DreamGen: Unlocking Generalization in Robot Learning through Video World Models

TL;DR

DreamGen is a novel pipeline that uses video world models to generate synthetic robot data, enabling policies to generalize across behaviors and environments with minimal real data, advancing robot learning scalability.

Contribution

We propose DreamGen, a 4-stage pipeline utilizing image-to-video generative models and pseudo-action recovery to enhance robot policy generalization across diverse tasks and environments.

Findings

Robots performed 22 new behaviors in unseen environments.

Strong correlation between video generation benchmark and policy success.

Effective generalization achieved with minimal teleoperation data.

Abstract

We introduce DreamGen, a simple yet highly effective 4-stage pipeline for training robot policies that generalize across behaviors and environments through neural trajectories - synthetic robot data generated from video world models. DreamGen leverages state-of-the-art image-to-video generative models, adapting them to the target robot embodiment to produce photorealistic synthetic videos of familiar or novel tasks in diverse environments. Since these models generate only videos, we recover pseudo-action sequences using either a latent action model or an inverse-dynamics model (IDM). Despite its simplicity, DreamGen unlocks strong behavior and environment generalization: a humanoid robot can perform 22 new behaviors in both seen and unseen environments, while requiring teleoperation data from only a single pick-and-place task in one environment. To evaluate the pipeline systematically, we introduce DreamGen Bench, a video generation benchmark that shows a strong correlation between benchmark performance and downstream policy success. Our work establishes a promising new axis for scaling robot learning well beyond manual data collection. Code available at https://github.com/NVIDIA/GR00T-Dreams.