Impedance control lets a collaborative robot adapt its motion to interaction forces rather than behave only as a rigid position-following machine. Delivering that behavior reliably is challenging when flexible transmissions and motor friction affect the torque applied at each joint. A cascaded control architecture addresses these effects with an inner joint torque servo based on active disturbance rejection and an outer impedance loop that sets the robot’s compliant response.
The approach is intended for flexible-joint manipulators operating both without contact and in constrained tasks such as insertion. Its practical value lies in treating actuator-side uncertainty as a disturbance to estimate and compensate, instead of requiring a detailed model of every nonlinear effect.
Why impedance control is challenging in flexible-joint robots
Conventional industrial robots are often designed to perform repeatable position-controlled tasks. Collaborative applications, by contrast, may require controlled force at the end effector as the robot contacts a fixture, workpiece, or other environment. Joint torque control can support this interaction, but the actuator is not an ideal torque source.
Two effects are particularly important. First, a harmonic reducer introduces torsional compliance between the motor and the robot link. The motor-side and link-side positions can therefore differ, and the joint behaves as part of a coupled flexible system. Second, friction in the motor and transmission—including Coulomb friction and stiction—can distort the relationship between commanded motor torque and measured output torque.
These effects can cause tracking error, vibration, or unwanted resistance during contact. Their impact may vary with operating conditions, and link motion itself can disturb the actuator torque loop. A controller that depends on a highly accurate model of all these effects can be difficult to design and maintain.
Using active disturbance rejection in the joint torque servo
The proposed architecture separates control into two loops. An inner loop regulates joint torque, while an outer loop commands the torque needed for the desired impedance. The inner loop uses active disturbance rejection control (ADRC) to estimate and compensate for the combined effects of uncertain actuator dynamics and external disturbances.
The role of the extended state observer
At the heart of ADRC is an extended state observer (ESO). It estimates the joint torque and the total disturbance affecting the actuator. That disturbance can include motor friction, link motion, actuator dynamics, and other unmodeled effects. Rather than separately identifying each source, the controller treats their combined influence as an additional state to be estimated in real time.
The controller then uses the observer’s estimate to counteract the disturbance while a proportional-derivative (PD) feedback component tracks the reference torque. In effect, disturbance compensation helps the torque loop behave more like a predictable, linear system, even when friction or motion at the robot link would otherwise interfere.
Observer bandwidth is an important tuning consideration. A higher bandwidth can improve disturbance estimation, but it is constrained by noise in the torque feedback signal. The torque-loop bandwidth also needs to be selected with awareness of actuator limits and sensor noise: increasing it can improve tracking but may demand greater motor torque and make the system more sensitive to measurement noise.
Building the outer impedance loop
Once the inner servo can track joint torque more reliably, the outer controller can focus on the robot’s desired interaction behavior. In the described framework, a PD controller with gravity compensation generates the reference joint torque from position and velocity errors. The stiffness and damping settings determine how the robot responds to displacement and motion, allowing its apparent compliance to be adjusted for different tasks.
This division of responsibilities simplifies control design. The inner loop addresses actuator-level disturbances; the outer loop specifies the robot’s motion and compliant behavior. The analysis in the paper establishes input-output stability conditions for the cascaded system, including the effect of actuator dynamics and disturbances. It also shows why sufficient bandwidth in the closed-loop torque actuators matters to applying a rigid-body controller to a flexible-joint robot.
In practical terms, the control structure is relevant where a robot must move freely toward a target and then respond safely and consistently as contact begins. Stiffness can be selected according to the task direction and interaction requirements rather than treating every axis as equally rigid.
Experimental findings in a peg-in-hole task
The method was evaluated on a seven-degree-of-freedom collaborative robot with torque sensing at the joints. In joint-level tests, the ADRC-based torque controller was compared with a conventional PID controller and a passivity-based method. Simulations showed that the ESO could estimate the combined disturbance and improve reference torque tracking under simulated link motion and motor friction. In the reported comparison, the ADRC controller also had lower robustness-to-total-disturbance error than the two alternatives, while its computation time remained suitable for the test system’s real-time cycle.
Experiments on a single joint involved pressing the robot’s end effector against a fixture and tracking square-wave and harmonic torque references. Compared with the passivity-based controller, ADRC produced more stable torque output and better precision in the tested conditions. The observer helped compensate for link-motion effects and motor friction; it also addressed steady-state deviation associated with error in the motor torque constant.
The complete impedance framework was then tested in a two-dimensional peg-in-hole assembly task. The robot approached the hole in free space, made contact with its upper surface, and continued through constrained motion as the peg entered. The recorded joint velocities remained stable through the transition. During insertion, the robot maintained contact with the hole wall, demonstrating the use of the torque servo and impedance loop in a task that combines unconstrained and constrained motion.
Practical considerations for automation teams
For engineers and integrators considering a similar control strategy, the findings point to several design priorities:
- Measure joint torque directly where possible. The described testbed uses joint torque sensors to close the inner feedback loop.
- Treat friction and link motion as meaningful disturbances. Both can affect torque tracking, especially around direction changes and during contact.
- Tune observer and control bandwidths together. Tracking speed, sensor noise, and actuator saturation all influence the usable settings.
- Separate force behavior from motion behavior. An inner torque servo and outer impedance controller give each loop a distinct role.
- Validate across contact conditions. Free-space motion and constrained interaction can present different disturbance dynamics.
Conclusion
Impedance control on a flexible-joint collaborative robot depends on more than choosing an outer-loop stiffness and damping. Joint flexibility, friction, and link motion can compromise the torque that the robot actually applies. Combining an ESO-based ADRC torque servo with an outer PD impedance controller provides a structured way to estimate and reject these disturbances while retaining a relatively simple outer control law. For automation teams, the key implication is to treat torque-loop performance, sensor noise, and bandwidth as foundational design considerations for reliable contact tasks such as assembly and insertion.
Access Official Technical PDF