Maximal Average-Reward Policies for a Class of Semi-Markov Decision Processes with Arbitrary State and Action Space,
Abstract
The report discusses the problem of maximizing the long-run average (also the long-run average expected) reward per unit time in a Semi-Markov Decision Process with arbitrary state and action space. The main result states that one need only to consider the set of stationary policies in that for each epsilon > 0 there is a stationary policy which is epsilon-optimal. This result is derived under the assumptions that (roughly) expected rewards and expected transition times are uniformly bounded over all states and actions, and that there is a state such that the expected length of time until the system returns to this state is uniformly bounded over all policies. The existence of an optimal stationary policy is established under the additional assumption of countable state and finite action space. Applications to queueing reward systems are given. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 01, 1970
- Accession Number
- AD0714833
Entities
People
- Steven A. Lippman
Organizations
- University of California, Los Angeles