In human activity space, there are obstacles with various sizes and configurations. The purpose of this research is generation of the motion for walking robot in response to the obstacle environment. In this research, generation of the motion for walking robot carry out by the optimization of stepping points and the generation of toe trajectory between every stepping point for the toes. We make the neural networks to generate the stepping points in response to the obstacle recognition. Training data for the neural networks are optimum stepping points based on mechanism of robot model generated by genetic algorithm. The sets of the representative points of obstacle is input data and the sets of stepping points is out put for the trained neural networks. In the case of using the unused training data for input data, the neural network output the optimum stepping points successfully.