視覚運動統合学習に基づく複数対象物群の混和マニピュレーション

抄録

This paper presents a manipulation method for interacting swarm of objects based on visual-motor integration learning. Here, a mixing manipulation problem is treated, in which a single probe mixes the swarm of objects in a closed environment. First, from images of the swarm of objects, an index of mixing level is obtained employing the gray level cooccurrence matrix method. Then, the probe trajectory to increase the mixing level as high as possible in a time limit is obtained by trial and error using reinforcement learning. The proposed method is validated in numerical experiments. The proposed method has a potential to recognize physical features of objects through their interacting behaviors, and to optimize the probe trajectories based on such features.