Diffusion Policy

This paper introduces EnergyFlow, a framework that unifies generative action modeling with inverse reinforcement learning by parameterizing a scalar energy function whose gradient is the denoising field. We establish that under maximum-entropy optimality, the score function learned via denoising score matching recovers the gradient of the expert's soft Q-function, enabling reward extraction without adversarial training. Formally, we prove that constraining the learned field to be conservative reduces hypothesis complexity and tightens out-of-distribution generalization bounds. We further characterize the identifiability of recovered rewards and bound how score estimation errors propagate to action preferences. Empirically, EnergyFlow achieves state-of-the-art imitation performance on various manipulation tasks while providing an effective reward signal for downstream reinforcement learning that outperforms both adversarial IRL methods and likelihood-based alternatives. These results show that the structural constraints required for valid reward extraction simultaneously serve as beneficial inductive biases for policy generalization. The code is available at https://github.com/sotaagi/EnergyFlow.

Unlike chatbots, physical AI must act while the world keeps evolving. Therefore, the inter-chunk pause of synchronous executors are fatal for dynamic tasks regardless of how fast the inference is. Asynchronous execution -- thinking while acting -- is therefore a structural requirement, and real-time chunking (RTC) makes it viable by recasting chunk transitions as inpainting: freezing committed actions and consistently generating the remainder. However, RTC with flow-matching policy is structurally suboptimal: its inpainting comes from inference-time corrections rather than the base policy, yielding little pre-training benefit, specific fine-tuning, heuristic guidance, and extra computation that inflates the latency. In this work, we observe that discrete diffusion policies, which generate actions by iteratively unmasking, are natural asynchronous executors that resolve all limitations at once: they are fine-tuning free since inpainting is their native operation, while early stopping further provides adaptive guidance and reduces inference cost. We propose DiscreteRTC, which replaces external corrections with native unmasking, and show on dynamic simulated benchmarks and real-world dynamic manipulation tasks that it achieves higher success rates than continuous RTC and other baselines. In summary, DiscreteRTC is simpler to implement with 0 lines of additional code to enable async inpainting, faster at inference with only ~0.7 computation compared with generating actions from scratch, and better at execution with 65% higher success rate in real-world hockey defend task compared with flow-matching RTC, and 30% higher compared with training-time flow-matching RTC. More visualizations are on https://outsider86.github.io/DiscreteRTCSite/.

Contact-rich manipulation is central to many everyday human activities, requiring continuous adaptation to contact uncertainty and external disturbances through multi-modal perception, particularly vision and tactile feedback. While imitation learning has shown strong potential for learning complex manipulation behaviors, most existing approaches rely on action chunking, which fundamentally limits their ability to react to unforeseen observations during execution. This limitation becomes especially critical in contact-rich scenarios, where physical uncertainty and high-frequency tactile feedback demand rapid, reactive control. To address this challenge, we propose Tube Diffusion Policy (TDP), a novel reactive visual-tactile policy learning framework that bridges diffusion-based imitation learning with tube-based feedback control. By leveraging the expressive power of generative models, TDP learns an observation-conditioned feedback flow around nominal action chunks, forming an action tube that enables fast and adaptive reactions during execution. We evaluate TDP on the widely used Push-T benchmark and three additional challenging visual-tactile dexterous manipulation tasks. Across all benchmarks, TDP consistently outperforms state-of-the-art imitation learning baselines. Two real-world experiments further validate its robust reactivity under contact uncertainty and external disturbances. Moreover, the step-wise correction mechanism enabled by action tube significantly reduces the required denoising steps, making TDP well suited for real-time, high-frequency feedback control in contact-rich manipulation.

Bridging high-level semantic understanding with low-level physical control remains a persistent challenge in embodied intelligence, stemming from the fundamental spatiotemporal scale mismatch between cognition and action. Existing generative VLA policies typically adopt a "Generation-from-Noise" paradigm, which disregards this disparity, leading to representation inefficiency and weak condition alignment during optimization. In this work, we propose ResVLA, an architecture that shifts the paradigm to "Refinement-from-Intent." Recognizing that robotic motion naturally decomposes into global intent and local dynamics, ResVLA utilizes spectral analysis to decouple control into a deterministic low-frequency anchor and a stochastic high-frequency residual. By anchoring the generative process on the predicted intent, our model focuses strictly on refining local dynamics via a residual diffusion bridge. Extensive simulation experiments show that ResVLA achieves competitive performance, strong robustness to language and robot embodiment perturbations, and faster convergence than standard generative baselines. ResVLA also demonstrates strong performance in real-world robot experiments.

Some of the most performant reinforcement learning algorithms today can be prohibitively expensive as they use test-time scaling methods such as sampling multiple action candidates and selecting the best one. In this work, we propose FASTER, a method for getting the benefits of sampling-based test-time scaling of diffusion-based policies without the computational cost by tracing the performance gain of action samples back to earlier in the denoising process. Our key insight is that we can model the denoising of multiple action candidates and selecting the best one as a Markov Decision Process (MDP) where the goal is to progressively filter action candidates before denoising is complete. With this MDP, we can learn a policy and value function in the denoising space that predicts the downstream value of action candidates in the denoising process and filters them while maximizing returns. The result is a method that is lightweight and can be plugged into existing generative RL algorithms. Across challenging long-horizon manipulation tasks in online and batch-online RL, FASTER consistently improves the underlying policies and achieves the best overall performance among the compared methods. Applied to a pretrained VLA, FASTER achieves the same performance while substantially reducing training and inference compute requirements. Code is available at https://github.com/alexanderswerdlow/faster .

