Human skill evaluation by characteristics of input command decision

This paper addresses a human skill quantification technique with human command decision characteristics in human-machine systems. In machine operation systems, getting operation skills is not easy and required learning time. Proposed technique aims at realizing assist system to human skill achievement. However, in previous works, the operator skill is evaluated by tracking error for tracking tasks. In this paper, a new evaluation indexes for human operation skill based on prediction errors by a human model consists of multiple ANNs. The ANNs have input signals of different time series respectively. By observing these prediction errors, we can get how long or what time information is used for input command decision during the task. By some experiments with 20 participants, the distribution of prediction error is change as getting skills and this trend is not depend on the tracking errors.