Finding sub-optimal policies faster in multi-agent systems

Abstract

Recently, delayed reinforcement learning (RL) has been proposed as a strong method for learning in multi-agent systems (MASs). In this method, agents are concerned with the problem of discovering an optimal policy, a function mapping states to actions. The most popular RL technique, Q-learning, has been proven to produce an optimal policy under certain conditions. In this paper, we present a new method called FQ-learning for more quickly learning agents acting in MAS. In experimental results done on the pursuit domain, we demonstrate the superiority of the proposed algorithm over standard Q-learning method in terms of convergence speed and number of convergence steps.