Skill representation and acquisition

摘要

This paper presents a new approach for representing and acquiring skill by a robot based on experimental data. Skill is represented by describing a feasible state transition region embedded in experimental data as an union of hyper-ellipsoidal subregions of various sizes and shapes. Multi-resolution radial basis competitive and cooperative network (MRCCN) is formulated for self-organizing hyper-ellipsoidal subregions and for providing accurate forward and backward state transitions through interpolation. Skill acquisition is performed by finding an optimal path from the current to the goal state in the feasible state transition region. The search for an optimal path is based on bidirectional dynamic path planning algorithm proposed in this paper.