Best-Effort Policies for Robust Markov Decision Processes
This work addresses a tie-breaking issue in robust decision-making for researchers and practitioners in reinforcement learning, offering an incremental improvement over existing robust MDP methods.
The paper tackles the problem of selecting among multiple optimal robust policies in robust Markov decision processes (RMDPs) by proposing a refined criterion that maximizes worst-case expected return while also achieving maximal expected return under non-adversarial transition probabilities, resulting in an algorithm with manageable overhead compared to standard robust value iteration.
We study the common generalization of Markov decision processes (MDPs) with sets of transition probabilities, known as robust MDPs (RMDPs). A standard goal in RMDPs is to compute a policy that maximizes the expected return under an adversarial choice of the transition probabilities. If the uncertainty in the probabilities is independent between the states, known as s-rectangularity, such optimal robust policies can be computed efficiently using robust value iteration. However, there might still be multiple optimal robust policies, which, while equivalent with respect to the worst-case, reflect different expected returns under non-adversarial choices of the transition probabilities. Hence, we propose a refined policy selection criterion for RMDPs, drawing inspiration from the notions of dominance and best-effort in game theory. Instead of seeking a policy that only maximizes the worst-case expected return, we additionally require the policy to achieve a maximal expected return under different (i.e., not fully adversarial) transition probabilities. We call such a policy an optimal robust best-effort (ORBE) policy. We prove that ORBE policies always exist, characterize their structure, and present an algorithm to compute them with a manageable overhead compared to standard robust value iteration. ORBE policies offer a principled tie-breaker among optimal robust policies. Numerical experiments show the feasibility of our approach.