Emotional Understanding and Behavior Learning for Haru via Social Reinforcement Learning

As a new type of human companion, social robots are becoming more and more popular and expected to being fully integrated with human daily life in the near future. Being able to correctly perceive the emotions of users and react to it can increase the sense of trust, affinity, and social presence of human-robot interaction. In this paper, we propose a human-centered reinforcement learning strategy to train social robots to achieve autonomous emotion understanding and behavior shaping. Our whole study was conducted on the social robot Haru, which has a large library of routines to express different emotions. Our experimental results show that autonomous emotion understanding and behavior shaping of social robots can be achieved through continuous interaction with humans.