Research map / Graph learning / Graph algorithms and clustering
Learning Heuristic Functions with Graph Neural Networks for Numeric Planning
Valerio Borelli, Alfonso Gerevini, Enrico Scala, Ivan Serina
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| feed numeric values to the per-condition MLPs on edges (h_n) instead of abstracting numeric conditions as boolean (h_c) (ablation) | GNN numeric planning heuristic · IPC 2023 numeric domains | total coverage: 21 -> 75 | Improved |
| add numeric structure (h_c) over ignoring it (h_0) (ablation) | GNN numeric planning heuristic · IPC 2023 numeric domains | total coverage: 11 -> 21 | Improved |
| use numeric-value encoding h_n instead of h_c on a domain with few numeric conditions (ablation) | GNN numeric planning heuristic · Mprime | coverage: 9 -> 8 | Hurt |
| use smooth-max aggregation instead of add aggregation (ablation) | GNN numeric planning heuristic · numeric planning domains | coverage | Hurt |
| use mean or max aggregation instead of add (ablation) | GNN numeric planning heuristic · numeric planning domains | coverage | No clear effect |
| train on suboptimal plans in addition to optimal plans (ablation) | GNN numeric planning heuristic · numeric planning training data | validation error | Hurt |
| decompose n-ary numeric goals into a chain of n-1 overlapping binary goals | GNN numeric planning heuristic · Fo-Farmland | average expanded nodes: 531201 -> 56.30 | Improved |
| use learned h_n instead of traditional h_mrp on Fo-Farmland | greedy best-first search · Fo-Farmland | coverage: 5 -> 14 | Improved |
| use learned h_n instead of traditional h_mrp on Sailing | greedy best-first search · Sailing | coverage: 20 -> 2 | Hurt |
| use h_n instead of graph-kernel h_ccWLF_rank on domains with numeric goals and preconditions | greedy best-first search · Counters | coverage: 3 -> 10 | Improved |
| use h_n instead of h_ccWLF_rank on boolean-goal domains | greedy best-first search · Sailing | coverage: 5 -> 2 | Hurt |
| use h_n instead of h_ccWLF_rank on Mprime | greedy best-first search · Mprime | coverage: 13 -> 8 | Hurt |
Closest papers by meaning
- Learning Domain-Independent Heuristics for Grounded and Lifted Planning (AAAI 2024 · similarity 0.91)
- Graph Learning for Numeric Planning (NeurIPS 2024 · similarity 0.91)
- State Encodings for GNN-Based Lifted Planners (AAAI 2025 · similarity 0.90)
- Graph Neural Network Based Action Ranking for Planning (NeurIPS 2025 · similarity 0.88)
- PDBs Go Numeric: Pattern-Database Heuristics for Simple Numeric Planning (AAAI 2025 · similarity 0.87)
- Learning More Expressive General Policies for Classical Planning Domains (AAAI 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
