Learning Feasible and Time-optimal Trajectories for Robot Manipulators

Classical time-optimal path tracking algorithms make use of an a priori known model description to compute a minimum-time trajectory. Due to a possible model mismatch, this result may not be time-optimal or even be infeasible – torque limits could be exceeded and yield large tracking errors or even halt operation. To avoid such situations, this paper presents an algorithm that combines iterative learning of the model and time-optimal path tracking.status: publishe