Generation of a socially aware behavior of a guide robot using reinforcement learning

This paper proposes a generation of guiding behaviors of a guide robot under a social force framework that is aware of a human social aspect. This framework is supported by a Q-learning algorithm to optimize the social force parameters to deal with a variety of stimulations. It implies that we let our robot learn by itself by interacting with the environments directly. We named this framework as Q-Learning based Social Force Guiding Model (QL-SFGM). However, let the real robot learn in the real environments under Q-learning framework is difficult, time-consuming, and hazardous. Therefore, in this study, we utilize a realistic simulator, V-Rep, for both training and testing. The simulation results show that our proposed framework is effective to reduce over-reactive behavior of our guide robot so that smoothness, safety, and comfort can be achieved.