Title: Navigating Uncertainty in Planning: Leveraging Symmetries, Inferring Policies, and Compressing Solutions

Abstract: Fully-observable non-deterministic (FOND) planning serves as a foundational element in artificial intelligence planning under uncertainty, characterizing unpredictable outcomes through actions with non-deterministic effects. This study introduces a suite of methods that position explicit best-first policy-space search as a competitive approach for tackling FOND planning challenges. We investigate the definition of equivalence relations among policies to facilitate the pruning of the search space. Furthermore, we demonstrate the application of group theory techniques to efficiently calculate canonical symmetries between states. Beyond policy-space search, we offer two additional contributions: a procedure capable of inferring a solution policy function in polynomial time from its domain specification alone, and an integer-programming formulation that, when provided with a solution policy defined over complete states, generates a set of resource-efficient models. These models are designed to identify a partial-state policy that unambiguously represents the original policy while utilizing the minimum number of partial states.

Source: arXiv Generated at: 2026-06-03 00:00:00 UTC