Experiments on the end-point control of a two-link robot with elastic drives

摘要

This paper discusses the use of end-point feedback to improve the position control performance of manipulators that have lumped sources of flezibility within the drive trains or joints. The links are assumed to be rigid. Four different control strategies are discussed and compared. Classical control using only position and rate sensors colocated at the motors gives limited performance and poor rejection of disturbances at the end-effector. Linear-Quadratic-Regulator full-state feedback with end-point position and rate sensing dramatically improves performance and disturbance rejection, but only for a certain class of trajectories in which linearized models are effective. Robust controllers designed for larger regions of motion improve the average performance of the manipulator but at the cost of specific performance. None of the above three constantgain controllers performs well for a large class of desired trajectories in which the non-linear dynamics dominate. Non-linear computed-torque methods offer better performance for these cases. An ezperimental two-link manipulator has been constructed and ezperimental verification of these control strategies is in progress.