環境モデルにおける物体の見え方と見方

抄録

This paper presents a scheme of environment understanding for intelligent robots. The purpose of this scheme is to enable intelligent robots to make appropriate plans for their actions in unknown environments. To realize this scheme, we extend the environment models by introducing concepts of "visibility" and" view". By "visibility", we mean scene image available for intelligent robots as input or predicted by model-based vision technique. The concept "view" is key concern of this paper. "View" is positive action toward environment by robots in order to make the incomplete model more correct. The overall model-management is based on non-monotonic logic inference system, which maintains incomplete data and carrys out necessary model-modification. The paper describes advantages of the establishment of this scheme, which frees robot systems from too much dependence on image understanding.