A Case Study in Hybrid Multi-threading and Hierarchical Reinforcement Learning Approach for Cooperative Multi-agent Systems

This paper describes a case study about a multi-agent system for cooperative tasks, i.e. a mixing color task given three different sources of color. A reinforcement learning approach was performed by the agents, however, this type of learning exploits exponentially when the number of states in the environment is very large. In that sense, the paper proposes to use the MaxQ-Q hierarchical reinforcement learning algorithm to obtain a suitable policy for agents in order to minimize the time process to achieve the goal, and to reduce the state space. In addition, since the multi-agent system runs in a software application, a multi-threading paradigm was proposed to use. Experimental results show that this multi-agent system can reduce the time process and still maintain independence of agents.