Parallel Lifted Planning via Semi-Naive Datalog Evaluation
Researchers have developed a new execution model for lifted classical planning that introduces two levels of parallelism: rule-level and grounding parallelism. This approach builds on prior work linking lifted planning to Datalog evaluation. The solver uses a grounder based on clique enumeration, extended to support semi-naive Datalog evaluation. Experimental evaluation with greedy best-first search and the FF heuristic shows the implementation solves more tasks than the baseline. The work aims to address the traditional slowness of lifted planning by reducing the need for repeated instantiation of ground structures during search.
Key facts
- Lifted classical planners avoid computationally demanding grounding steps.
- Lifted planning is typically slower due to repeated instantiation of ground structures.
- Core components of lifted planning have been studied through Datalog evaluation.
- The new model has two levels of parallelism: rule-level and grounding parallelism.
- The solver uses a grounder based on clique enumeration.
- The grounder supports semi-naive Datalog evaluation.
- Experimental evaluation used greedy best-first search with the FF heuristic.
- The implementation solves more tasks than the baseline.
Entities
Institutions
- arXiv