Learning of social skills for Human-Robot Interaction by hierarchical HMM and interaction dynamics

Abstract

In Human-Robot Interaction, an intelligent robot should be able to learn social skills and reproduce such skills according to dynamic human's behaviors. To this end, both motion trajectories of a human and a robot are autonomously segmented, after which social skills are represented by combining hierarchical hidden Markov models and interaction dynamics (i.e., mass-spring-damper) to include three abilities of recognition, reproduction, and adaptation. To validate this, we present the experimental results when using a humanoid robot that performs several social skills.