Neural Networks for the Moving Objects Control

Abstract

This research describes the methods of neural networks with the aim of generating recommendations for the optimal choice of the learning method for a neural network construction. Two-wheeled Lego robot Mindshtorms nxt 2.0 is considered as an example of a moving object. Back propagation and genetic algorithm methods are considered and compared as methods to train a neural network. A comparative analysis of experimental results was done, which was resulted in the optimal method of back propagation learning algorithm. The optimal network architecture for this result was defined. The results of the training of the neural network can be used in the development of the control systems for moving objects that are currently being used to solve a huge range of tasks. Further results can be improved by selecting the optimal trajectory of the robot, and the use of neuro-fuzzy learning methods, which should provide a more effective training of a neural network.