Development of simulator for human-robot interaction learning based on programming by demonstration and Q-learning

This paper proposes a simulation system that makes a virtual robot effectively learn human-robot interaction (HRI) in a virtual environment. Given that a virtual robot requires a second entity in order to for learning interactions, the virtual robot interacts with a virtual human. An experiment is performed to teach a virtual human and a virtual robot to verify the possibility of HRI learning using the proposed simulation system. Human behaviors are simulated through a virtual human that has learned through programming by demonstration (PbD). A virtual robot selects the proper behaviors depending on the status of the virtual human and itself by applying Q-learning, a type of reinforcement learning. For the experiment, a house is built in the virtual space of the simulator and a virtual human performs daily activities in the house. Furthermore, a process in which a virtual robot helps a virtual human is simulated. The simulation system proposed in this study can be applied to teach a variety of robots that support human beings through HRI.