Development of a Mobile Robot Following a Specific Person and Arriving at Destinations Based on Deep Reinforcement Learning

Abstract

This study aims to develop a mobile robot to follow a specific human while avoiding obstacles and reaching to destinations. The behavior models for robot navigation are obtained by deep reinforcement learning in simulated environments representing simplified real-world conditions. The mobile robot in this study assumes to equip an RGB-D camera as an external sensor. The trained behavior model is then implemented in the actual robot for real-world navigation. Detecting and tracking of a specific target person is performed by combination of YOLO and SORT algorithms. Additionally, the RTAB-Map, which enables online simultaneous localization and mapping, allows the relative navigation from the current position without prior mapping.