Bipedal Locomotion

A robot's ability to walk, run, and balance using only two legs — a basic skill for humanoid robots.

Bipedal locomotion means a robot walks, runs, climbs stairs, and recovers its balance after a push using only two legs. A two-legged support base is small and the center of mass is high, making it inherently unstable — the hardest category of legged locomotion. Early mainstream methods centered on the zero moment point (ZMP), the point where the ground reaction force produces zero horizontal moment: the robot plans its center of mass and footholds so the ZMP always stays within the foot's support region, the approach Honda's ASIMO used. In recent years the mainstream has shifted to reinforcement learning: training in simulation across large numbers of randomized environments, then transferring zero-shot to the real robot. A 2023 UC Berkeley project, for instance, used a causal Transformer to let an Agility Digit robot walk across plazas, grass, and other varied outdoor terrain. Bipedal locomotion is a prerequisite for humanoid robots to do mobile manipulation.

ExampleThe Berkeley team led by Ilija Radosavovic trained a walking controller entirely in simulation and deployed it directly to an Agility Digit humanoid with no fine-tuning; the robot walked stably on sidewalks, tracks, and grass, and could withstand being shoved.

Also called

Humanoid Locomotion

Related

Legged Locomotion · Humanoid Robot · Zero Moment Point · RL-based Locomotion Control · Sim-to-Real Transfer · Balance Control

Sources

Real-World Humanoid Locomotion with Reinforcement Learning (arXiv 2303.03381)
Zero moment point - Wikipedia