Torque Control

Commanding a joint directly with ‘how much torque to output,’ rather than ‘which angle to turn to.’

Torque control means the controller's output is directly the joint torque τ (units N·m), and the drive is responsible for making the motor actually produce it. A motor's torque is roughly proportional to its winding current, so this is often implemented at the motor level as a closed current loop — hence the alternate name ‘current control’; joints with a large gear ratio have significant gear friction, so current alone isn't accurate enough, requiring a joint torque sensor at the output to close the loop instead. Compared to position control (where the drive internally forces the angle to a target), torque control leaves ‘how stiff or soft’ up to the higher-level algorithm, forming the basis of impedance control, whole-body control, and operational space control, and behaves more compliantly on contact or impact — at the cost of the higher level having to compensate for gravity and friction itself, since an inaccurate model causes sagging or drift. Simulation frameworks such as legged_gym also list it as an action type alongside position and velocity.

ExampleFranka arms' libfranka interface provides a 1 kHz joint-torque command: the user's callback computes 7 joint torques, and the documentation notes these should exclude gravity and friction — the robot adds its own gravity and friction compensation on top.

Also called

Effort Control, Torque Mode, Current Control (motor-level)

Related

Position Control · Velocity Control · Impedance Control · Gravity Compensation · Joint Torque Sensor · Cascade Control

Sources

Lynch & Park, Modern Robotics (2017 preprint), Fig. 11.1 与 11.5 Force Control (Chinese)
libfranka robot.h（torque control / 1 kHz / without gravity and friction）
legged_gym: legged_robot.py（control_type P / V / T）