Show, Attend and Interact: Perceivable Human-Robot Social Interaction through Neural Attention Q-Network

摘要

For a safe, natural and effective human-robot social interaction, it isessential to develop a system that allows a robot to demonstrate theperceivable responsive behaviors to complex human behaviors. We introduce theMultimodal Deep Attention Recurrent Q-Network using which the robot exhibitshuman-like social interaction skills after 14 days of interacting with peoplein an uncontrolled real world. Each and every day during the 14 days, thesystem gathered robot interaction experiences with people through ahit-and-trial method and then trained the MDARQN on these experiences usingend-to-end reinforcement learning approach. The results of interaction basedlearning indicate that the robot has learned to respond to complex humanbehaviors in a perceivable and socially acceptable manner.