EVA: Aligning Video World Models with Executable Robot Actions via Inverse Dynamics Rewards

2026-03-18

AI Summary

This work addresses the limitation of existing video world models, which generate temporally coherent visual sequences but lack constraints ensuring robotic executability, often leading inverse dynamics models to decode actions that violate rigid-body or kinematic constraints. To bridge this executability gap, the authors propose the Executable Video Alignment (EVA) framework, which introduces a reinforcement learning reward mechanism based on an inverse dynamics model during the post-training phase of the video world model. EVA repurposes the inverse dynamics model as a reward function to explicitly align generated trajectories with physically feasible actions, optimizing velocity, acceleration, and jerk while penalizing behaviors that exceed embodiment constraints. Experiments on the RoboTwin benchmark and a real dual-arm robot demonstrate that EVA significantly reduces embodiment artifacts in generated rollouts and improves downstream task success rates.

Technology Category

• Machine Learning: Imitation Learning & Inverse Reinforcement Learning
• Intelligent Robots: Embodied AI
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Video generative models are increasingly used as world models for robotics, where a model generates a future visual rollout conditioned on the current observation and task instruction, and an inverse dynamics model (IDM) converts the generated frames into executable robot actions. However, current video world models lack explicit executability constraints. As a result, visually coherent rollouts may still violate rigid-body and kinematic consistency, producing unstable or infeasible control commands when decoded by an IDM. We refer to this mismatch between visual generation and physically executable control as the executability gap. While this gap can be mitigated at inference time using techniques such as rejection sampling, such approaches are inefficient due to the high cost of video generation. In this paper, we leverage the executability gap as a training signal and introduce Executable Video Alignment (EVA), a reinforcement-learning post-training framework for aligning video world models. EVA trains an inverse dynamics model on real robot trajectories and repurposes it as a reward model that evaluates generated videos through the action sequences they induce, encouraging smooth motions measured by velocity, acceleration, and jerk while penalizing actions that violate embodiment constraints. Importantly, the reward remains informative even when generated videos contain severe visual artifacts, since such artifacts typically translate into unstable or out-of-bound actions. Experiments on the RoboTwin benchmark and a real bimanual robot show that EVA reduces embodiment-specific artifacts in generated rollouts and improves downstream task execution success.

Problem

• executability gap
• video world models
• inverse dynamics model
• robot actions
• kinematic consistency

Innovation

• Executable Video Alignment
• inverse dynamics model
• world model alignment
• executability gap
• reinforcement learning