Decoupling Planning and Tracking Lets Humanoid Robots Move Like Athletes

Researchers introduced a method that separates high-level motion planning from low-level balance tracking, letting humanoid robots keep a specific athletic style without losing stability. An adaptive speed controller and training at randomized speeds in simulation prepared the robots for real-world tasks without external sensors.

• Planner handles motion style while tracker maintains balance
• Training at randomized speeds in simulation improves resilience to delays
• Motions are extracted from broadcasts and motion capture, then retargeted to robot skeletons
• Tests on two platforms ran without external sensors or cameras