Online learning control strategies: Theoretical and experimental study

Many industrial applications of robot manipulators involve iteratively repeated cycles of tasks. To minimize tracking errors in trajectory tracking of such manipulators, suitable learning strategies can be applied. In this paper, a novel family of online learning control laws is developed, which is primarily based on the combination of directly iterative learning control as a feedforward part and different PD based control law as a feedback part. Specifically, fixed PD gain online learning control, nonlinear PD gain online learning control, and adaptive switching PD gain online learning control are examined and compared. The convergence analysis is also provided for fixed PD gain online learning control. Experimental studies for trajectory tracking of a 2-DOF closed-loop robot manipulator to examine and verify the effectiveness of the control strategies are carried out, and the experimental results show that all these control strategies are effective. It also demonstrates that, among these three control laws, the adaptive switching PD gain online learning control is the best in terms of tracking errors.