humanoidsdata.com

Search

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

← All glossary terms

Hardware & control

Torque control

Torque control is a control mode in which desired joint torques, or equivalent linear forces, are the command variables sent to a robot’s low-level actuators. Unlike position control, it does not directly command where each joint should be. Motion and contact instead emerge from the applied efforts, robot dynamics, feedback, constraints and any higher-level controller.

Also known as: Joint torque control, Effort control

Updated

What the interface commands

Torque control names an actuator interface, not a complete control algorithm. A joint-space feedback controller, inverse-dynamics controller, impedance controller or learned policy can all produce torque commands. Modern Robotics shows one example: computed-torque control combines a dynamics model with feedback to turn desired motion into joint efforts.

Robots expose this interface differently. Franka’s ROS 2 hardware layer separates desired joint torque from measured link-side joint torque and states that its command excludes gravity. The surrounding controller must follow that platform-specific convention.

Torque is not end-effector force

A commanded joint torque does not directly specify a Cartesian contact force. The resulting motion and wrench also depend on the robot’s configuration, inertia, gravity, friction, transmission, other joints and environmental constraints. A Jacobian can relate joint efforts to an end-effector wrench under stated assumptions, but the two values are not interchangeable.

Commanded, estimated and sensor-measured torque are also different signals. The Universal Robots ROS 2 documentation warns that its standard joint-state effort field contains motor current rather than physical joint torque.

What torque-control data needs

A useful record keeps torque commands and feedback in separate channels, with joint order, units, sign, loop rate, saturation, filtering and delay. It also states whether gravity, friction or other feedforward terms have already been added.

For learned humanoid control, these details define the action space. Nominal torque labels without the actuator dynamics and safety limits can describe a command that another robot cannot reproduce.

Sources