The LAMA Planner: Guiding Cost-Based Anytime Planning with Landmarks (1401.3839v1)

Abstract: LAMA is a classical planning system based on heuristic forward search. Its core feature is the use of a pseudo-heuristic derived from landmarks, propositional formulas that must be true in every solution of a planning task. LAMA builds on the Fast Downward planning system, using finite-domain rather than binary state variables and multi-heuristic search. The latter is employed to combine the landmark heuristic with a variant of the well-known FF heuristic. Both heuristics are cost-sensitive, focusing on high-quality solutions in the case where actions have non-uniform cost. A weighted A* search is used with iteratively decreasing weights, so that the planner continues to search for plans of better quality until the search is terminated. LAMA showed best performance among all planners in the sequential satisficing track of the International Planning Competition 2008. In this paper we present the system in detail and investigate which features of LAMA are crucial for its performance. We present individual results for some of the domains used at the competition, demonstrating good and bad cases for the techniques implemented in LAMA. Overall, we find that using landmarks improves performance, whereas the incorporation of action costs into the heuristic estimators proves not to be beneficial. We show that in some domains a search that ignores cost solves far more problems, raising the question of how to deal with action costs more effectively in the future. The iterated weighted A* search greatly improves results, and shows synergy effects with the use of landmarks.

• The paper introduces the LAMA planner, which fuses landmark heuristics with a cost-sensitive FF-based approach for iterative plan quality enhancement.
• It employs an iterated weighted A* search that balances action costs with goal distance estimates, ensuring continuous improvement over time.
• Empirical evaluations on IPC benchmarks demonstrate LAMA’s effectiveness across domains while identifying opportunities for further heuristic refinement.

Overview of The LAMA Planner: Guiding Cost-Based Anytime Planning with Landmarks

The paper provides a comprehensive examination of the LAMA (Landmarks-Based Anytime Multi-heuristic Algorithm) planning system, a prominent classical planning system that operates on heuristic forward search. LAMA’s distinctive feature is its utilization of landmarks—propositional formulas that must necessarily hold in every solution of a planning task—to steer its search.

Core Contributions:

• Integration of Landmark Heuristics: The key innovation in LAMA is the combination of a landmark-based pseudo-heuristic with the established FF (Fast-Forward) heuristic. Importantly, both heuristics are adapted to be cost-sensitive.
• Iterated Anytime Search: LAMA employs a weighted A* search that iteratively reduces the weight parameter, facilitating continuous improvement of the plan quality until the search is terminated.
• Performance Optimization: The system leverages Fast Downward’s finite-domain state variables approach and multi-heuristic search to optimize heuristic estimates and preferred operators, significantly enhancing planning performance.

Detailed System Description:

Landmarks and Heuristics:

Landmarks in LAMA are utilized in tandem with a heuristic based on the FF heuristic. Specifically:

• Landmark Heuristic Construction: The system builds on Fast Downward by extending the causal graph heuristic to include landmarks. It identifies landmarks and their orderings through a back-chaining process complemented by domain transition graph analysis and relaxed planning graph pruning.
• Action Costs and Goal Distance Combination: LAMA balances action cost awareness with goal distance estimations. This combined heuristic approach is pivotal for handling zero-cost operators effectively, adapting cost-sensitive planning without exclusively focusing on costs.

Anytime Search Mechanism:

• Greedy Best-First Search and Weighted A*: After an initial plan is found via a greedy best-first search, LAMA transitions to iterated weighted A* searches with decreasing weights. This process iteratively enhances the plan quality, exploiting the synergy between multi-queue heuristic search and deferred heuristic evaluation.
• Preferred Operators: The use of preferred operators, derived from heuristic functions, ensures that the most promising actions are prioritized in the search, further optimizing performance.

Empirical Evaluation:

The empirical assessment, focused on the International Planning Competition (IPC) 2008 benchmarks, reveals several insights:

• Comparison with FF/Add Heuristic: The pure cost-sensitive FF/add heuristic (Fc) performs poorly compared to the cost-unaware version (F) due to significant reductions in coverage. The combination heuristic (FLc) combines distance and cost estimates, achieving nearly optimal IPC performance.
• Iterated Search Effectiveness: Iterated search significantly improves solution quality by allowing continuous refinement. For instance, the Openstacks domain exemplifies how initial suboptimal plans due to the landmark heuristic can be substantially improved through iterative search.

Domain-Specific Observations:

• Elevators: The cost-sensitive heuristics suffer due to their focus on slow elevators and the intricacies of capacity constraints, which introduce plateaus and local minima. Landmarks mitigate some issues but do not fully resolve them.
• PARC Printer: Similar problems occur due to lengthy relaxed plans for cheap paths. However, landmarks improve performance more substantially here.
• Cyber Security: A domain where cost-sensitive configurations notably outperform the cost-unaware ones, reflecting accurate budget representations in heuristic estimations.
• Openstacks: This domain highlights the downside of landmarks in practice, where the goal distance estimation can lead to high-cost plans. Nevertheless, the landmark heuristic’s speed allows iterative search to significantly refine the plans.

Implications and Future Work:

The LAMA planner’s use of cost-sensitive heuristics and landmarks provides crucial benefits for planning efficiency and solution quality. However, the challenges identified suggest several areas for future research:

• Refinement of Cost-Sensitive Heuristics: Addressing the inaccuracies in the FF/add heuristic and exploring alternative heuristic approaches to more effectively balance cost and distance estimations.
• Enhanced Landmark Identification: Extending landmarks to include multiple occurrences and complex propositional formulas, and improving the definition and utility of reasonable orderings.
• Further Synergy Exploration: Investigating more sophisticated interactions between landmarks and other heuristic methods or search strategies.

In conclusion, LAMA’s innovative integration of landmarks, cost-sensitive heuristics, and an iterated anytime search mechanism marks significant progress in heuristic-driven classical planning. Future advancements could further refine and amplify these benefits, contributing to the development of more sophisticated planning systems.