Whole-body control is the part of humanoid robotics that coordinates many joints and contacts at the same time. Instead of asking the legs to walk, the arms to reach and the torso to stay upright as separate problems, a whole-body controller treats those motions as one coupled physical task. That matters because moving an arm changes the robot's mass distribution, a foot contact limits where the body can accelerate, and a manipulation force can disturb balance.
Why humanoid robots need whole-body control
A humanoid has many degrees of freedom and must obey several constraints simultaneously. The feet may need to remain planted, the center of mass must stay recoverable, a hand may need to follow a tool path, and joint limits and actuator capabilities still apply. A controller that optimizes only one body part can produce a motion that is locally correct but physically incompatible with the rest of the robot.
Whole-body control solves this coordination problem. It provides a common layer between high-level goals such as "walk to the table and pick up the box" and the joint commands sent to the robot's actuators.
Tasks, priorities and constraints
Many whole-body controllers represent behavior as a set of tasks. One task may regulate the robot's center of mass, another may keep a foot fixed to the floor, and another may move a hand toward a target. The controller then computes joint motion or torque commands that satisfy these objectives while respecting joint, contact and stability constraints.
Some systems use strict task hierarchies, meaning a safety-critical balance or contact objective has priority over a lower-priority hand motion. Others combine objectives in an optimization problem with different weights. The exact mathematics varies, but the engineering question is similar: how can the robot achieve a useful action without violating the physical conditions that keep it stable?
Whole-body control and balance
Humanoid robot balance is not a separate afterthought. It is one of the central constraints on whole-body motion. Reaching far with an arm, lifting an object or rotating the torso can shift the center of mass and change the contact forces at the feet. A whole-body controller must account for those effects and may compensate with the hips, knees, ankles, torso or a change in foot placement.
For dynamic tasks, stability cannot be reduced to keeping a point inside a fixed region. The controller must also consider momentum, contact timing and whether the robot can recover from the motion it is creating.
Whole-body control and locomotion
Humanoid locomotion focuses on walking, running, gait generation and terrain adaptation. Whole-body control is broader. It can coordinate locomotion with upper-body goals, such as carrying a long object while walking, turning the torso without losing balance or using the arms to assist a recovery step.
This overlap is why modern humanoid systems increasingly discuss whole-body locomotion and loco-manipulation. The ownership boundary is still useful: locomotion owns how the robot moves through the environment, while whole-body control owns how competing full-body objectives are coordinated.
Model-based whole-body control
Model-based approaches use robot kinematics and dynamics explicitly. They may solve inverse kinematics, inverse dynamics, quadratic programs or model-predictive-control problems to choose feasible motions and forces. Their strengths include interpretable constraints and the ability to encode physical requirements directly. Their limitations include model error, computational cost and the difficulty of describing every real-world behavior with hand-designed objectives.
Learned whole-body control
Learning-based controllers can acquire complex full-body skills from simulation, demonstrations or motion data. A policy may learn to track reference motions, recover from disturbances or switch between movement modes. Modern systems often combine learning with classical control rather than replacing all model-based structure.
Training commonly relies on robot simulation, while deployment must handle the differences described by sim-to-real robotics. The controller may also use tactile, proprioceptive and visual feedback to correct errors that were not present in the training environment.
Whole-body control for manipulation
Manipulation becomes a whole-body problem when the task requires more than a stationary arm. Opening a heavy door, lifting a large object or reaching while stepping can demand coordination from the feet through the torso to the hand. This connects whole-body control to dexterous manipulation, but the two concepts are not the same. Dexterous manipulation focuses on controlling objects through hand and contact behavior; whole-body control focuses on coordinating the complete robot around the task.
What sensors does a controller use?
Typical inputs include joint encoders, inertial measurements, force or torque sensing, foot-contact information and perception from cameras or depth sensors. Proprioceptive feedback tells the controller what the body is doing, while exteroceptive perception helps it understand terrain, obstacles and task targets. The useful sensor set depends on the behavior and the robot architecture.
Why whole-body control is becoming more important
Humanoid robots are moving from isolated demonstrations toward longer tasks that combine walking, reaching, carrying and manipulation. Those behaviors expose the coupling between every part of the body. A capable whole-body controller is therefore less about making a robot perform a spectacular single motion and more about making many objectives coexist safely and continuously.
