Multi-agent systems: Integrating reinforcement learning, bidding and genetic algorithms

Multi-agent systems are research areas of the distributed artificial intelligence. Multi-agent systems have the advantages of parallelism, robustness, and scalability. Agents in multi-agent systems are not pre-arranged to help each other with all resources and capabilities that they possess. They may coordinate their activities with others to achieve their own local goals.; This dissertation presents a multi-agent reinforcement learning bidding approach (MARLBS) for performing multi-agent reinforcement learning. MARLBS integrates reinforcement learning, bidding and genetic algorithms. The general idea of our multi-agent systems is as follows: There are a number of individual members in a team. Each member has two modules: Q module and CQ module. Each member can select actions to be performed at each step, which is done by the Q module. While CQ module determines at each step whether the agent should continue or relinquish control. Once an agent relinquishes its control, the new agent is selected by the bidding algorithms. After a number of training episodes, we apply genetic algorithm to these teams to facilitate learning.; The members in this system interact and cooperate with each other through bidding as well as individual reinforcement learning. A member calls upon another member when such an action leads to higher reinforcement.; We implement the GA-based reinforcement learning bidding system (GMARLBS) and the GP-based reinforcement learning bidding system to two complex problems: the Backgammon game and the TSP problem. The experiment results show MARLBS can achieve a certain level of performance in game-playing and ordering problems while the system uses zero built-in knowledge.