著者

海保 諒 森岡 一幸

出版者

一般社団法人 日本機械学会

雑誌

ロボティクス・メカトロニクス講演会講演概要集

巻号頁・発行日

vol.2020, pp.2A1-N10, 2020

---

<p>The paper introduces a method of automatic map generation used for learning of action policy in mobile robot navigation. Road contours and waypoint candidates are extracted as map components by image processing from an image of existing electronic map such as Google Map. Road contours are extracted using topological structural analysis of the binarized electronic map image. Also, waypoint candidates used as respawn or destination points in learning system are randomly selected from pixels of road areas. The generated map can be applied to learning simulator based on deep reinforce learning system. The paper describes an abstract of navigation system based on reinforcement learning, a proposed map generation method.</p>