The Proceedings of JSME annual Conference on Robotics and Mechatronics (Robomec)
Online ISSN : 2424-3124
2021
Session ID : 1P1-L10
Conference information

Mobile Robot Navigation System Based on Adaptive Selection of Action Models Acquired by Deep Reinforcement Learning in Unknown Real World
*Ren YUASAKoki YOKOYAMAKazuyuki MORIOKA
Author information
CONFERENCE PROCEEDINGS RESTRICTED ACCESS

Details
Abstract

The purpose of study is autonomous mobile robot navigation in any unknown real environments. A proposed system is based on switching of action models according to situations around the mobile robot. Action models for predefined basic shapes are trained in advance by deep reinforcement learning. In addition, mobile robot selects suitable action model based on shapes around the robot. Autonomous robot navigation experiments were performed in two real environments that include the place mainly consisted of basic shapes and the other place including complicated shape. The results of the experiments are shown in this paper.

Content from these authors
© 2021 The Japan Society of Mechanical Engineers
Previous article Next article
feedback
Top