Trajectory Optimization (TO) solvers exploit known system dynamics to compute locally optimal trajectories through iterative improvements. A downside is that each new problem instance is solved independently; therefore, convergence speed and quality of the solution found depend on the initial trajectory proposed. To improve efficiency, a natural approach is to warm-start TO with initial guesses produced by a learned policy trained on trajectories previously generated by the solver. Diffusion-based policies have recently emerged as expressive imitation learning models, making them promising candidates for this role. Yet, a counterintuitive challenge comes from the local optimality of TO demonstrations: when a policy is rolled out, small non-optimal deviations may push it into situations not represented in the training data, triggering compounding errors over long horizons. In this work, we focus on learning-based warm-starting for gradient-based TO solvers that also provide feedback gains. Exploiting this specificity, we derive a first-order loss for Sobolev learning of diffusion-based policies using both trajectories and feedback gains. Through comprehensive experiments, we demonstrate that the resulting policy avoids compounding errors, and so can learn from very few trajectories to provide initial guesses reducing solving time by $2\times$ to $20 \times$. Incorporating first-order information enables predictions with fewer diffusion steps, reducing inference latency.

Diffusion policies are becoming mainstream in robotic manipulation but suffer from hard negative class imbalance due to uniform sampling and lack of sample difficulty awareness, leading to slow training convergence and frequent inference timeout failures. We propose VADF (Vision-Adaptive Diffusion Policy Framework), a vision-driven dual-adaptive framework that significantly reduces convergence steps and achieves early success in inference, with model-agnostic design enabling seamless integration into any diffusion policy architecture. During training, we introduce Adaptive Loss Network (ALN), a lightweight MLP-based loss predictor that quantifies per-step sample difficulty in real time. Guided by hard negative mining, it performs weighted sampling to prioritize high-loss regions, enabling adaptive weight updates and faster convergence. In inference, we design the Hierarchical Vision Task Segmenter (HVTS), which decomposes high-level task instructions into multi-stage low-level sub-instructions based on visual input. It adaptively segments action sequences into simple and complex subtasks by assigning shorter noise schedules with longer direct execution sequences to simple actions, and longer noise steps with shorter execution sequences to complex ones, thereby dramatically reducing computational overhead and significantly improving the early success rate.

3D policy learning promises superior generalization and cross-embodiment transfer, but progress has been hindered by training instabilities and severe overfitting, precluding the adoption of powerful 3D perception models. In this work, we systematically diagnose these failures, identifying the omission of 3D data augmentation and the adverse effects of Batch Normalization as primary causes. We propose a new architecture coupling a scalable transformer-based 3D encoder with a diffusion decoder, engineered specifically for stability at scale and designed to leverage large-scale pre-training. Our approach significantly outperforms state-of-the-art 3D baselines on challenging manipulation benchmarks, establishing a new and robust foundation for scalable 3D imitation learning. Project Page: https://r3d-policy.github.io/

Large-scale Vision-Language-Action (VLA) models offer semantic generalization but suffer from high inference latency because they adopt a low-frequency batch-and-execute paradigm. This frequency mismatch creates an execution blind spot, causing failures in dynamic environments where targets move during the open-loop execution window. We propose TIDAL (Temporally Interleaved Diffusion and Action Loop), a hierarchical framework that decouples semantic reasoning from high-frequency actuation. TIDAL operates as a backbone-agnostic scheduler for diffusion-based VLAs, using a dual-frequency architecture to redistribute the computational budget. Specifically, a low-frequency macro-intent loop caches semantic embeddings, while a high-frequency micro-control loop interleaves single-step flow integration with execution, conditioning on the latest state fused with motion cues. To handle the resulting latency shift, we introduce a temporally misaligned training strategy where the policy learns stalenessaware compensation, conditioning on stale semantic intent alongside real-time proprioception. TIDAL is architectural, making it orthogonal to system-level optimizations. Experiments show an average 2.5x performance gain over open-loop baselines in dynamic interception tasks. Despite a slight decrease in static success rates, our approach yields a 4x increase in feedback frequency and extends the effective horizon of semantic embeddings beyond the native action chunk size. Under nonpaused physics and on a real robot, TIDAL remains robust to unmodeled dynamics, while standard open-loop baselines fail due to latency-induced error accumulation.

