Robot Learning from Video

Vision-language models (VLMs) have shown impressive capabilities across diverse tasks, motivating efforts to leverage these models to supervise robot learning. However, when used as evaluators in reinforcement learning (RL), today's strongest models often fail under partial observability and distribution shift, enabling policies to exploit perceptual errors rather than solve the task. We introduce SOLE-R1 (Self-Observing LEarner), a video-language reasoning model explicitly designed to serve as the sole reward signal for online RL. Given only raw video observations and a natural-language goal, SOLE-R1 performs per-timestep spatiotemporal chain-of-thought (CoT) reasoning and produces dense estimates of task progress that can be used directly as rewards. To train SOLE-R1, we develop a large-scale video trajectory and reasoning synthesis pipeline that generates temporally grounded CoT traces aligned with continuous progress supervision. This data is combined with foundational spatial and multi-frame temporal reasoning, and used to train the model with a hybrid framework that couples supervised fine-tuning with RL from verifiable rewards. Across four different simulation environments and a real-robot setting, SOLE-R1 enables zero-shot online RL from random initialization: robots learn previously unseen manipulation tasks without ground-truth rewards, success indicators, demonstrations, or task-specific tuning. SOLE-R1 succeeds on 24 unseen tasks and substantially outperforms strong vision-language rewarders, including Robometer, RoboReward, ReWiND, GPT-5, and Gemini-3-Pro, while exhibiting markedly greater robustness to reward hacking. We release all models, data, code, and demos at the anonymous page: https://philip-mit.github.io/sole-r1/

Vision-language-action (VLA) models have shown great potential in building generalist robots, but still face a dilemma-misalignment of 2D image forecasting and 3D action prediction. Besides, such a vision-action entangled training manner limits model learning from large-scale, action-free web video data. To address these issues, we propose DeFI, a novel framework that Decouples visual Forward and Inverse dynamics pretraining to exploit respective data sources, wherein video generation and action prediction are disentangled. We introduce the General Forward Dynamics Model (GFDM), pretrained on diverse human and robot videos for future prediction, and the General Inverse Dynamics Model (GIDM), trained via self-supervised learning to infer latent actions from unlabeled video transitions. These models are then integrated into a unified architecture for end-to-end finetuning on downstream tasks. In this manner, GFDM and GIDM first shine separately and then cooperate for mutual benefit. Extensive experiments on CALVIN ABC-D and SimplerEnv demonstrate state-of-the-art performance, with DeFI achieving an average task length of 4.51 for CALVIN, 51.2% success rate on SimplerEnv-Fractal benchmark and 81.3% success rate in real-world deployment, significantly outperforming prior methods.

Action-conditioned video models offer a promising path to building general-purpose robot simulators that can improve directly from data. Yet, despite training on large-scale robot datasets, current state-of-the-art video models still struggle to predict physically consistent robot-object interactions that are crucial in robotic manipulation. To close this gap, we present PlayWorld, a simple, scalable, and fully autonomous pipeline for training high-fidelity video world simulators from interaction experience. In contrast to prior approaches that rely on success-biased human demonstrations, PlayWorld is the first system capable of learning entirely from unsupervised robot self-play, enabling naturally scalable data collection while capturing complex, long-tailed physical interactions essential for modeling realistic object dynamics. Experiments across diverse manipulation tasks show that PlayWorld generates high-quality, physically consistent predictions for contact-rich interactions that are not captured by world models trained on human-collected data. We further demonstrate the versatility of PlayWorld in enabling fine-grained failure prediction and policy evaluation, with up to 40% improvements over human-collected data. Finally, we demonstrate how PlayWorld enables reinforcement learning in the world model, improving policy performance by 65% in success rates when deployed in the real world.

Human videos could reduce the reliance on expensive robot demonstrations, but extracting task-relevant features remains challenging. We introduce 3PoinTr, a method that provides robot policies with a global 3D point-track plan predicted from human videos. Given an initial point cloud, 3PoinTr predicts dense 3D point tracks spanning the full task and compresses them into a small set of tokens that remain fixed throughout execution. At each policy query, the policy combines these plan tokens with the current observation to predict robot actions. This formulation enables the policy to extract useful motion cues from human videos. We find this formulation transfers exceptionally well, even when human demonstration motions are infeasible for robots. In controlled simulation and real-world evaluations, 3PoinTr achieves an average success rate 25.0 percentage points higher than the strongest behavior-cloning and video-pretraining baselines on real-world tasks, and an improvement of 29.6 percentage points in simulation. Ablations establish the benefits of our method, including predicting non-embodiment 3D tracks once as a global plan, Perceiver-IO compression, and a residual cross-attention mechanism. Our point-track prediction network also outperforms a strong 3D-flow baseline by retaining supervision for temporarily occluded points. Project page: https://adamhung60.github.io/3PoinTr/

