Evolution of Multiobjective Neuromodulated Neurocontrollers for Multi-Robot Systems

Abstract

This thesis focuses on advancing our understanding of the evolution of multiobjective neurocontrollers that have the ability to perform unsupervised learning while operating.We begin with some biologically-inspired modifications to a standard neuroevolution algorithm, that add unsupervised learning and inheritance of said learning for the benefit of offspring generations. We then explore and analyze the relationship between neurocontroller topology and function. Next, a series of modifications to improve the performance of the evolutionary algorithm, and to adapt it for coevolution are presented, along with the results of a series of experiments used to demonstrate their effectiveness.Finally, we present an experiment designed to determine the ability of the complete method to cross the boundary between simulation and reality. Synaptic plasticity has been shown to facilitate unsupervised learning by adapting neural network weights. Neuromodulation is a biologically-inspired technique that can adapt the per-connection learning rates of synaptic plasticity. Multiobjective evolution of neural network topology and weights has been used to design neurocontrollers for autonomous robots. Lamarckian inheritance has been demonstrated with neuroevolution to pass on learned behaviour from parent to offspring generations. Two previous investigations are presented here. Firstly, multiobjective evolution of network weights and topologies (NEAT-MODS) is augmented with neuromodulated iii weight connection of the pre-synaptic neuron and the post-synaptic neuron x x-axis location of robot Cartesian coordinate y y-axis location of robot Cartesian coordinate