Development of a Web-Based Education System for Deep Reinforcement Learning-Based Autonomous Mobile Robot Navigation in Real World

摘要

The technology that combined deep reinforcement learning and robotics is increasing interests in recent years. Although several online tools for studying this technology can be found, it is difficult for beginners to develop actual robot systems for autonomous navigation in the real world. In this study, we developed a web-based educational system that is able to help users to study mobile robot navigation based on deep reinforcement learning and develop actual robot systems. The proposed web system provides the following functions: setting the parameters of reinforcement learning for autonomous robot navigation, running learning scripts and monitoring status of the learning. The first experiment that a user develops an actual robot system was performed. In the experiment, the user tuned parameters on the web page started the training and obtained action policy models. The experimental results indicate the proposed system can be applied to develop an actual autonomous navigation system. Also, the user could decide better parameters through the trial and error process using the proposed system.