New SAT Encoding Strategies for Factored Tasks Boost AI Planning Efficiency
Researchers reveal which task transformations improve SAT-based planners and which hurt performance.
Factored task specifications (FTS) allow compact representation of AI planning problems, supporting disjunctive preconditions, conditional effects, and angelic nondeterminism. Until now, planners for factored tasks relied solely on heuristic search—this work by João Filipe, Álvaro Torralba, and Gregor Behnke bridges that gap by investigating SAT-based approaches. The authors propose several ways to encode the factored transition relation into propositional logic, each balancing formula size and solver efficiency. They also explore parallelism exploitation at multiple levels within the SAT encoding, and systematically study how common task transformations (like lifting or splitting) impact solver performance.
Experimental results on standard benchmarks show that certain encodings dramatically reduce solver runtime, while others can increase clause count and slow down search. The paper identifies specific transformation patterns that consistently help or hurt—for example, transforming away conditional effects can simplify the SAT instance, while aggressive factoring may introduce redundant constraints. This work provides a practical roadmap for building high-performance SAT-based planners for factored tasks, enabling more expressive planning domains without sacrificing computational tractability. The findings are especially relevant for researchers working on AI planning, propositional satisfiability, and automated reasoning.
- Proposes three novel SAT encoding strategies for factored transition relations, each with different trade-offs in formula size and solver efficiency
- Demonstrates parallelism exploitation at multiple levels (action, state, and task) within the SAT-based planning framework
- Identifies specific task transformations (e.g., removing conditional effects) that improve SAT solver performance by 2-5x, while others can degrade it by adding redundant constraints
Why It Matters
This work enables SAT solvers to handle richer planning problems efficiently, opening new applications for automated reasoning in AI.