Balance Control

Keeps a legged or humanoid robot from falling over while standing, walking, or being pushed.

Balance control is the most fundamental capability of a legged robot: keeping the body from tipping over under gravity and ground-reaction forces. Traditional methods rely on simplified models: the zero moment point (ZMP, introduced by Vukobratović in 1968) is the point where the ground-reaction force produces no horizontal moment, and as long as it stays inside the support polygon (the region enclosed by the feet on the ground), the robot won't tip over the edge of its feet — early humanoids such as Honda's ASIMO planned their gaits this way; Pratt and colleagues introduced the capture point in 2006, answering where the foot should land after a push in order to stop. In response to a disturbance, robots can recover, roughly in order of increasing push strength, with ankle adjustments, hip swinging, or stepping. Today's humanoids and quadrupeds mostly use reinforcement learning: the robot is randomly pushed in simulation and the policy learns on its own to stay upright and to step to recover, then transfers to the real machine. In whole-body control, balance is usually the highest-priority task.

ExampleA standing humanoid gets pushed from the side by a person: if the push is light, the ankles alone pull the center of mass back; if it's too strong for the ankles to handle, the robot steps sideways, planting its foot near the capture point to regain balance.

Also called

Balance Maintenance, Postural Balance Control

Related

Zero Moment Point · Capture Point · Push Recovery · Support Polygon · Linear Inverted Pendulum Model (LIPM) · Ankle, Hip and Stepping Strategies

Sources

Wikipedia: Zero moment point
Pratt et al., Capture Point: A Step toward Humanoid Push Recovery (Humanoids 2006)