A consideration of virtual learned action fitting in real world using person navigation

Abstract

Reinforcement learning is well known with respect to acquisition of action for autonomous robot control. However, it is required much trial for its convergence. In our research, we have proposed the method of reinforcement learning addition of semi-supervised learning. The method yielded that autonomous robot learning in real environment quickly corresponds to environmental changes to goal achievement. The effecitiveness was shown in the results in the simulation. In this paper, we verify the learned action in the simulation to the real environment.