In this study, we propose an autonomous navigation system for a snake-like robot with the aim of efficiently searching for victims during disasters. This system was developed using PPO (Proximal Policy Optimization), a type of deep reinforcement learning algorithm. First, we constructed an environment in the simulator with a destination and obstacles, and placed a 3D model of the snake-like robot within it. Then, by repeatedly running the snake-like robot from the starting point to the destination, we confirmed that it acquired the ability to autonomously reach the destination while avoiding obstacles.
Kohayakawa et al. (Thu,) studied this question.