Preprint reports world-model locomotion on Unitree G1

A robotics preprint reports that a world-model-assisted locomotion policy was deployed on a physical Unitree G1 humanoid using onboard proprioception and one depth stream. The authors say the robot traversed gaps, stepping stones, and stairs with an average success rate of 93.3%. The system jointly trains a recurrent world model and PPO policy so a predictive feature can guide foot placement without explicit foothold labels.