Multiple contextual task recognition for sharing autonomy to assist mobile robot teleoperation

Abstract

To efficiently facilitate autonomy sharing for assisting mobile robot teleoperation, in this paper we propose a method to recognize four contextual task types executed by the human operator: doorway crossing, object inspection, wall following and robot docking, which extends our previous approach, where only the first two task types were considered. We employ a set of simple but highly distinctive task features to efficiently describe each task type, which is adopted by a Gaussian Mixture Regression (GMR) model combined with a recursive Bayesian filter (RBF) to infer the most probable task the human operator executes across multiple candidates during operation. We demonstrate the effectiveness of the approach with a variety of tests in a cluttered indoor scenario.