Solving Multi-Model MDPs by Coordinate Ascent and Dynamic Programming

TL;DR

This paper introduces CADP, a novel algorithm combining coordinate ascent and dynamic programming to efficiently solve multi-model MDPs, improving policy robustness and outperforming existing methods.

Contribution

The paper presents CADP, a new approach that guarantees monotone policy improvements for multi-model MDPs using coordinate ascent and dynamic programming.

Findings

CADP outperforms existing algorithms on benchmark problems.

CADP guarantees monotone policy improvements.

Theoretical analysis shows CADP is at least as good as previous methods.

Abstract

Multi-model Markov decision process (MMDP) is a promising framework for computing policies that are robust to parameter uncertainty in MDPs. MMDPs aim to find a policy that maximizes the expected return over a distribution of MDP models. Because MMDPs are NP-hard to solve, most methods resort to approximations. In this paper, we derive the policy gradient of MMDPs and propose CADP, which combines a coordinate ascent method and a dynamic programming algorithm for solving MMDPs. The main innovation of CADP compared with earlier algorithms is to take the coordinate ascent perspective to adjust model weights iteratively to guarantee monotone policy improvements to a local maximum. A theoretical analysis of CADP proves that it never performs worse than previous dynamic programming algorithms like WSU. Our numerical results indicate that CADP substantially outperforms existing methods on several benchmark problems.