Current methods in robot learning are fundamentally bottlenecked by one or more of: hand-designed rewards, simulation modeling, or action supervision (e.g. teleoperation) each requiring significant domain expertise, engineering effort, and robot-operator labor. Towards reducing these bottlenecks, this work pursues observational learning via Inverse Reinforcement Learning from Observation (IRLfO) in which only access to task observations (e.g. video) is assumed. Due to the challenging setting and limitations of RL methods, IRLfO has thus far remained impractical for real-world robot learning. Here, we present the first IRL method to learn visual manipulation in the real world from scratch, and the first real-world demonstration of positive online transfer across visual manipulation tasks from scratch. In under 40 minutes, MPAIL2 learns pick-and-place from scratch to 82%, where RL and BC with equal interaction and demonstration budgets reach only 0% and 12% despite their reward and action supervision. Interactive project page with training videos: https://uwrobotlearning.github.io/mpail2/

General-purpose robot learning requires dense, instruction-conditioned feedback that can distinguish meaningful task progress from stalled, failed, or partially completed behavior. Yet obtaining such feedback at scale remains difficult, since existing approaches often rely on manual progress annotations, task-specific demonstrations, or reward models trained on curated robot datasets. We introduce TOPReward, a training-free progress reward method that probes pretrained Video-Language Models (VLMs) through their internal token probabilities rather than asking them to generate numerical progress values. Given a video prefix and a language instruction, TOPReward measures the model's likelihood that the instructed task has been completed, converting latent video-language understanding into a dense reward signal without task-specific reward-model training or manually annotated progress labels. We evaluate TOPReward on ManiRewardBench, our real-world manipulation benchmark spanning 130 unique tasks and four robot platforms, as well as on Open X-Embodiment datasets. Across these settings, TOPReward substantially outperforms prior training-free VLM reward methods on open-source models and is competitive with a trained reward-model baseline on progress-estimation metrics, while requiring no reward-model training. Additional analyses show that the reward is sensitive to the specified instruction and is not explained by time index alone. Finally, TOPReward supports downstream applications including success detection and offline reward-weighted behavior cloning.

State-of-the-art Vision-Language-Action (VLA) models excel at semantic generalization but struggle to generalize to unseen physical motions in novel environments. We introduce DreamZero, a World Action Model (WAM) built upon a pretrained video diffusion backbone. Unlike VLAs, WAMs learn physical dynamics by predicting future world states and actions, using video as a dense representation of how the world evolves. By jointly modeling video and action, DreamZero learns diverse skills effectively from heterogeneous robot data without relying on repetitive demonstrations. This results in over 2x improvement in generalization to new tasks and environments compared to state-of-the-art VLAs in real robot experiments. Crucially, through model and system optimizations, we enable a 14B autoregressive video diffusion model to perform real-time closed-loop control at 7Hz. Finally, we demonstrate two forms of cross-embodiment transfer: video-only demonstrations from other robots or humans yield a relative improvement of over 42% on unseen task performance with just 10-20 minutes of data. More surprisingly, DreamZero enables few-shot embodiment adaptation, transferring to a new embodiment with only 30 minutes of play data while retaining zero-shot generalization.

Learning highly dynamic motions from video provides a scalable alternative to expensive data collection for humanoid robots, yet it is challenged by height floating, missing kinematic and force information, and limited robustness. We present KungfuAthleteBot (KAB), a unified framework that systematically addresses these issues. We first build a high-dynamic motion dataset in collaboration with professional martial artists, alleviating data scarcity while improving motion quality. To eliminate height drift and global jitter during both aerial and landing phases, we introduce a physics-guided parabolic trajectory correction that enforces physical plausibility. To bridge the discrepancy between human and robot actuation in highly dynamic behaviors, we design a three-stage curriculum that progressively enables stable acquisition of such motions. In addition, we propose a direct, teacher-free training paradigm that jointly learns disturbance rejection and fall recovery within a single policy. To the best of our knowledge, KAB achieves the fastest recovery time reported for a unified policy ($\approx 0.7$s). Experiments on a real humanoid robot demonstrate successful learning of highly dynamic motions from video and strong robustness, including rapid recovery under severe disturbances.

We present Actron3D, a framework that enables robots to acquire transferable 6-DoF manipulation skills from monocular, uncalibrated, RGB-only human demonstration videos. Our key idea is to represent manipulation knowledge within a video as a continuous neural function over object space. At the core of Actron3D lies the Neural Affordance Function, which distills geometry, visual features, contact priors, and action flows from diverse demonstration videos into a compact 3D neural representation. During deployment, we adopt a hierarchical pipeline that retrieves the matched affordance function and transfers encoded manipulation knowledge to novel objects through coarse-to-fine differentiable optimization. Leveraging the continuous nature of Neural Affordance Function, the framework performs spatial queries over multimodal features to align demonstrations with observations and generates precise 6-DoF manipulation policy. Experiments in both simulation and the real-world demonstrate that Actron3D significantly outperforms prior methods, achieving a 14.9 percentage point improvement in the average success rate across 13 tasks while requiring only 2-3 demonstration videos per task.