPs wherein the components of the model, including the observation function, reward function, transition function, and initial state distribution function, can be modeled as low-complexity probabilistic graphical models in the form of a short probabilistic program. Our strategy to learn these programs uses an LLM as a prior, generating candidate probabilistic programs that are then tested against the empirical distribution and adjusted through feedback. We experiment on a number of classical toy POMDP problems, simulated MiniGrid domains, and two real mobile-base robotics search domains involving partial observability. Our results show that using an LLM to guide in the construction of a low-complexity POMDP model can be more effective than tabular POMDP learning, behavior cloning, or direct LLM planning.

A robot navigating an outdoor environment with no prior knowledge of the space must rely on its local sensing, which is in the form of a local metric map or local policy with some fixed horizon. A limited planning horizon can often result in myopic decisions leading the robot off course or worse, into very difficult terrain. In this work, we make a key observation that long range navigation only necessitates identifying good frontier directions for planning instead of full map knowledge. To address this, we introduce Long Range Navigator (LRN), which learns to predict ‘affordable’ frontier directions from high-dimensional camera images. LRN is trained entirely on unlabeled egocentric videos, making it scalable and adaptable. In off-road tests on Spot and a large vehicle, LRN reduces human interventions and improves decision speed when integrated into existing navigation stacks.

Vision-Language-Action (VLA) models, pre-trained on large-scale imitation learning datasets, have demonstrated remarkable capabilities in visuomotor control. However, these models exhibit diverse failure modes in unstructured real-world environments, limiting the widespread adoption of VLAs in robotics. Efforts to enhance the robustness and generalization of VLAs have gradually shifted from the pre-training to the post-training phase. Yet, the potential of scaling test-time compute remains underexplored. In this paper, we investigate test-time scaling for robotics through the lens of sampling and verification. We first demonstrate that the relationship between action error and the number of generated samples follows an exponentiated power law across a range of VLAs, indicating the existence of inference-time scaling laws. Building on this insight, we propose a synthetic data generation pipeline for training a Vision-Language Model (VLM)-based action verifier, and show that scaling the synthetic dataset consistently improves verification and downstream accuracy. We then introduce RoboMonkey, a test-time scaling framework for VLAs. At deployment, RoboMonkey samples a small set of actions from a VLA, applies Gaussian perturbations and majority voting to construct an action proposal distribution, and then uses the VLM-based verifier to select the optimal action. Through extensive evaluations across simulated and real-world environments, we show that pairing existing VLAs with RoboMonkey yields significant performance gains, achieving a 25\% absolute improvement on out-of-distribution tasks and 8\% higher average success rate on in-distribution tasks. Additionally, when adapting to new robot setups, we show that fine-tuning both VLAs and action verifiers yields a 7\% performance increase compared to fine-tuning VLAs alone.

Bimanual manipulation has been widely applied in household services and manufacturing, which enables the complex task completion with coordination requirements. Recent diffusion-based policy learning approaches have achieved promising performance in modeling action distributions for bimanual manipulation. However, they ignored the physical safety constraints of bimanual manipulation, which leads to the dangerous behaviors with damage to robots and objects. To this end, we propose a test-time trajectory optimization framework named SafeBimanual for any pre-trained diffusion-based bimanual manipulation policies, which imposes the safety constraints on bimanual actions to avoid dangerous robot behaviors with improved success rate. Specifically, we design diverse cost functions for safety constraints in different dual-arm cooperation patterns including avoidance of tearing objects and collision between arms and objects, which optimizes the manipulator trajectories with guided sampling of diffusion denoising process. Moreover, we employ a vision-language model (VLM) to schedule the cost functions by specifying keypoints and corresponding pairwise relationship, so that the optimal safety constraint is dynamically generated in the entire bimanual manipulation process. SafeBimanual demonstrates superiority on 8 simulated tasks in RoboTwin with a 11.1\% increase in success rate and a 18.9\% reduction in unsafe interactions over state-of-the-art diffusion-based methods. Extensive experiments on 4 real-world tasks further verify its practical value by improving the success rate by 32.5\%.

Underactuated soft robot hands offer inherent safety and adaptability advantages over rigid systems, but developing dexterous manipulation skills remains challenging. While imitation learning shows promise for complex manipulation tasks, traditional approaches struggle with soft systems due to demonstration collection challenges and ineffective state representations. We present KineSoft, a framework enabling direct kinesthetic teaching of soft robotic hands by leveraging their natural compliance as a skill teaching advantage rather than only as a control challenge. KineSoft makes two key contributions: (1) an internal strain sensing array providing occlusion-free proprioceptive shape estimation, and (2) a shape-based imitation learning framework that uses proprioceptive feedback with a low-level shape-conditioned controller to ground diffusion-based policies. This enables human demonstrators to physically guide the robot while the system learns to associate proprioceptive patterns with successful manipulation strategies. We validate KineSoft through physical experiments, demonstrating superior shape estimation accuracy compared to baseline methods, precise shape-trajectory tracking, and higher task success rates compared to baseline imitation learning approaches. KineSoft's results demonstrate that embracing the inherent properties of soft robots leads to intuitive and robust dexterous manipulation capabilities.