Practical and robust incremental model predictive control for flexible-joint robots

Abstract

This article proposes an optimization-based method for robust yet efficient control of flexible-joint robots by using the model predictive control approach. The time-delay estimation (TDE) technique is used to approximate uncertain and nonlinear dynamic equations, where neither concrete knowledge of mathematical system model parameters is required in the approximation, thus granting the model-free property for dynamics compensation and real-time system linearization. TDE is integrated with model predictive control, which is designated as the incremental model predictive control (IMPC) framework. This approach guarantees the tracking performance of the flexible joint robot with input and output constraints, such as motor torque and joint states. Moreover, the proposed controller can practically circumvent high-order derivatives in implementation while providing robust tracking, a capability that conventional methods for flexible joint robots often face challenges due to the inherent nature of their high-order dynamics. The input-to-state stability of IMPC in a local region around the reachable reference trajectory is theoretically proven, and the high approximation accuracy of the resulting incremental system is analyzed. Finally, a series of experiments is conducted on a flexible-joint robot to verify the practical effectiveness of IMPC, and superior performance in terms of high accuracy, high computational efficiency, and constraint admissibility is demonstrated.