Iterative Depth-First Search for FOND Planning

Abstract

Fully Observable Non-Deterministic (FOND) planning models uncertainty through actions with non-deterministic effects. Existing FOND planning algorithms are effective and employ a wide range of techniques. However, most of the existing algorithms are not robust for dealing with both non-determinism and task size. In this paper, we develop a novel iterative depth-first search algorithm that solves FOND planning tasks and produces strong cyclic policies. Our algorithm is explicitly designed for FOND planning, addressing more directly the non-deterministic aspect of FOND planning, and it also exploits the benefits of heuristic functions to make the algorithm more effective during the iterative searching process. We compare our proposed algorithm to well-known FOND planners, and show that it has robust performance over several distinct types of FOND domains considering different metrics.

PRP Rebooted: Advancing the State of the Art in FOND Planning

“…The code, benchmarks, and detailed analysis can be found at mulab.ai/pr2 . We compare against the state of the art in FOND planning: MyND (Mattmüller et al 2010), FONDSAT (Geffner and Geffner 2018), PRP (Muise, McIlraith, and Beck 2012), and Paladinus (Pereira et al 2022). We configured each planner to its best settings based on aggregate performance across all domains, including using a modern SAT solver for FOND-SAT (improving its coverage by a fair margin).…”

“…FOND planners based on weak planning procedures carry very little of the FOND setting to the classical planner subcalls (Pereira et al 2022). PR2 addresses this issue in several ways.…”

“…The final category of FOND solvers is the policy-space search of MyND and Paladinus (Mattmüller et al 2010;Pereira et al 2022). By many measures, Paladinus represents the state of the art for this class of solution strategy.…”

“…This was accomplished by representing families of states as conjunctive formulae and employing regression rewriting, a form of pre-image computation, to establish relevance (Waldinger 1977;Reiter 2001;Fritz and McIlraith 2007). Subsequent techniques explored new concepts for computing solutions to FOND problems: from the backwards search of GRENDEL (Ramírez and Sardiña 2014), to the SAT encoding of FONDSAT (Geffner and Geffner 2018), to the policy-based search of MyND and Paladinus (Mattmüller et al 2010;Pereira et al 2022). These new planners were often published with additional benchmark problems that showcased the new planner's potential and, in a number of cases, their superior performance compared to the incumbent, PRP.…”

Non-deterministic Planning for Hyperproperty Verification

“…Non-deterministic planning provides a powerful intermediate language that encompasses problems such as reactive synthesis (Camacho et al 2018), controller synthesis in MDPs, epistemic planning (Engesser and Miller 2020), and generalized planning (Hu and Giacomo 2011). Consequently, many methods and tools have been developed (Pereira et al 2022;Messa and Pereira 2023;Camacho et al 2017;Geffner and Geffner 2018;Rodriguez et al 2021;Muise, McIlraith, and Beck 2012;Kuter et al 2008), with some also supporting planning in partially observable domains (Bertoli et al 2006;Cimatti et al 2003;Bonet and Geffner 2011).…”