Variable Impedance Control

Impedance control whose stiffness and damping change in real time with the task phase or sensed force, stiff when needed and soft when needed.

Impedance control makes a robot's end-effector behave like a spring and damper: F = K(x_d − x) + D(ẋ_d − ẋ), where x_d is the desired position, x is the actual position, K is stiffness (higher means ‘stiffer’), and D is damping. Ordinary impedance control keeps K and D fixed; variable impedance control instead lets them change over time, with the task phase, or with force feedback — an idea inspired by how humans actively adjust their arm stiffness. This balances precision and safety: stiff in free space for accurate tracking, soft on contact so the robot doesn't damage what it touches. The difficulty is that time-varying stiffness can inject energy into the system and break stability, which is usually kept in check with passivity theory and so-called ‘energy tanks.’ K and D can be hand-designed, learned from demonstrations, or output directly as a reinforcement-learning action: the 2019 VICES method has the policy output both the end-effector motion and the impedance gains, and is more sample-efficient and safer than fixed impedance on contact-rich tasks like wiping and door-opening. Federico Abu-Dakka and Matteo Saveriano wrote a systematic survey in 2020.

ExamplePeg-in-hole assembly: the arm uses high stiffness while carrying the peg for fast, accurate positioning. Once the peg touches the rim of the hole, lateral stiffness is lowered so it can slide along the chamfer into the hole, while stiffness along the insertion direction stays higher to keep pushing forward.

Also called

VIC, Time-Varying Impedance Control

Related

Impedance Control · Compliance Control · Admittance Control · Cartesian Impedance Control · Force-aware / Compliant Policy Learning · Peg-in-Hole Insertion

Sources

Abu-Dakka, Saveriano: Variable Impedance Control and Learning—A Review (Frontiers in Robotics and AI, 2020)
Martín-Martín et al.: Variable Impedance Control in End-Effector Space: An Action Space for RL in Contact-Rich Tasks (arXiv:1906.08880)