Active Exploration in Building Hierarchical Neural Networks for Robotics

During early robot learning, several mappings need to be set up for sensorimotor coordinations and transformation of sensory information from one modality to another. Usually these mappings are nonlinear and traditional passive learning approaches can not deal with these problems well. In this paper, a hierarchical clustering technique is introduced to group large mapping error locations and these error clusters drive the system to actively explore details of these clusters. Higher level local growing radial basis function subnetworks are used to approximate the mapping residual errors from previous mapping levels. Plastic radial basis function networks construct the substrate of the learning system and a simplified node-decoupled extended Kalman filter algorithm is presented to train these radial basis function networks. Experimental results are given to compare the performance between active learning and passive learning