1C1-4 進化型人工神経回路網を用いたマルチロボットシステムの協調行動獲得に関する一考察(1C1 強化学習)

概要

論文の詳細を見る
Evolutionary robotics is a promising way of developing the controllers of autonomous mobile robots. A series of computer simulations are conducted with evolving artificial neural network (EANN) approach to building the controllers for a multi-robot system. However, in general, it is not so easy to generate meaningful collective behavior in a multi-robot system when a standard GA and popular layered-type neural networks are adopted. Therefore, in our experiments, not only weights but also topology are evolved simultaneously by using MBEANN, which is the novel method we are developing. In this paper, we are interested in analyzing the behavior shown in the system, from the viewpoint of the influence of cumulative penalties in the fitness function toward the evolutionary dynamics. We have found from the results that the cumulative penalties make the problem difficult to achieve appropriate collective behavior.
社団法人日本機械学会の論文
2006-09-25