Abstract
Inserting a peg into a hole, snapping a connector into a socket, or mating a gear onto a shaft all require a robot to sustain deliberate contact with its environment while remaining stable under unpredictable reaction forces. Position-controlled robots struggle here: small geometric errors produce large force spikes that either damage parts or trigger fault stops. Impedance control reframes the problem by commanding a desired dynamic relationship between end-effector displacement and contact force, letting the robot yield compliantly in the force direction while remaining stiff in directions that matter for task geometry. The technique is now mature enough that industrial teams are deploying it on production lines, though careful tuning and sensing remain essential.
Impedance Control Fundamentals
Hogan’s original 1985 formulation defined impedance as the mapping from motion to force - the robot behaves like a mass-spring-damper with tunable parameters. In Cartesian impedance control, the controller renders a virtual spring connecting the actual end-effector pose to a reference pose, with a virtual damper resisting velocity and a virtual inertia shaping the transient. Choosing high stiffness along task-relevant axes (the insertion axis in a peg-in-hole task) and low stiffness in the lateral directions lets the robot accommodate fixture tolerances passively.
Modern implementations on torque-controlled arms - including the KUKA LBR iiwa, Franka Emika Panda, and ABB YuMi - close the impedance loop at joint torque level rather than at the outer Cartesian position loop. This gives bandwidth above 500 Hz and enables forces below five Newtons to be modulated reliably, which is difficult to achieve with purely position-commanded hardware through a force-torque sensor.
Combining Impedance with Force Feedback
Pure impedance control is open-loop with respect to contact force: the commanded stiffness determines the steady-state force implicitly rather than tracking a setpoint. For tighter force regulation - for example, keeping the insertion force below a damage threshold while maximizing speed - hybrid impedance/force control adds an explicit force loop in the contact-normal direction. The Robotics and Biology Laboratory at TU Berlin has demonstrated hybrid controllers on Panda arms that insert USB-C connectors with sub-millimeter tolerance at speeds competitive with manual operators.
Compliant assembly also benefits from environmental impedance estimation. If the robot can identify the stiffness of the part it is pressing against - by exciting small oscillations and observing the force response - it can adapt its virtual spring to stay in a stable contact regime. Adaptive schemes using recursive least squares have shown convergence in under 200 milliseconds in benchtop experiments, fast enough to be useful within a single assembly cycle.
Practical Challenges in Industrial Deployment
Translating lab demonstrations to production environments surfaces several issues. Joint friction in industrial arms introduces nonlinearity that distorts the rendered impedance at low velocities - a Panda arm executing a slow, force-sensitive insertion behaves differently from the same arm moving at nominal speed. Gravity compensation accuracy matters too: errors in the inertial model of the tool or workpiece appear as constant force offsets that bias the contact interaction.
Calibration drift in the wrist force-torque sensor is a maintenance concern over thousands of cycles. Teams at Bosch Research and at FANUC’s collaborative robotics group have addressed this by fitting drift models as a function of thermal state, corrected with periodic zero-force reference checks between work cycles. Fixture repeatability - the mechanical side of the problem - often dominates once the controller is well-tuned, reinforcing that impedance control is most effective as part of a systems engineering effort rather than a standalone algorithmic fix.