Design of central pattern generator for humanoid robot walking based on multi-objective GA

Recently, the field of humanoid robotics attracts more and more interest and the research on humanoid locomotion based on central pattern generators (CPG) reveals many challenging aspects. This paper describes the design of CPG for stable humanoid bipedal locomotion using an evolutionary approach. In this research, each joint of the humanoid is driven by a neuron that consists of two coupled neural oscillators, and corresponding joint´s neurons are connected by strength weight. To achieve natural and robust walking pattern, an evolutionary-based multi-objective optimization algorithm is used to solve the weight optimization problem. The fitness functions are formulated based on zero moment point (ZMP), global attitude of the robot and the walking speed. In the algorithms, real value coding and tournament selection are applied, the crossover and mutation operators are chosen as heuristic crossover and boundary mutation respectively. Following evolving, the robot is able to walking in the given environment and a simulation shows the result