Approximate State Abstraction for Markov Games