Legged Locomotion and Sim-to-Real Reinforcement Learning: Zero-Moment Point, Centroidal MPC, Actuator Transparency (Quasi-Direct Drives), Domain Randomization, and Rough-Terrain Gaits

Deep End Era. Engineering with Sankalp · Oct 29, 2026 · 1h 12m

Put a wheeled robot in front of a flight of steep refinery stairs or a trench of loose crushed rock, and its mission is over. Put an early humanoid robot on that same rock pile using classical flat-footed Zero-Moment Point walking, and the first stone that rolls under its ankle topples fifty thousand dollars of aluminum and harmonic drives onto the ground. Today we trace the fifty-five-year journey of legged locomotion from Vukobratović and Raibert to MIT Cheetah's quasi-direct drives, convex MPC, and ETH Zurich's sim-to-real reinforcement learning policies.

Chapters

00:00:00 Opening: Wet…