2A2-D21 強化学習による安定的協調行動の獲得に関する研究

概要

論文の詳細を見る
Acquisition of cooperative behaviors for multiple intelligent robots is very important to develop intelligent artificial systems. For this purpose, many studies and approaches have been proposed in recent years. As a most succesful approach, the reinforcemet learning algorithm, such as Q-learning, is well known. In these studies, main target is getting how to learn a cooperative behaviors. However the robustness of acqired cooperative solutions had not been discussed very much till now. Thus, there are many feasible configurations by which a cooperative task is achieved, but the quality of a cooperative solution is deferent from that of each other. Therefore in this study, we focus on not only a way of getting cooperative solution, but also the robustness of acquired solutions. We try to develop a new reinforcement learning mechanism which solves the complex cooperative tasks and finds its more robust solutions. As an experiment, we treat the cooperative robots systems in which there are multiple autonomous mobile robots, and the seesaw balancing task is given. This problem is an example of cooperative tasks to find the appropriate locations for multiple mobile robots. Each robot agent on a seesaw keeps being balanced state. The experimental result shows the effectiveness of our approaching method.
2009-05-25