Stable Multi-Terrain Humanoid Walking via Convex Model Predictive Control

· 1 min read

Developed a convex MPC-based motion control framework for stable multi-terrain humanoid walking based on Cheetah-Software, using single rigid body dynamics model and computing ground reaction force for the stance leg at 200 Hz.
Implemented heuristic foothold planning, cosine swing-leg trajectories, and linear Kalman filter-based state estimation.
Enabled stable multi-terrain walking on tiles, soil, asphalt, and grass with ground contact detection.
Validated robustness against external disturbances including pushes, kicks, steel pipes, speed bumps, and curbs.
The system was showcased at the 19th Asian Games Hangzhou and demonstrated to visiting dignitaries and senior oﬀicials.

Authors

PhD Student

Hi, I am Kunzhao Ren, a first-year PhD student in Mechanical Engineering at the University of Wisconsin–Madison, with interests in legged robotics.