Semi-Communicate Social Navigation using Deep Q Networks

The research of mobile robot and pedestrian social interaction aware in human-robot interaction is focused on solving the problem of pedestrian trajectory prediction based on RNN or reinforcement learning. This method has problems such as a low success rate and an uneven path in predicting and avoiding the trajectory of a new environment and an object that is not pre-trained. Because of these problems, it is very difficult to navigate and control existing mobile robots using reinforcement learning. However, many previous reinforcement learning experiments did not consider the precise positioning design of robots and pedestrians to have a mobile robot navigation system with high success rate and safety. In order to alleviate this dilemma, this study aims to improve driving efficiency and learning safety by setting states through precise positioning design of robots and dynamic objects in Deep Q Networks.