Multiagent AGVs dispatching system using multilevel decisions method

An AGV dispatching system needs dynamic and distributed dispatching policies in FMS. This research addresses a multilevel decision and cooperative learning method to solve this problem. Each AGV is treated as a rational agent, which has two level decisions. On the option level an agent will make decisions to execute a sub-task with the best response to the other AGVs current option. On the action level, an agent will learn an optimal policy of actions for achieving his planned option. To implement the multi-level decision, we use Markov games and the reinforcement learning (RL) algorithm on the option level and a memory based algorithm RL on the action level. We apply our method to an AGV dispatching simulation and show the results.