価値関数を用いた非線形手動制御系の学習過程に関する考察

概要

論文の詳細を見る
This paper describes analysis on human operator's trial and error learning process to control a nonholonomic system. A novel analysis technique using the value function of reinforcement learning is proposed. According to the transition of the value function, human operators tend to explore an objective trajectory first, and then shift to the following control of the trajectory and accelerate it. This acceleration disturbs the objective trajectory, and inducts another exploration phase to converge on the better solution. These results of the stepped approach of human learning may inspire an improvement to reinforcement learning.原著Original Paper
東北大学の論文