Host: The Japan Society of Mechanical Engineers
Name : [in Japanese]
Date : June 01, 2022 - June 04, 2022
The purpose of this study is to develop a mobile robot navigation system based on switching of action models according to characteristics of environments. Action models according to several environments are trained by deep reinforcement learning. For example, crowded environments such as train stations, open environments such as parks and narrow environments such as office buildings are assumed as environments for action models training. In the preliminary system, one floor of the university building is divided into two environments, and the action models are trained in each environment. The mobile robot travels using the action model corresponding to the current environment in Unity simulation.
