深層強化学習を用いた抑草ロボットの作業経路計画に関する研究

抄録

This paper describes a work path planning using deep reinforcement learning study to derive the optimal work path for a paddy weed suppression robot with an automatic driving function to efficiently perform weed suppression work. In addition to the paths obtained by learning, simulations were performed using the robot's equations of motion to evaluate the paths for a simple path that repeats reciprocation in the vertical and horizontal directions. As a result, we were able to derive the work path using deep reinforcement learning. The path of the proposed method was characterized by a spiral shape. Therefore, it was found to be superior in terrain that can be divided into several rectangles. Compared with the conventional path, the tracking accuracy of the robot on the spiral path was improved.