Learning a Slow Dynamic System

Abstract

Humans learning to control slow dynamic systems (e.g. a large excavator) perceive complex system dynamics in combination with, the less intuitive, rate control method. Earlier research has described that humans control these systems by predicting the slow response on basis of an internal model of the system dynamics. Learning this internal model can be a long and therefore a costly process in practical applications as excavators. This paper presents a training method based on the concept of Haptic Shared Control to support learning of the system dynamics. Literature on learning with Haptic Shared Control has found no consensus if this improves learning performance. However, recent literature have shown that improved learning performance is possible in timing based tasks with guidance-as-needed support. This study hypothesis that providing Haptic Shared Control training trials, used for training the desired control input, will result in lower control activity and higher performance after training. Two groups (n=10 each) had to pursuit a trajectory by controlling the slow dynamic system with a 1-DOF manipulator. The results show that subjects learning with Haptic Shared Control had similar performance, and similar excessive control activity in the frequency domain. On the other hand also a noisier joystick input, which may indicate a less developed model of the system dynamics. It is concluded that the used form of Haptic Shared Control can be used to increase the safety and performance during the slow dynamic system learning process, but is ineffective as training method.