humanoidsdata.com

Search

Search datasets, articles, and glossary terms for humanoids and embodied AI.

← All glossary terms

Hardware & control

Whole-body control

Whole-body control coordinates a humanoid’s full multibody state, actuated joints and environmental contacts to pursue several motion or force objectives while respecting constraints such as balance, joint limits and friction. It is a family of hierarchical or optimisation-based methods, not one algorithm; implementations may output joint positions, accelerations or torques.

Also known as: WBC, Whole body control

Updated

Why the whole body enters the problem

Moving one humanoid limb changes more than that limb’s pose. Reaching with a hand shifts momentum and balance, while feet, hands or knees may carry contact forces. The floating base is not directly actuated, so its motion must be influenced through joint torques and environmental contacts.

The Sentis and Khatib framework therefore models supporting contacts, operational tasks and residual posture together. Typical objectives include hand or foot motion, centre-of-mass behaviour, gaze and contact force, subject to structural and environmental constraints.

Hierarchies and optimisation are different implementations

Classical hierarchical control gives critical constraints higher priority, then projects lower-priority tasks into the remaining motion space. This can stop a posture objective from overriding balance or a contact condition. The original hierarchical whole-body formulation describes constraints, operational tasks and posture as distinct priority levels.

Optimisation-based controllers instead express dynamics, contacts, limits and tasks as equalities, inequalities and costs. Dynamic whole-body motion generation demonstrates a strict hierarchy with full multibody dynamics, while Task Space Inverse Dynamics treats prioritised motion and force control. Kinematic whole-body controllers may output positions or velocities; inverse-dynamics versions commonly output torque. “WBC” alone does not identify the interface.

What whole-body training data needs

A reference pose sequence is insufficient for contact-rich control. Useful records synchronise base state, joint state, task targets, contact schedule, contact wrench, controller priorities or weights, active constraints, model version and final commands.

Learned controllers also need a clear split between observations available on hardware and simulator-only privileged state. Preserving these details shows whether a policy learned coordinated physical control or merely reproduced kinematics with hidden information.

Sources