Thesis for the degree of Master of Science in Complex Adaptive Systems Particle swarm optimization of artificial neural networks for autonomous robots

Artificial neural networks (ANNs), especially when they have feedback connections, are potentially able to produce complex dynamics, and therefore have received attention in control applications. Although ANNs are powerful, designing a network can be a difficult task, and the more complex desired dynamics are, the more difficult the design of the network becomes. Many researchers have sought to automate ANN design process using computer programs. The problem of finding the best parameter set for a network to solve a problem can be seen as a search and optimization problem. This thesis concerns a comparison of two widely used stochastic algorithms, genetic algorithms (GAs) and particle swarm optimization (PSO), applied to the problem of optimizing parameters of ANNs for a simulated autonomous robot. For this purpose, a mobile robot simulator has been developed. The neural network-based controller of the robot is optimized using GAs and PSO in order to enable the robot to accomplish complex tasks. The results show that both algorithms are able to optimize the network and solve the problem. PSO excels in smaller networks, while GAs perform better for larger networks.