Efficient SAT Encodings for Hierarchical Planning

Evaluation of high level methods for efficient planning as satisfiability

“…Probably one of the biggest contributors to speed in planning as satisfiability is parallel planning and it is presented in detail in [20]. Moreover, there is compositional planning as investigated in [21], and hierarchical planning [22]. Big performance improvements in planning as satisfiability can be unlocked when methods are altered on the low, SAT solving level.…”

Lilotane: A Lifted SAT-based Approach to Hierarchical Planning

“…Independently and almost simultaneously, Schreiber et al (2019a) developed a SAT encoding for TOHTN planning problems which exploits incremental SAT solving by simulating a stack machine of tasks and using the number of stack machine transitions as the problem horizon to increase. An enhancement of this approach resulted in the Tree-REX planner (Schreiber et al, 2019b) which was shown to be much more efficient than its precursor while capable to find the shortest possible plan at the first solvable layer.…”

“…This planning approach, after two decades of inactivity since its initial proposal (Mali & Kambhampati, 1998), recently received a significant amount of attention. On the one hand, new techniques for much more compact SAT encodings were found Schreiber, Pellier, Fiorino, & Balyo, 2019a;Schreiber, Pellier, Fiorino, et al, 2019b) while on the other hand new grounding approaches (Ramoul, Pellier, Fiorino, & Pesty, 2017;Behnke, Höller, Schmid, Bercher, & Biundo, 2020) serve as a catalyst to improve on SAT-based approaches, not only speeding up the processing of a planning description but also leading to smaller encodings and therefore better performance.…”

Tree-REX: SAT-Based Tree Exploration for Efficient and High-Quality HTN Planning

“…Their encodings share a clause and variable complexity cubic in the amount of tasks, rendering it not viable for today's HTN planning problems of non-trivial size. Current works on SAT-based totally ordered HTN planning include totSAT (Behnke, Höller, and Biundo 2018a) as well as the encodings proposed in (Schreiber et al 2019) which introduced incremental SAT solving to hierarchical planning and served as a basis for the work at hand. A detailed comparison of Tree-REX to the latter can be found in (Schreiber 2018).…”

“…In contrast to SAT-based classical planning, research on SAT-based HTN planning lay idle for nearly two decades after its initial proposal (Mali and Kambhampati 1998). Recently, the topic was revisited by (Behnke, Höller, and Biundo 2018a) and (Schreiber et al 2019) who each proposed new, modern SAT encodings for HTN problems with totally ordered subtasks. However, these approaches do not fully exploit the potential of modern SAT solving yet, and they produce plans of improvable quality.…”