State Space Segmentation and Motion Generation Based on Dependency among Image Features

This paper presents a method that robot acquires a knowledge of environment by interacting with objects without designer’s knowledge, and moves autonomously its arms to an effective position for manipulating objects such as a pushing based only on acquired knowledge of environment. Specifically, the knowledge is acquired by moving its arm slightly around an object and calculating affordances probabilistically through the behavior of feature points in images of environment. Furthermore, the robot moves its arm widely and do that as well as repeatedly. The proposed method was verified through experiment of state segmentation and motion generation.