Robot Behavior Adaptation for Human-Robot Interaction based on Policy Gradient Reinforcement Learning

Abstract

When humans interact in a social context, there are many factors apart from the actual communication that need to be considered. Previous studies in behavioral sciences have shown that there is a need for a certain amount of personal space and that different people tend to meet the gaze of others to different extents. For humans, this is mostly subconscious, but when two persons interact, there is an automatic adjustment of these factors to avoid discomfort. In this paper we propose an adaptation mechanism for robot behaviors to make human-robot interactions run more smoothly. We propose such a mechanism based on policy gradient reinforcement learning, that reads minute body signals from a human partner, and uses this information to adjust interaction distances, gaze meeting, and motion speed and timing in human-robot interaction. We show that this enables autonomous adaptation to individual preferences by the experiment with twelve subjects.