World Action Models are Zero-shot Policies

TL;DR

DreamZero introduces a world action model that predicts future states and actions, enabling zero-shot generalization and real-time control in robotics, significantly outperforming vision-language-action models.

Contribution

The paper presents DreamZero, a novel world action model built on video diffusion that learns physical dynamics from diverse robot data, enabling zero-shot generalization and efficient embodiment transfer.

Findings

Over 2x improvement in generalization to new tasks and environments.

Real-time control at 7Hz with a 14B parameter model.

42% improvement in unseen task performance with minimal data.

Abstract

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.