Optimum Architecture of Neural Networks lane following system

Recently, neural networks have demonstrated their ability to achieve excellent performance for the control of mobile robots. In fact, the recourse of this control method by learning has become a necessity because control systems obtain then, proceed by collecting empirical data, storing and removing the knowledge contained in it and using this knowledge to respond to new situations. However, the problem of choosing an optimal number of hidden layers as well as choosing neurons per layer is very critical for these networks. So here we propose to determine the settings for the optimum architecture of neural network. In the course of our experiments, we have shown that the error of learning as well as the one of the validation provides a satisfactory criterion for the optimization of network architecture.