Lilotane: A Lifted SAT-based Approach to Hierarchical Planning

Abstract

One of the oldest and most popular approaches to automated planning is to encode the problem at hand into a propositional formula and use a Satisfiability (SAT) solver to find a solution. In all established SAT-based approaches for Hierarchical Task Network (HTN) planning, grounding the problem is necessary and oftentimes introduces a combinatorial blowup in terms of the number of actions and reductions to encode. Our contribution named Lilotane (Lifted Logic for Task Networks) eliminates this issue for Totally Ordered HTN planning by directly encoding the lifted representation of the problem at hand. We lazily instantiate the problem hierarchy layer by layer and use a novel SAT encoding which allows us to defer decisions regarding method arguments to the stage of SAT solving. We show the correctness of our encoding and compare it to the best performing prior SAT encoding in a worst-case analysis. Empirical evaluations confirm that Lilotane outperforms established SAT-based approaches, often by orders of magnitude, produces much smaller formulae on average, and compares favorably to other state-of-the-art HTN planners regarding robustness and plan quality. In the International Planning Competition (IPC) 2020, a preliminary version of Lilotane scored the second place. We expect these considerable improvements to SAT-based HTN planning to open up new perspectives for SAT-based approaches in related problem classes.

Scalable SAT Solving in the Cloud

“…Secondly, while Mallob's performance with just a single solver backend is a remarkable result, we expect to further improve Mallob's performance by integrating further state-of-the-art solver backends. Thirdly, we intend to advance Mallob by adding support for incremental SAT solving and for related applications such as automated planning [26]…”

“…Today's applications of SAT solving are manifold and include areas such as cryptography [23], formal software verification [11], and automated planning [26]. Application-specific SAT encoders generate formulae which represent the problem at hand stated in propositional logic.…”

“…Application-specific SAT encoders generate formulae which represent the problem at hand stated in propositional logic. Oftentimes, multiple formulae which represent different aspects or horizons of the problem are generated [11,26]. The difficulty of these individual formulae ranges from trivial to extremely difficult and is usually not known beforehand.…”

Making Translations to Classical Planning Competitive with Other HTN Planners

“…We start with the TO benchmark set of the IPC 2020. We have compared our approach against the five IPC competitors (HyperTensioN (Magnaguagno, Meneguzzi, and de Silva 2021), Lilotane (Schreiber 2021), PDDL4J (Pellier and Fiorino 2021), SIADEX (Fernandez-Olivares, Vellido, and Castillo 2021), pyHiPOP (Lesire and Albore 2021)) as well as against the original encoding HTN2STRIPS, the translation-based approach by Höller (2021) (TOAD), and the most recent versions of progression search (Greedy RC Höller and Behnke 2021)) and SATbased HTN planning (pandaPisatt-1iB (Behnke 2021a)). In Tab.…”

HDDL 2.1: Towards Defining an HTN Formalism with Time

“…Over the last years, much progress has been made in the field of hierarchical planning (Bercher, Alford, and Höller, 2019). Novel systems based on the traditional, search-based techniques have been introduced (Bit-Monnot, Smith, and Do, 2016;Ramoul et al, 2017;Shivashankar, Alford, and Aha, 2017;Bercher et al, 2017;Höller et al, 2019Höller et al, , 2020bHöller and Bercher, 2021), but also new techniques like the translation to STRIPS/ADL (Alford, Kuter, and Nau, 2009;Alford et al, 2016;Behnke et al, 2022), or revisited approaches like the translation to propositional logic Biundo, 2018, 2019;Schreiber et al, 2019;Schreiber, 2021;Behnke, 2021).…”