On the practical implementation of time-optical control for robotic manipulators

This paper presents experimental results for time optimal control of robotic manipulators along specified paths. It is shown that 1) time optimal trajectories yield far better tracking accuracy than the commonly used trapezoidal velocity profile at comparable motion times; 2) the delays caused by the unmodeled motor dynamics may cause significant tracking errors at high speeds; and 3) it is necessary to track time optimal trajectories with some form of model-based feedforward controller to produce the typically discontinuous optimal controls. In this paper, we track time optimal trajectories using trajectory preshaping, compensating for the motor dynamics using a simplified friction model. Implemented for the UCLA Direct Drive Arm, this is shown to reduce the tracking errors to 15% of the error obtained with a simple PD controller and without compensating for motor dynamics, and 16% of the error obtained for a trapezoidal velocity profile with comparable motion time.