Interconnection structure optimization for neural oscillator based biped robot locomotion

One of the problems in neural oscillator based humanoid locomotion is the interconnection structure and its weights. They influence the locomotion performance. This paper proposes an evolutionary algorithm for determining the interconnection structure in humanoid robot locomotion based on neural oscillator. The aim of this paper is to form the interconnection structure of motor neurons in order to produce the locomotion pattern in humanoid biped robot. The evolutionary system forms the connection and determines the synapse weight values of the 12 motor neurons distributed to 6 joint angles (two hip-x joints, two hip-y joints, two knee joints). One chromosome has 53 genes, where 50 genes represent the weight values between motor neurons and 3 genes represent the gain parameters in hip-y, hip-x, and knee joint. Center of gravity and speed walking analysis are required for fitness evaluation. In order to prove the effectiveness of the system model, we realized it in a computer simulation. The experimental result shows the comparison result with our previous model. The stabilization level and speed resulted by using this system are increased.