Torque Control
- Also called: Current Control, Effort Control
- Category: Control › Force & interaction control
- Tags: control, actuation, safety, legged-robots, compliance
Commanding how hard a joint pushes rather than where it goes — the control mode that lets a robot be compliant and feel what it touches.
Definition
Torque Control commands the force a joint applies instead of the position it should reach. A position-controlled joint answers the question "go here and hold it"; a torque-controlled joint answers "push this hard and let the world decide where you end up". The distinction changes what a robot can safely do.
Under position control, contact with an unexpected object produces whatever force is needed to reach the commanded position, which is how machines break things and hurt people. Under torque control, contact simply stops the motion at the commanded force. That makes it the foundation of compliant behaviour: force-limited assembly, safe physical interaction, legged robots absorbing ground impacts, and any task where the environment is not known precisely in advance.
In practice torque is commanded through motor current, since current and torque are proportional in a brushless motor. That means the quality of torque control depends on how faithfully the transmission passes torque through: a low-ratio, backdrivable drive lets current be a good proxy for joint torque, while a high-ratio gearbox adds friction and windup that swamp the signal. This is exactly why quasi-direct-drive actuators exist, and why joints intended for torque control are built very differently from ones intended for precision positioning.
Why it matters
Torque control is the dividing line between machines that must be caged and machines that can work beside people. It is also what made modern legged robotics possible: a leg that commands force rather than position can absorb a landing instead of fighting it. When a datasheet advertises torque control, it is implicitly claiming a whole chain of design decisions — a backdrivable transmission, current sensing, and a fast enough control loop — not merely a firmware feature.
Key applications
- Collaborative robots that must limit contact force for safety
- Legged robots absorbing ground impacts and estimating contact
- Assembly tasks like insertion where position alone would jam or break parts
- Teleoperation and force feedback, where the operator must feel the remote contact
- Exoskeletons and rehabilitation devices that assist rather than override the wearer
In practice
A torque-controlled joint typically pairs a brushless motor with field-oriented control, current sensing on each phase, and a low-ratio transmission so the relationship between current and output torque stays honest. Control runs fast — commonly a kilohertz or more — because contact events happen quickly. Many systems use it underneath a higher-level position or impedance controller rather than alone: the outer loop decides where to go, the inner torque loop decides how hard to insist.
Common misconceptions
- Torque control means the robot is safe — it makes safe behaviour possible, but the commanded torque still has to be chosen sensibly
- Any motor can be torque controlled — you need current sensing and a transmission that passes torque honestly; a high-ratio worm drive cannot
- It replaces position control — most systems run torque as an inner loop beneath a position or impedance controller
- Motor current equals joint torque — friction, gearing and temperature all sit in between, which is why joint torque sensors exist
Who should know this
- Anyone building a robot that touches unpredictable objects or people
- Builders choosing actuators for legged or collaborative machines
- Anyone whose robot breaks parts on contact instead of stopping
- Engineers reading actuator specs that advertise torque or current control
- Students learning why control mode and mechanical design cannot be chosen separately
Related terms
Related projects
- Drake: MIT robotics toolbox for model-based dynamics, optimization, and control system design
- MuJoCo MPC: DeepMind's interactive real-time model predictive control framework for MuJoCo
- PyPose: PyTorch library for differentiable robotics with deep learning and Lie group optimization
- Extreme Parkour with Legged Robots: RL system for legged robot parkour trained in under 20 hours
- ros2_control: Hardware abstraction framework for ROS 2 enabling portable robot control code
- safe-control-gym: Benchmark suite for safe learning-based control: RL and MPC under one API
Videos
- Modern Robotics, Chapter 11.4: Motion Control with Torque or Force Inputs (Part 1 of 3): Northwestern Robotics