Cooperative Q Learning Based on Blackboard Architecture

摘要

This paper presents cooperative Q learning based on the blackboard architecture. The learning process is executed at the blackboard architecture making use of all the robots in the training scenario to explore the learning space and collect experiences. Three kinds of Q value update strategies are proposed in order to deal with variation of Q value that may happen in the multi-robot learning scenario. Simulation experiments verify that the learning process converges faster and the control strategy obtained has better performance.