Classifying contact states for recognizing human assembly task

Long development time and efforts are required to complete robot programs. To shorten this programming time and minimize the efforts, we have been developing the system, referred to as "assembly plan from observation (APO) system". The system first observes human actions, understands them and then automatically generates robot programs to mimic these actions. For understanding them, we use transitions of contact states. Since there are infinite contact states in the real world, we need to classify contact states. In this paper, we describe the method to classify contact states obtained from observers.