Dynamic analysis of multiagent {\it Q}-learning with {\&}epsilon;-greedy exploration