Semantic Partial Grounding via LLMs
arXiv:2602.22067v1 Announce Type: new Abstract: Grounding is a critical step in classical planning, yet it often becomes a computational bottleneck due to the exponential growth in grounded actions and atoms as task size increases. Recent advances in partial grounding have addressed this challenge by incrementally grounding only the most promising operators, guided by predictive models. However, these approaches primarily rely on relational features or learned embeddings and do not leverage the textual and structural cues present in PDDL descriptions. We propose SPG-LLM, which uses LLMs to analyze the domain and problem files to heuristically identify potentially irrelevant objects, actions, and predicates prior to grounding, significantly reducing the size of the grounded task. Across seven hard-to-ground benchmarks, SPG-LLM achieves faster grounding-often by orders of magnitude-while delivering comparable or better plan costs in some domains.
arXiv:2602.22067v1 Announce Type: new Abstract: Grounding is a critical step in classical planning, yet it often becomes a computational bottleneck due to the exponential growth in grounded actions and atoms as task size increases. Recent advances in partial grounding have addressed this challenge by incrementally grounding only the most promising operators, guided by predictive models. However, these approaches primarily rely on relational features or learned embeddings and do not leverage the textual and structural cues present in PDDL descriptions. We propose SPG-LLM, which uses LLMs to analyze the domain and problem files to heuristically identify potentially irrelevant objects, actions, and predicates prior to grounding, significantly reducing the size of the grounded task. Across seven hard-to-ground benchmarks, SPG-LLM achieves faster grounding-often by orders of magnitude-while delivering comparable or better plan costs in some domains.
Executive Summary
The article proposes SPG-LLM, a novel approach to semantic partial grounding via large language models (LLMs) in classical planning. By analyzing PDDL descriptions, SPG-LLM heuristically identifies and prunes irrelevant objects, actions, and predicates, significantly reducing the size of the grounded task. This results in faster grounding times, often by orders of magnitude, while maintaining comparable or better plan costs across seven hard-to-ground benchmarks. The approach leverages textual and structural cues in PDDL descriptions, addressing the computational bottleneck in grounding.
Key Points
- ▸ SPG-LLM uses LLMs to analyze PDDL descriptions
- ▸ Heuristically identifies and prunes irrelevant objects, actions, and predicates
- ▸ Achieves faster grounding times and comparable or better plan costs
Merits
Efficient Grounding
SPG-LLM significantly reduces grounding time, making it a promising approach for large-scale planning tasks.
Demerits
Limited Domain Knowledge
The approach relies on the quality of the PDDL descriptions and the LLM's ability to understand the domain, which may be limited in certain cases.
Expert Commentary
The proposed SPG-LLM approach represents a significant advancement in semantic partial grounding, addressing the long-standing challenge of computational bottlenecks in classical planning. By leveraging the capabilities of LLMs, SPG-LLM can efficiently prune irrelevant objects, actions, and predicates, resulting in faster grounding times and improved plan costs. However, the approach's reliance on high-quality PDDL descriptions and the LLM's domain knowledge may limit its applicability in certain cases. Further research is needed to explore the potential of SPG-LLM in various planning domains and to address its limitations.
Recommendations
- ✓ Further evaluation of SPG-LLM in diverse planning domains to assess its robustness and applicability
- ✓ Investigation of techniques to improve the LLM's domain knowledge and understanding of PDDL descriptions
