Analysis and control of flexible-joint lightweight robots

Abstract

The focus of this work is to analyse, implement, and compare model-based controllers for a flexible joint robot (FJR). The motivation is that lightweight robots have intrinsically elastic joints and limited motor torques. If the motor dynamics would be neglected, instabilities can occur. One approach to control an FJR is by applying Singular Perturbation (SP). This method is based on the assumption that the slow dynamics can be decoupled from the fast dynamics. Since the stiffness in the joint is finite, this decoupling process is not perfect. In this work it will be analysed when the SP approach can be applied. A second part of this thesis is to address the physical limitations. If the controller demands a higher commanded motor torque that the physical system can provide, the closed-loop system can become unstable. One approach to address this physical limitation is Model Predictive Control (MPC). Prospering on the knowledge of the future trajectory and the physical system, an MPC can predict the dynamics and find a suitable control input that takes constraints on the state and input in account. Three different MPC methods are proposed. First, an MPC is considered that can propagate the full dynamics. Second, a composite controller is implemented, for which the slow component is regulated by MPC and the fast component is predicted with the theory from SP. Third, a composite controller is considered, for which an augmented PD controller is chosen as the slow component, and the MPC is embedded to dampen out fast oscillations. The controllers are compared in both simulations and a test bench.