一人称視点画像を用いたDeep Q-Networkによる群行動の創発

抄録

Many animals including birds and fish form different type of swarm and many studies have tried to create swarm behaviors which is like real swarm on computer simulation. The most famous model is Boids proposed by Reynolds. In Boids, individuals just follow three local rules. The model can show the realistic and dynamic swarm behaviors, but it is impossible that all individuals can sense the distances to the others precisely in a real world. The most individual must depend on their vision information. A purpose of this study is to build autonomous agents that behave based on visual images and to create swarm behavior. We use Deep Network for agent's decision making and create swarm behavior. Our results show that from visual image, agents can create swarm behavior.