Gain-scheduling control of a monocular vision-based human-following robot

Abstract

This paper presents a gain-scheduling controller for a human-following robot. Typical controllers use either a point-to-point approach where the relative orientation between human and platform is uncontrolled, or a direction-based approach which corrects orientation errors at the expense of additional actuation. The authors describe the flaws and benefits of each and argue that a gain-scheduling controller combining the two is better equipped to deal with the challenges of human-following. A model of our feature-based, single camera vision system is presented and used to show that the gain-scheduling controller offers better performance than its components, and actual responses to a human following task are used to corroborate this.