The application of a robot manipulator to the task of parts assembling or collaboration with human workers requires compliant control and intrinsic safety. As a result, it is necessary to exert accurate torque on each joint of the robot through torque sensing and implementing closed-loop joint torque control. This torque servo system is required to track reference torque signals while operating under the influence of motor friction, flexibility of the harmonic drive, noise from the sensor, robot dynamics modelling error and other unknown certainties, resulting in large control efforts.
This paper focuses on providing better compliance control for collaborative robots and proposes a joint torque controller design under development with active disturbance rejection concept. The controller is designed through a novel extended state observer to estimate and compensate for the unmodelled dynamics of the system, nonlinearly variable motor friction, and other uncertainties. Then, a simple proportional differential controller is designed to produce control law. In spite of the remarkable performance in dealing with the mechanical dynamics of the joint actuator, the original controller does not work well with the electrical factor of the joint actuator due to the limited current loop bandwidth in the hardware of motor and driver. To eliminate the detrimental effect of the time delay in current servo, a predictive output method based on a nonlinear tracking differentiator (TD) is used to improve the controller within the framework of active disturbance rejection control. Both simulations and experiments are conducted on a prototype one degree of freedom manipulator with a joint torque sensor. The results demonstrate the enhancement of both the system stability and disturbance rejection performances. Based on the proper treatment of actuator delay, the dominant effect of the motor friction and the flexibility of the harmonic drive has been reduced to insignificance. Moreover, the proposed controller is easy to implement because the explicit dynamic model of the system is not required.