arXiv subjects

Kehan Wen

Publications and source records attributed to Kehan Wen.

7 recordsLinked to original sources

arXiv2026

JEPA-x: Cross-Predictive Physics Grounding for Forecastable Latent Dynamics

Latent world models plan by predicting how candidate actions advance learned latent dynamics. In self-predictive models, however, the encoder and predictor are optimized jointly and can co-adapt to latent transitions that are easy to predict but weakly constrained by the physical evolution of the scene. We introduce the cross-predictive JEPA (JEPA-x), which grounds latent dynamics in privileged physical trajectories. JEPA-x treats visual observations and physical states as corresponding views of the same action-conditioned trajectory, advances both through a shared predictor, and matches each prediction to the future representations of both modalities. This encourages the action-conditioned predictor to learn a common transition rule across the two views. Privileged physical state is used only during training, leaving a visual-only model at deployment. Empirical results show that JEPA-x reduces the rollout drift of a newly fitted predictor from $0.361$ to $0.104$ and increases mean control success from $53.6\%$ to $78.2\%$ on a multi-task suite spanning six evaluation subfamilies. We additionally show that direct physical-state regression improves decodability without improving forecastability or control, indicating that the benefit comes from shaping latent dynamics rather than merely encoding physical variables.

cs.LG↗

arXiv2026

RARM: Confidence-Gated Progress Reward Modeling for RL in Manipulation

Reinforcement learning for robot manipulation is often bottlenecked by reward design, especially in long-horizon tasks: sparse success rewards provide weak supervision, while hand-crafted dense rewards are tedious to design and generalize poorly across tasks. Progress-based reward models offer a promising alternative by estimating how far an observation has advanced toward task completion, but existing approaches often require task-specific demonstrations or progress labels, and can assign high rewards to visually plausible but physically incorrect states. We introduce the Reference-Anchored Reward Model (RARM), a lightweight visual comparator that converts a single successful demonstration into a dense, progress-aware reward. RARM is trained once on general-purpose videos with a contrastive temporal objective, requiring no robot-specific data, task-specific reward labels, or per-task reward engineering. At deployment, RARM matches rollout clips to reference clips and rewards only confident forward progress, suppressing uncertain matches that may otherwise produce false-positive rewards. Across 9 simulated manipulation tasks from LIBERO and MetaWorld and 4 real-world tasks, RARM achieves the best overall success rates in subsequent RL training, with particularly large gains on long-horizon tasks such as cloth folding, where unreliable progress estimates are especially harmful.

cs.RO↗

arXiv2026

Learning Whole-Body Humanoid Locomotion via Motion Generation and Motion Tracking

Whole-body humanoid locomotion is challenging due to high-dimensional control, morphological instability, and the need for real-time adaptation to various terrains using onboard perception. Directly applying reinforcement learning (RL) with reward shaping to humanoid locomotion often leads to lower-body-dominated behaviors, whereas imitation-based RL can learn more coordinated whole-body skills but is typically limited to replaying reference motions without a mechanism to adapt them online from perception for terrain-aware locomotion. To address this gap, we propose a whole-body humanoid locomotion framework that combines skills learned from reference motions with terrain-aware adaptation. We first train a diffusion model on retargeted human motions for real-time prediction of terrain-aware reference motions. Concurrently, we train a whole-body reference tracker with RL using this motion data. To improve robustness under imperfectly generated references, we further fine-tune the tracker with a frozen motion generator in a closed-loop setting. The resulting system supports directional goal-reaching control with terrain-aware whole-body adaptation, and can be deployed on a Unitree G1 humanoid robot with onboard perception and computation. The hardware experiments demonstrate successful traversal over boxes, hurdles, stairs, and mixed terrain combinations. Quantitative results further show the benefits of incorporating online motion generation and fine-tuning the motion tracker for improved generalization and robustness.

cs.RO↗

arXiv2025

Constrained Style Learning from Imperfect Demonstrations under Task Optimality

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). Specifically, we optimize 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.

cs.RO↗

arXiv2025

Mini Diffuser: Fast Multi-task Diffusion Policy Training Using Two-level Mini-batches

We present a method that reduces, by an order of magnitude, the time and memory needed to train multi-task vision-language robotic diffusion policies. This improvement arises from a previously underexplored distinction between action diffusion and the image diffusion techniques that inspired it: In image generation, the target is high-dimensional. By contrast, in action generation, the dimensionality of the target is comparatively small, and only the image condition is high-dimensional. Our approach, \emph{Mini Diffuser}, exploits this asymmetry by introducing \emph{two-level minibatching}, which pairs multiple noised action samples with each vision-language condition, instead of the conventional one-to-one sampling strategy. To support this batching scheme, we introduce architectural adaptations to the diffusion transformer that prevent information leakage across samples while maintaining full conditioning access. In RLBench simulations, Mini-Diffuser achieves 95\% of the performance of state-of-the-art multi-task diffusion policies, while using only 5\% of the training time and 7\% of the memory. Real-world experiments further validate that Mini-Diffuser preserves the key strengths of diffusion-based policies, including the ability to model multimodal action distributions and produce behavior conditioned on diverse perceptual inputs. Code available at mini-diffuse-actor.github.io

cs.RO↗

arXiv2025

M$^3$PC: Test-time Model Predictive Control for Pretrained Masked Trajectory Model

Recent work in Offline Reinforcement Learning (RL) has shown that a unified Transformer trained under a masked auto-encoding objective can effectively capture the relationships between different modalities (e.g., states, actions, rewards) within given trajectory datasets. However, this information has not been fully exploited during the inference phase, where the agent needs to generate an optimal policy instead of just reconstructing masked components from unmasked ones. Given that a pretrained trajectory model can act as both a Policy Model and a World Model with appropriate mask patterns, we propose using Model Predictive Control (MPC) at test time to leverage the model's own predictive capability to guide its action selection. Empirical results on D4RL and RoboMimic show that our inference-phase MPC significantly improves the decision-making performance of a pretrained trajectory model without any additional parameter training. Furthermore, our framework can be adapted to Offline to Online (O2O) RL and Goal Reaching RL, resulting in more substantial performance gains when an additional online interaction budget is provided, and better generalization capabilities when different task targets are specified. Code is available: https://github.com/wkh923/m3pc.

cs.LG↗

arXiv2024

DexDribbler: Learning Dexterous Soccer Manipulation via Dynamic Supervision

Learning dexterous locomotion policy for legged robots is becoming increasingly popular due to its ability to handle diverse terrains and resemble intelligent behaviors. However, joint manipulation of moving objects and locomotion with legs, such as playing soccer, receive scant attention in the learning community, although it is natural for humans and smart animals. A key challenge to solve this multitask problem is to infer the objectives of locomotion from the states and targets of the manipulated objects. The implicit relation between the object states and robot locomotion can be hard to capture directly from the training experience. We propose adding a feedback control block to compute the necessary body-level movement accurately and using the outputs as dynamic joint-level locomotion supervision explicitly. We further utilize an improved ball dynamic model, an extended context-aided estimator, and a comprehensive ball observer to facilitate transferring policy learned in simulation to the real world. We observe that our learning scheme can not only make the policy network converge faster but also enable soccer robots to perform sophisticated maneuvers like sharp cuts and turns on flat surfaces, a capability that was lacking in previous methods. Video and code are available at https://github.com/SysCV/soccer-player

cs.RO↗