Epsilon-greedy babbling

摘要

Motor babbling allows an agent sampling trajectory data without a priori knowledge about self-body dynamics. We discuss about the efficiency of motor babbling through the example of drawing task. We propose an exploitation babbling and ϵ-greedy babbling. In order to implement the proposed babblings, we developed dynamics learning tree (DLT). DLT is an online incremental learning algorithm that has constant calculation order O(1). The proposed exploitation babbling and ϵ-greedy babbling improved the rate of effective data at 8 and 7 % from previous babbling respectively. ϵ-greedy babbling converged its prediction error fastest among the three babblings. Using ϵ-greedy babbling, a humanoid robot with wired flexible hand successfully drew a figure without a priori knowledge about the dynamics among self-body, pen, and pen tablet.