A Method for Co-Evolving Morphology and Walking Pattern of Biped Humanoid Robot

In this paper, we present a method for co-evolving structures and controller of biped walking robots. Currently, biped walking humanoid robots are designed manually on trial-and-error basis. Although certain control theory exists, such as zero moment point (ZMP) compensation, these theories assume humanoid robot morphology is given in advance. Thus, engineers have to design control program for apriori designed morphology. If morphology and locomotion are considered simultaneously, we do not have to spare time with trial-and-error. Therefore a method useful for designing the robot is proposed . At first, the simple models of both morphology and controller are used for the dynamic simulation, which are multi-link model as morphology and two kinds of controllers. One is a layered neural network and the other is neural oscillator. The robots with the optimal energy efficiency of walking are designed with Genetic Algorithm. As a result, various combinations of morphologies and gaits are generated, and obtained relationship between length of each link and moving distance which gives the optimal energy efficiency. Moreover, the robots are encoded from limited size of chromosomes. Keywords—Biped Walking, Genetic Algorithm, Neural Network, Oscillator.