Skill acquisition from human demonstration using FCM clustering of qualitative contact states

The reliance on precision positioning of current robots creates problems in both controlling and programming. The motivation behind this work is to transfer force-based skills to robots and to realize programming by demonstration for tasks in contact states. For this purpose, we model the skill as qualitative contact states and desired transitions between them. Element Contact Formation (ECF) is proposed to describe contact state. For different ECF has different force clustering, the modified FCM clustering is used for automatic classification of ECF in demonstration based forces information. Then from demonstration data the sequence of contact states and desired distances and velocities for transitions are obtained. The ECF identifier can be built with the similar degree between the force information and clustering. The skill acquisition module is implemented in robot controller with open architecture. Petri net is used for merging different sequences and simplifying the skill.