Autonomous Navigation of Rescue Robot on International Standard Rough Terrain by Using Deep Reinforcement Learning

Rescue robots perform rescue and search operations at disaster sites. These robots should be able to navigate autonomously because remote control is difficult. The objective of this research was to enable rescue robots to navigate autonomously on international standard rough terrain. To achieve this objective, we built a learning environment in a simulator using Unity, a physics engine, and conducted deep reinforcement learning using the machine learning framework of Unity and ML-Agents. A comparative verification with remote control demonstrated that autonomous navigation was superior to remote control in terms of both time and success rates because of the difference in the motion of the robot.