In robot applications such as assembly, welding, or bonding, position and velocity controls are required simultaneously. Conventional control methods such as fixed-gain linear feedback controllers, or computed torque method cannot face these control performances. Recently, iterative learning control techniques are applied to robotic manipulators, but these techniques have the limit of learning for the arbitrary trajectory tracking.
In this study, therefore, a learning algorithm, which is aimed at an arbitrary desired trajectory tracking within an allowable error tolerance is developed. A SCARA type robot that has dominant dynamic effects and heavy nonlinearities is selected as a model for this study.
In this paper, Our learning scheme obtains learning data from the outputs simulated with respect to some typical trajectory patterns using a conventional controller. This newly developed controller is learned from one specific trajectory, then this control scheme is simulated with respect to arbitrary desired trajectories. Simulation results represent that the trajectory tracking error is sufficiently small throughout the whole workspace.
