A Social Human-Robot Interaction Simulator for Reinforcement Learning Systems

Abstract

Social robotics represents a branch of human-robot interaction that aims to develop robots to operate in unstructured environments in direct partnership with human beings. Social robots must interact with human beings by understanding social signals and responding appropriately to promote a natural and socially acceptable interaction among humans and robots. In this article, we propose a simulator for Deep Reinforcement Learning and Social Robotics, Sim-DRLSR, aiming to provide a system development tool for human-robot interaction with a self-learning paradigm. The simulator SimDRLSR is capable of providing an environment for social robots to learn and to identify, through vision, human interactive behaviors and to act accordingly to them. We use the Multimodal Deep Reinforcement Learning (MDQN) architecture for training and validating the simulated robot. Preliminary experiments show the proposed simulator can assist in testing and developing phases of social robots for interactions using vision, saving the use of real robots in the early stages of projects.