Human Preferences' Optimization in pHRI Collaborative Tasks

摘要

Humans and robots working together have mutual benefits. The first is great at adaptation to new situations and has very high intellectual capabilities, while robots are very effective in assisting humans with heavy/repetitive tasks. In physical Human-Robot Interaction (pHRI), one challenge is to tune the robot's controller to make the interaction with the human as comfortable and natural as possible. Indeed, robots' perception is different from human to human. Moreover, depending on the target task, different robot tuning may be preferred. In this work, an assistive Game-Theoretic based controller is presented, and its parameters are tuned according to different subjects' preferences. Preference Based Optimization (PBO) allows optimization based on human preferences and feelings. The aim of this study is twofold: present a methodology for fast tuning of a pHRI controller according to the subjective preferences in different situations, and study general human preferences according to two different tasks. The two tasks investigated are a precise path-following task and a fast-reaching task. Experimental evaluations are conducted, and subjective and objective performances are evaluated and discussed. Finally, a questionnaire is proposed to the subjects to evaluate the applicability of the proposed method.