Vision-based imitation learning has enabled impressive robotic manipulation skills, but action imitation alone provides limited supervision of the geometric consequences of robot behavior. To address this limitation, we introduce Implicit Scene Supervision (ISS) Policy, a 3D visuomotor diffusion policy with a DiT backbone that predicts continuous action sequences from point-cloud observations. ISS augments action diffusion with a supervised robot motion predictor that maps generated actions and robot-state context to end-effector motion, and then uses the predicted motion together with gripper intent to forecast future point-cloud representations. By explicitly modeling the intermediate transition from action to robot motion, ISS encourages the policy to capture how its actions affect the surrounding 3D scene. We further introduce asymmetric gradient routing to separate direct motion regression from scene-level policy supervision, together with a change-balanced objective that accounts for variations in scene-change magnitude. These auxiliary objectives provide dynamics-aware geometric supervision using only expert demonstrations, without requiring additional annotations or auxiliary modules at inference time. ISS Policy achieves state-of-the-art performance on single-arm manipulation tasks in MetaWorld and dexterous manipulation tasks in Adroit, while real-world dual-arm experiments further demonstrate its effectiveness on physical robotic manipulation. The resulting framework preserves the scalable DiT backbone and standard diffusion-policy control interface. Code and videos will be released.

Diffusion models provide an expressive framework for sampling from complex, unnormalized distributions. In this work, we extend Maximum Entropy Reinforcement Learning (ME-RL) to diffusion-based policies by introducing Diffusion-Augmented Markov Decision Processes (DA-MDPs). DA-MDPs interpret each reverse-diffusion transition as an individual reinforcement-learning decision, while only the final denoised action is executed in the environment. Our DA-MDPs follow from a principled derivation based on the variational-inference formulation of ME-RL. By augmenting policy and target trajectories with intermediate diffusion variables, we obtain a tractable reverse-KL upper bound via the data-processing inequality. This bound decomposes across denoising transitions, yielding diffusion-augmented variants of soft rewards, value functions, and local policy objectives. This provides a general framework for adapting ME-RL algorithms to diffusion policies while differentiating through only one diffusion transition at a time. We instantiate the framework with PPO, REPPO, and a maximum-entropy extension of WPO. Experiments demonstrate improved continuous-control performance, benefits from additional diffusion steps, and memory-efficient training. On the StackCube and PushT manipulation tasks, DA-MDP methods learn alternative successful strategies from the same initial state and achieve higher success rates and generally higher success-weighted mode entropy than the Gaussian ME-RL baseline. We also demonstrate successful training when using action chunking.

Ad hoc teamwork (AHT) requires agents to collaborate with previously unseen teammates, which is crucial for many real-world applications. The core challenge of AHT is to develop an ego agent that can predict and adapt to unknown teammates on the fly. Conventional RL-based approaches optimize a single expected return, which often causes policies to collapse into a single dominant behavior, thus failing to capture the multimodal cooperation patterns inherent in AHT. In this work, we introduce PADiff, a diffusion-based approach that captures agent's multimodal behaviors, unlocking its diverse cooperation modes with teammates. However, standard diffusion models lack the ability to predict and adapt in highly non-stationary AHT scenarios. To address this limitation, we propose a novel diffusion-based policy that integrates critical predictive information about teammates into the denoising process. Extensive experiments across three cooperation environments demonstrate that PADiff outperforms existing AHT methods significantly.

Visuomotor policies trained via behavior cloning are vulnerable to covariate shift, where small deviations from expert trajectories can compound into failure. Common strategies to mitigate this issue involve expanding the training distribution through human-in-the-loop corrections or synthetic data augmentation. However, these approaches are often labor-intensive, rely on strong task assumptions, or compromise the quality of imitation. We introduce Latent Policy Barrier, a framework for robust visuomotor policy learning. Inspired by Control Barrier Functions, LPB treats the latent embeddings of expert demonstrations as an implicit barrier separating safe, in-distribution states from unsafe, out-of-distribution (OOD) ones. Our approach decouples the role of precise expert imitation and OOD recovery into two separate modules: a base diffusion policy solely on expert data, and a dynamics model trained on both expert and suboptimal policy rollout data. At inference time, the dynamics model predicts future latent states and optimizes them to stay within the expert distribution. Both simulated and real-world experiments show that LPB improves both policy robustness and data efficiency, enabling reliable manipulation from limited expert data and without additional human correction or annotation.

As robots become more integrated in society, their ability to coordinate with other robots and humans on multi-modal tasks (those with multiple valid solutions) is crucial. Such behaviors can be learned from expert demonstrations via imitation learning (IL), but when expert demonstrations are multi-modal, standard IL approaches usually average across modes or collapse to a single mode, preventing effective coordination. Being inspired by diffusion models' ability to capture complex multi-modal trajectory distributions in single-agent settings, we develop a diffusion-based framework for coordinated multi-modal behavior in multi-agent systems. However, existing multi-agent diffusion approaches typically require a centralized planner or explicit communication among agents. This assumption can fail in real-world scenarios where robots must operate independently or with agents like humans that they cannot directly communicate with. Therefore, we propose MIMIC-D, a joint training with decentralized execution paradigm for multi-modal multi-agent IL via diffusion. We jointly train all agents' policies with only local information to achieve implicit coordination. In simulation and hardware experiments, our method exhibits robust multi-modal coordination behavior in various tasks and environments, improving upon state-of-the-art baselines.