GENETIC ALGORITHM BASED QUASI DYNAMIC WALKING OPTIMIZATION FOR SELF MADE HUMANOID ROBOT

Abstract

Nowadays, humanoid robot has developed significantly and has been implemented in various applications. For some applications, such as humanoid robot used to check dangerous place or soccer humanoid robot, the requirement is not only a stable walking. It also needs a precise and long forward displacement to do the job perfectly. Based on that problem, this paper will mainly discuss the application of Genetic Algorithm (GA) to optimize the quasi dynamic walking of a self-made humanoid robot in terms of precision and forward displacement. The self-made humanoid robot is constructed by 10 servos, and several servo brackets as the chassis. Arduino UNO is used as the controller. Then, the first (base) walking gait of humanoid robot is formed by using the forward kinematic method. Later, this base walking gait is modified to generate further five individuals as the initial population to start the GA process. A fitness function is devised with the weight on increasing the forward displacement and the walking precision. The GA process is conducted for four cycles. Parent selection is conducted by using the tournament methods with 25% exchange gene probability during uniform crossover. Uniform crossover rate itself is 100%, and mutation rate is 10%. The result of this final project shows that the GA has successfully increased the fitness function of walking gait from 11.70 cm of the base walking gait to 25.42 cm of the final best walking gait.