Trajectory preshaping for high-speed articulated systems

This paper presents a method for reducing the tracking errors of articulated systems, moving along specified paths at high speeds. It consists of preshaping the reference trajectory to account for the dynamics of the feedback controller. The trajectory is assumed to be feasible, satisfying the known manipulator dynamics and the actuator constraints. The correction term, added to the nominal trajectory, is computed by filtering the nominal control inputs through the inverse of the feedback controller. A learning procedure is also presented to account for unknown parameters of the feedback controller and for unmodeled dynamics. It consists of an iterative parameter optimization that minimizes the tracking error along the path at each joint. The initial guesses for this optimization are selected to ensure that the tracking error is not worse than the error with no preshaping. The method is demonstrated experimentally for the UCLA Direct Drive Arm and for the AdeptOne industrial robot, achieving mean tracking errors as low as 0.11 mm at top speeds of 1.8 mJs.