We extend the potential-based shapingmethod fromMarkov decision processes to multi-player general-sum stochastic games. We prove that the Nash equilibria in a stochastic game remains unchanged after potential-based shaping is applied to the environment. The property of policy invariance provides a possible way of speeding convergence when learning to play a stochastic game.

Journal of Artificial Intelligence Research
Department of Systems and Computer Engineering

Lu, X. (Xiaosong), Schwartz, H.M, & Givigi Jr., S.N. (Sidney N.). (2011). Research note: Policy invariance under reward transformations for general-sum stochastic games. Journal of Artificial Intelligence Research, 41, 397–406. doi:10.1613/jair.3384