Friction Compensation

Estimating the friction torque inside a joint and adding it into the control command ahead of time to cancel it out.

The motors, gearboxes, and bearings inside a joint all have friction, which causes poor tracking at low speed and sticking on direction reversal (stick-slip); a 2005 survey by Bona and Indri notes this is especially critical for industrial robots. The most common model is τ_f = F_c·sgn(q̇) + F_v·q̇: F_c is Coulomb friction, constant in magnitude and opposing the direction of motion; F_v is the viscous friction coefficient, with torque proportional to joint velocity q̇; finer models add the Stribeck effect (extra static friction at breakaway) or dynamic models such as LuGre. The approach is to identify these parameters first, then add the estimated friction torque as a feedforward term on top of the motor command, or estimate it online with an observer. Because it actively cancels a modeled effect rather than correcting after an error appears (as an integral term does), it's distinct from that kind of feedback; zero-force drag, sensorless force estimation, and actuator modeling all depend on it.

ExampleWhen a collaborative arm enters kinesthetic-teaching mode, the controller adds the estimated friction torque for each joint on top of gravity compensation, so pushing the arm feels light and smooth rather than jerky and uneven.

Related

Gravity Compensation · Feedforward Control · Zero-Force Drag · Coulomb Friction · Static Friction (Stiction) and Stribeck Effect · System Identification

Sources

Bona & Indri, Friction Compensation in Robotics: an Overview (CDC-ECC 2005)
BME Robot Applications, Chapter 8: Models of Friction