An exact iterative search algorithm for constrained Markov decision processes

作者：Chang Hyeong Soo^*

来源：Automatica, 2014, 50(5): 1531-1534.

DOI：10.1016/j.automatica.2014.03.020

• Markov decision processes
• Policy iteration
• Dynamic programming
• Constrained optimization

This communique provides an exact iterative search algorithm for the NP-hard problem of obtaining an optimal feasible stationary Markovian pure policy that achieves the maximum value averaged over an initial state distribution in finite constrained Markov decision processes. It is based on a novel characterization of the entire feasible policy space and takes the spirit of policy iteration (PI) in that a sequence of monotonically improving feasible policies is generated and converges to an optimal policy in iterations of the size of the policy space at the worst case. Unlike PI an unconstrained MDP needs to be solved at iterations involved with feasible policies and the current best policy improves all feasible policies included in the union of the policy spaces associated with the unconstrained MDPs.

• 出版日期2014-5