Abstract
The purpose of this research is generation of stepping points for quadrupedal walking robot in obstacle environment. It is difficult to prepare the optimum stepping points for various configurations and shapes of obstacle in advance. We make the neural networks to learn the sets of obstacle environment and its optimum stepping points. We use optimum stepping points generated by genetic algorithm as supervised data. The available stepping points for the obstacle environment by the learned neural network. Stepping points for the same obstacle environment are generated by three different methods. The methods are genetic algorithm, neural network and random transformation based on optimum stepping points. From the comparing fitness of three cases it is fined that the stepping points generated by the neural network is close to the optimum stepping points.
