画像分類に基づく走行環境識別に応じた行動モデル選択による移動ロボットの自律走行システム

抄録

This study aims to develop an autonomous mobile robot navigation system based on deep reinforcement learning in various environments. Especially, the mobile robot navigation system based on switching to the appropriate action model to the running environment is proposed. The action model selection is performed depending on the environment classification with camera images. Simulation environments including various environments, such as a hallway-like indoor environment, an urban outdoor environment, and a large outdoor environment like a park are made with Unity. On each environment, an action model that the mobile robot can reach the goal while avoiding the obstacles is trained respectively. Furthermore, the proposed system is tested on the experimental environments mixed with avobe three environments in the simulator.