A Survey of Asymptotic Value-Iteration for Undiscounted Markovian Decision Processes

摘要

The asymptotic behavior of undiscounted value-iteration in Markov decision problems with finite state and action spaces is reviewed. The asymptotic results concern both the value functions and the sets of optimizing policies as the length of the planning period tends to infinity. The discussion includes Lyapunov functions and other techniques for bounding a solution pair to the optimality equations