視覚と行動のend-to-end学習により経路追従行動をオンラインで模倣する手法の提案

―経路選択機能の追加―

抄録

We have proposed an online imitation method for path-following behavior based on end-to-end learning of vision and action. However, the proposed method aims to follow a fixed path and cannot dynamically select a path to make a robot move to a destination. In this study, we add a function to select a path to the method so that the robot can move to an arbitrary destination. We introduce the online imitation learning method with the additional function of selecting a path, and then verify whether the system can select a path by experiments using a simulator.