Developmental Approach to Spatial Perception for Imitation Learning: Incremental Demonstrator's View Recovery by Modular Neural Network

摘要

Imitation Learning is not simply one of the most promising ways to accelerate the behavior acquisition for humanoid robots but also one of the most inter-esting cognitive issues to model how we human beings learn to acquire various kinds of behaviors. As the first step towards developmental approach to spatial perception for imitation learning, this paper proposes a method of incremental recovery of the demonstra-tor’s view using a modular neural network by which the learner can organize spatial perception for the view-based imitation learning with the demonstrator in dif-ferent positions and orientations. 1