Using Economic Iterative Learning Control for Time-Optimal Control of a Redundant Manipulator

Industrial manipulators are deployed for a range of repetitive tasks in cluttered environments in which the robot must rapidly execute safe trajectories. While nominal robot models exist, true dynamic models of deployed manipulators are typically unavailable. This paper addresses the problem of generating dynamically feasible, collision-free, time-optimal kinematic reference signals for redundant manipulators with unknown dynamics. A novel economic iterative learning control approach is developed to leverage repeated task executions to learn a time-optimal control signal for an uncertain robot model. Simulation results demonstrate the performance of the approach for a 7-DOF manipulator. An experimental analysis is performed to understand the impact of the initial reference trajectory on converged performance.