[2510.14035] GammaZero: Learning To Guide POMDP Belief Space Search With Graph Representations

Computer Science > Artificial Intelligence arXiv:2510.14035 (cs) [Submitted on 15 Oct 2025 (v1), last revised 29 Mar 2026 (this version, v2)] Title:GammaZero: Learning To Guide POMDP Belief Space Search With Graph Representations Authors:Rajesh Mangannavar, Prasad Tadepalli View a PDF of the paper titled GammaZero: Learning To Guide POMDP Belief Space Search With Graph Representations, by Rajesh Mangannavar and 1 other authors View PDF HTML (experimental) Abstract:We introduce an uncertainty-aware graph representation framework for learning to guide planning in Partially Observable Markov Decision Processes (POMDPs). Unlike existing approaches that require domain or problem size specific neural architectures, GammaZero leverages a unified graph-based belief representation that enables generalization across problem sizes within a domain. Our key insight is that belief states can be systematically transformed into uncertainty-aware graphs where structural patterns learned on small problems transfer to larger instances. We employ a graph neural network with a decoder architecture to learn value functions and policies from expert demonstrations on computationally tractable problems, then apply these learned heuristics to guide Monte Carlo tree search on larger problems. Experimental results on standard POMDP benchmarks demonstrate that GammaZero achieves comparable performance to BetaZero when trained and tested on the same-sized problems, while enabling zero-shot generalizati...