P. Goldberg

This work reduces state-adversarial Markov decision processes to a strategically equivalent constrained zero-sum one-sided partially observable stochastic game and presents the first algorithmic route to computing $\epsilon$-approximations of initial-state dependent equilibria.