whole-body control

Whole-body control is a way of commanding a robot with many joints — like a humanoid with two arms, two legs, and a torso — so that all of those joints work together as one team to do several things at once. Instead of telling each joint what to do separately, you tell the robot what you want in plain goals: keep your balance, reach for that cup, hold your gaze steady, and do not push too hard on the table. The controller then solves, many times a second, for the single set of joint commands that best satisfies all those goals together. Think of it like a person reaching for a high shelf: your legs, hips, back, and shoulders all adjust at the same time, automatically, so you do not topple over while your hand goes up.

The reason it must be solved all at once is that a robot's tasks compete for the same body. Reaching far forward shifts the robot's weight, which threatens balance; turning the head changes nothing about the feet but everything about where the cameras point. A whole-body controller treats every goal as a task and works out joint motions and forces that serve them jointly, respecting hard limits along the way — joints can only bend so far, motors can only push so hard, and feet must not slip. It does this by setting up and solving a small optimization problem on every control cycle, asking which command best honors all the requests at the same instant.

Crucially, the goals are usually ranked by priority, because they cannot all be perfectly satisfied at once. Balance and not breaking your own arm sit at the top; reaching the exact target comes next; looking graceful comes last. A prioritized whole-body controller satisfies the high-priority tasks first and uses whatever freedom is left over to chase the lower ones, so a humanoid can keep its footing even if that means missing the cup by a centimeter. This is what lets legged robots walk over rubble, lift loads, and steady themselves all at the same time instead of doing them one at a time.