GE-Sim 2.0 tops WorldArena with 2B parameters for rapid robot training

GE-Sim 2.0 (Genie Envisioner World Simulator 2.0) is a closed-loop video world simulator designed to train robotic manipulation policies without real-world trials. Building on the action-conditioned video generation framework of Genie Envisioner, the model was re-trained on thousands of hours of real robot data spanning teleoperation, contact-rich interaction, and on-robot deployment. This dramatically improves action-following fidelity and trajectory coverage over its predecessor. With only 2 billion parameters, GE-Sim 2.0 tops the public WorldArena leaderboard, outperforming both dedicated robotic world models and closed-source general video generators, including much larger systems.

Three new modules close the loop from video simulation to policy learning: a state expert that decodes proprioceptive state from video latents for next-chunk prediction by downstream VLA policies; a world judge that scores generated rollouts against task instructions, providing machine-verifiable success signals; and an acceleration framework that produces a 25-frame rollout in 2.3 seconds on a single H100, with up to 4× frame skipping for long-horizon evaluation. Policies trained against GE-Sim 2.0's rollouts and rewards translate into measurable real-world gains, establishing it as a practical platform for scalable evaluation and closed-loop learning of manipulation policies.