How iCub Learns to Imitate Use of a Tool Quickly by Recycling the Past Knowledge Learnt During Drawing

Ajaz Ahmad Bhat, Vishwanathan Mohan

Abstract: Using a skill learning architecture being developed for the humanoid iCub, in this article we show through experiments how a cognitive robot can learn to imitate quickly the use of a tool after a teacher’s demonstration by recycling previously learnt knowledge from drawing experiments. The employed architecture incorporates novel principles for constructing a growing motor vocabulary in cognitive robots enabling “cumulative” and “swift” learning of skills through integration of multiple streams of learning like imitation and mental simulation. A central notion emphasized in the architecture is that movements can be represented in the form of ‘shapes’ instead of trajectories per se in order to liberate them from task specific details. The idea is to abstract out critical features/knowledge in the movement trajectory, which can later be reused in a context-independent manner.