humanoid locomotion
Model-based optimal control and reinforcement learning for dynamic humanoid walking.
We study how humanoids can walk, balance, and recover over discrete and uneven terrain. Our approach blends model-based whole-body control and trajectory optimization with reinforcement learning, and extends to biomechanically inspired skills such as running jumps and soccer kicking. A recurring theme is generic locomotion policies that transfer across gaits and terrains rather than being tuned for a single scenario.
Placeholder image — replace with a lab photo.