Improving Multi-Agent Cooperation Using Directed Exploration

In this work, we are addressing the problem of fully cooperative multi-agent system (MASs) with the same common goal for all agents. Coordination question is the main focus in such systems: how to ensure that the agents' own decisions contribute to the group's jointly optimal decisions? To solve this, a new multi-agent reinforcement learning algorithm, named TM LRVS Qlearning, is introduced and tested. The usefulness of this new method is shown using a simulated hunting game.