Motor development of an pneumatic musculoskeletal infant robot

This paper aims to clarify the mechanism of an infant´s locomotive development from the viewpoint of cognitive developmental robotics. We built up an infant-sized musculoskeletal robot driven by McKibben pneumatic actuators, which enable the robot to interact with its environment without any problems of mechanical damage and excessive heat in long-term experiments. We applied a learning algorithm based on central pattern generators and optimization method as a way for the robot to acquire its crawling motion. As a result of a developmental experiment, an efficient forward motion was acquired with the proposed method. We discuss how its musculoskeltal body, spinal structure and tonus of the artificial muscles affect its development of crawling behavior.