The capture point is where a legged robot must step to come to a complete stop — the modern balance concept behind push recovery and dynamic walking, letting robots stumble, catch themselves, and stay upright.
The capture point is the exact spot a robot needs to plant its foot to stop moving and balance. Get shoved? Step onto your capture point and you come to rest instead of falling — the same instinctive step a person takes when pushed.
Shove a person and they instinctively step in the direction they're falling to catch themselves. Teaching a robot that same reflex is the job of the capture point — a more dynamic balance idea than the classic zero-moment point.
The idea
Model the robot as an inverted pendulum — a mass (center of mass) balancing over a pivot foot. If the mass is moving, there's exactly one place to plant the foot such that the robot's momentum is arrested and its center of mass comes to rest directly above the new foot. That place is the capture point. Step there, and you stop; step short or long, and you keep toppling or have to step again.
Step where your momentum takes you
The capture point depends on the robot's velocity — the faster it's moving, the farther out it must step. Placing the foot there converts a stumble into a stable stop.
Why it matters
ZMP-based walking keeps the foot flat and never lets the robot "fall," which makes it stable but cautious. The capture point embraces the opposite view: a walking or shoved robot is falling, and balance means stepping to catch it. This unlocked:
Push recovery. A robot shoved off balance computes its capture point and takes a quick corrective step, exactly like a human — the reason modern humanoids can take a hit and stay up.
Dynamic walking. Each step naturally targets the capture point, giving faster, more natural, less stiff gaits.
Rough terrain. Reasoning about where to step, not just how to keep feet flat.
Refinements
The basic capture point assumes an idealized pendulum; real robots use extensions (the "N-step capturability" idea — can I stop in one step, or do I need several?) and blend it with whole-body control and ZMP methods. It's a key concept behind the balance controllers on robots like Atlas.
Why it matters
The capture point reframed legged balance from "never fall" to "know how to catch yourself." It's the foundation of push recovery and agile, human-like walking — a central idea in getting humanoids to move robustly in the real world.
