Abstract
The purpose of this research is the planning of walking trajectory for quadruped robot in response to obstacle environment. If quadruped robot adjusts stepping points to step over obstacles, the robot can avoid obstacles more smoothly. So we optimize stepping points for walking in obstacle environment like an animal. Generation problem of stepping points is reduced to a combinatorial optimization problem solved by using genetic algorithm. Learning data for neural network are information of obstacle environments and optimum stepping points. Optimum stepping points for learning are generated by genetic algorithm to variety of obstacle environments. Obstacle environments in this report have different configurations and different lengths. After learning optimum stepping points and obstacle environments, unlearned data are input to the neural network. The output data to unlearned input data is the valid stepping points. The proposed method of walking trajectory planning can generate stepping points to unlearned environments.
