humanoidsdata.com

Search

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

← All glossary terms

Hardware & control

Floating base

A floating-base model represents a robot’s root body with six unconstrained spatial degrees of freedom—three for translation and three for rotation—instead of fixing it to the world. Those base coordinates are not directly actuated; a legged robot changes them through joint motion and external contact forces.

Also known as: floating-base model, free-floating base, free-floating-base model

Updated

The root pose becomes part of the state

A fixed-base manipulator attaches its base frame to the world. A floating-base model instead adds the root body’s position and orientation to the robot’s generalized coordinates. The ETH Zurich notes describe these base coordinates separately from the actuated joint coordinates.

The word “floating” describes the model, not necessarily a robot suspended in the air. A standing humanoid is commonly represented with a floating base while foot contacts are imposed as external constraints. When a foot lifts, slips or lands, those constraints change without redefining the robot’s coordinate structure.

The base cannot command itself

There is no motor that directly applies an arbitrary six-dimensional command to a humanoid’s pelvis in free space. Joint actuators alter the internal configuration, while gravity and contact forces change whole-body momentum. This makes the floating-base system underactuated and couples limb motion to balance.

Free-floating-base dynamics therefore include both the unactuated root and actuated joints. Whole-body controllers solve for joint commands and feasible contact forces together rather than treating the base as an independently driven Cartesian stage.

Floating-base data needs an estimator and a frame

Joint encoders measure relative joint positions but do not directly measure the root pose in the world. A state estimator may combine an IMU, kinematics, contact assumptions, vision or external tracking to estimate base orientation, position and velocity.

A dataset should declare the chosen root link, pose representation, reference frame, estimator, covariance where available and whether the base state is measured, estimated or simulator ground truth. It should also distinguish a model’s floating-base coordinates from ROS names such as base_link or base_footprint, which define frames rather than actuation.

Sources