深層強化学習による単眼カメラ画像を入力とした移動ロボットの自律走行モデルの獲得

抄録

This paper introduces deep reinforcement learning to obtain action models for autonomous mobile robots in the simulation environment. The proposed robot system uses monocular camera images mounted on the mobile robot instead of 2D-LiDAR. Unity to create simulation environments and agents was adopted in this study. The training in Unity provides action models that the agents can reach the destinations in the environments using monocular camera images input. In addition, the trained models acquired in the simulation environment were adopted to a ROS-based navigation system in Unity.