Neuroevolution of Cooperation in a Multi-Agent Foraging Task

Teams of Artificially Intelligent agents can more effectively achieve their goals by learning to cooperate. Methods for learning cooperation can be centralised or decentralised but the trade-offs of applying these methods are not well understood. This thesis conducts an in-depth analysis of different learning methods. It introduces a new multi-agent task that facilitates the study of complex cooperation and implements it as part of a reusable experimental platform. It uses the platform to conduct a systematic comparison of centralised and decentralised learning, providing new insights into their trade-offs, and presents a theoretical model that generalises the findings to new contexts.