A robot agent that learns group interaction through a team-based virtual reality game using affective reward reinforcement learning

Abstract

In the near future, robots are expected to be integrated into people’s lives, interacting with them. To develop better robotics and artificial intelligence, this research focuses on the concept of teamwork. A robot agent was implemented in a virtual reality(VR) game to play the sport roundnet, a team-based sport similar to table tennis and volleyball [2]. The agent is trained with reinforcement learning with EDA skin sensor data [6] of players. The system is evaluated using a questionnaire on the player’s feeling during the experiment and compared with agents not trained with affective data. The system is implemented in Unity3D’s ML-Agents Toolkit.