Iterative Learning of Feasible Time-optimal Trajectories for Robot Manipulators

摘要

Time-optimal trajectories describe the minimum execution time motion along a given geometric path while taking system dynamics and constraints into account. By using a model of the real plant, inputs are provided that ought to yield minimal execution time and good tracking performance. In practice however, due to an imperfect model, the computed inputs might be suboptimal, result in poor tracking or even be infeasible in that they exceed given limits. This paper therefore presents a novel two-step iterative learning approach for industrial robots to find time-optimal, yet feasible trajectories and improve the tracking performance by repeatedly updating the nonlinear robot model and solving a time-optimal path tracking problem. The proposed learning algorithm is experimentally validated on a serial robotic manipulator, which shows that the developed approach results in reduced execution time and increased accuracy.