1A2-E08 仮想物理環境におけるバッタモデルの歩行・跳躍動作(進化・学習とロボティクス(2))

Abstract

This paper proposes how to acquire the composite behavior composed of the jumping, landing and walking behavior on artificial creature like a locust under the physical virtual environment. We acquires such behaviors by use of neuroevolution. Neuroevolution is one of the learning algorithm composed of artificial neural network(ANN) and real-coded genetic algorithm (RCGA) and we acquire optimized behavior of the locust model by use of it. In this study, we realize the jumping behavior first. After that, we realize the landing behavior to fix the posture of the locust model. Finally, we realize the walking behavior. We analyze the obtained behavior.