Experimental Results on Q-Learning for General-Sum Stochastic Games