Research map / Learning theory and algorithms / Combinatorial and search optimization
Dealing with Numeric and Metric Time Constraints in PDDL3 via Compilation to Numeric Planning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Compiles PDDL3 trajectory constraints with numeric and metric-time conditions into a plain numeric planning problem via regression (NTCORE), with an achiever-based optimization (NTCORE+), and solves it with ENHSP.
Problem
Planning with PDDL3 trajectory constraints that mix propositional and numeric conditions and metric time bounds, which existing planners handle poorly.
Approach
Compile PDDL3 constraints away using numeric regression over action effects, with a time variable for metric-time constraints, so any numeric planner can solve the result; an achiever analysis over-approximates which actions can affect each constraint to shrink the compiled problem.
Key results
- NTCORE+ solves 351 of 512 instances versus 71 for OPTIC, the best native PDDL3 system (about five times the coverage).
- The achiever-based optimization NTCORE+ raises coverage by 21.4% over basic NTCORE (289 solved) and is an order of magnitude faster on many instances.
- NTCORE+ adds up to two orders of magnitude fewer precondition and effect terms than NTCORE in several domains (e.g., PLANTWATERING preconditions 1766052.7 vs 44359.1).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add achiever-based optimization to the basic NTCORE compilation (ablation) | NTCORE · All domains (512 instances) | coverage: 289 -> 351; 21.4% | Helped |
| Add achiever-based optimization to the basic NTCORE compilation (ablation) | NTCORE · All domains (512 instances) | runtime: 319 instances | Helped |
| Add achiever-based optimization to the basic NTCORE compilation (ablation) | NTCORE · All domains (512 instances) | additional terms in preconditions and effects: up to two orders of magnitude fewer | Helped |
| Use NTCORE+ compilation instead of MIPS-XXL | NTCORE+ and MIPS-XXL · TPP | coverage: 12 vs. 25 | Hurt |
| Use NTCORE+ compilation instead of OPTIC | NTCORE+ and OPTIC · PLANTWATERING | plan length: 80 fewer actions | Helped |
| Use NTCORE+ compilation instead of MIPS-XXL | NTCORE+ and MIPS-XXL · TPP | plan length: 72.9% fewer actions | Hurt |
Surprising findings
- On mostly propositional domains such as TPP, OPTIC and MIPS-XXL outperform the compilation approach, which the authors attribute to ENHSP being tuned for numeric domains.
Limitations
- Compilation overhead and reliance on ENHSP, which is Java-based and has JVM startup cost.
- Equality numeric conditions cannot be fully handled by the achiever analysis since exact values are unknown at compile time.
- Support limited to instantaneous actions; temporal planning and LTL modulo theories are left for future work.
Experimental setup
Datasets: COUNTERS ROVERS DEPOTS TPP ZENOTRAVEL FARMLAND PLANTWATERING SAILING BLOCKGROUPING
Benchmarks: COUNTERS ROVERS DEPOTS TPP ZENOTRAVEL FARMLAND PLANTWATERING SAILING BLOCKGROUPING
Metrics: coverage runtime plan length number of additional precondition and effect terms
Baselines: OPTIC SGPLAN MIPS-XXL Metric-FF
Builds on: TCORE PDDL3 numeric regression PDDL 2.1
Closest papers by meaning
- Temporal Numeric Planning with Patterns (AAAI 2025 · similarity 0.89)
- PDBs Go Numeric: Pattern-Database Heuristics for Simple Numeric Planning (AAAI 2025 · similarity 0.88)
- Leveraging Environment Interaction for Automated PDDL Translation and Planning with Large Language Models (NeurIPS 2024 · similarity 0.86)
- Towards Practical Classical Planning Compilations of Numeric Planning (AAAI 2025 · similarity 0.86)
- Symbolic Numeric Planning with Patterns (AAAI 2024 · similarity 0.85)
- Situation Calculus Temporally Lifted Abstractions for Generalized Planning (AAAI 2025 · similarity 0.84)
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.