Multi-agent Learning in Coverage Control Games

摘要

Multi-agent systems have found a variety of industrial applications from economics to robotics. With the increasing complexity of multi-agent systems, multi-agent control has become a challenging problem in many areas. While studying multi-agent systems is not identical to studying game theory, there is no doubt that game theory can be a key tool to manage such complex systems. Game theoretic multi-agent learning is one of relatively new solutions to the complex problem of multi-agent control. In such learning scheme, each agent eventually discovers a solution on his own. The main focus of this thesis is on enhancement of multi-agent learning in game theory and its application in multi-robot control. Each algorithm proposed in this thesis, relaxes and imposes different assumptions to fit a class of multi-robot learning problems. Numerical experiments are also conducted to verify each algorithm's robustness and performance.