D-Robotics WM-LOCO hits 92.5% success on Unitree G1 terrain tests

🤖How can a humanoid robot navigate terrain where every foothold matters?

WM-LOCO jointly trains a recurrent world model and a PPO locomotion policy, using proprioception and a single onboard depth image to provide predictive context for foot placement.

From stepping stones and wide gaps to stairs and narrow balance beams, WM-LOCO achieved:
✅ 96.9% / 96.0% / 94.0% / 92.1% success rates across the four terrain types in simulation
✅ 92.5% success across 40 real-world trials on Unitree G1

View on X