2P1-G10 DCOB: Constructing Action Space from Basis Functions for Motion Acquisition of Robots by Reinforcement Learning : Application to Crawling Task of Actual Small Size Multi-link Robot

Abstract

This paper investigates an applicability of the discrete action set DCOB proposed by Yamaguchi et al. to motion acquisition task of an actual small size multi-link robot by reinforcement learning. We apply the DCOB to a crawling task of a spider robot (ROBOTIS Bioloid), and compare the DCOB with a grid action set as a conventional method. The experimental results demonstrate the advantage of the DCOB both in learning speed and ability to acquire performance.