1A1-P02 蝶の飛行制御におけるANNの汎化能力に関する研究(進化・学習とロボティクス)

概要

論文の詳細を見る
We propose a simple physics model of a butterfly and its flight control by the real-coded genetic algorithm (RCGA) and the artificial neural network (ANN). A physics model consists of two kinetic equations which are led by simplification of the fluid force. A butterfly's flight is controlled by an ANN. The RCGA optimizes weights of the ANN for obtaining the suitable flight. After evolution, the generality of the optimized ANN is investigated by changing an initial height at which a butterfly starts in flight. Then it is shown that the flying creature enough evolved can not keep at the aimed height when the initial height is set far from the initial height used in evolution. Investigation by changing generality of the ANN in evolution shows that too much optimization may reduce the generality of the ANN.
一般社団法人日本機械学会の論文
2011-05-26