Count Downward
Count Downward extends Numeric Fast Downward with abstraction heuristics for numeric planning. Its configurations won the agile and satisficing awards for simple numeric planning at IPC 2026, and a portfolio of them, Count Downward Together, won the optimal award.
- Authors
- Daniel Gnad, Markus Fritzsche, Mikhail Gruntov, Alexander Shleyfman
- Institutions
- Heidelberg University, Linköping University, Technion, Bar-Ilan University
- Kind of planning
- Classical and numeric
- Source
- Project page, github.com/dgnad/numeric-fast-downward (GPL-3.0)
How it works
Its abstract describes "a planner for cost-optimal, satisficing, and agile numeric planning built on top of the Numeric Fast Downward framework". For optimal planning it uses numeric abstraction heuristics, pattern databases and domain abstractions, under A*. Its satisficing and agile configurations use lazy greedy best-first search with an interval-relaxed FF heuristic, and the satisficing one adds weighted A* restarts.
Its authors now recommend PlanForge, a reimplementation in Rust, and say Count Downward receives only rudimentary maintenance.
What it handles
Classical PDDL with ADL features and action costs, numeric conditions and effects, and plan metrics. It does not handle durative actions. See the feature comparison.
In the competitions
In the simple numeric planning category of IPC 2026, Count Downward won the agile and satisficing tracks and Count Downward Together won the optimal track, according to the published results.
Results in our benchmarks
| Track | Problems | Count-Downward | Autoplan |
|---|---|---|---|
| IPC 1998–2018 classical | 2,742 | 1,914 | 2,450 |
| IPC 2023 classical | 140 | 32 | 109 |
| IPC 2023 numeric | 400 | 223 | 383 |
| IPC 2026 numeric | 320 | 234 | 270 |
Count-Downward 42de017: problems solved with a plan the VAL validator accepts, within 60 seconds and 4 GB of memory on one core. These are our own runs, not competition results, and a planner written for other goals or for a different kind of problem can score low here while being the right tool for its purpose. How we measured.
Papers
- Daniel Gnad, Markus Fritzsche, Mikhail Gruntov, Alexander Shleyfman. Count Downward. Planner abstract, IPC 2026 numeric track.
- Daniel Gnad, Lee-Or Alon, Eyal Weiss, Alexander Shleyfman. PDBs Go Numeric: Pattern-Database Heuristics for Simple Numeric Planning. Proceedings of AAAI 39(25), 26507-26515, 2025.
- Markus Fritzsche, Daniel Gnad, Mikhail Gruntov, Alexander Shleyfman. Managing Infinite Abstractions in Numeric Pattern Database Heuristics. Proceedings of AAAI 40(43), 36227-36235, 2026.
- Markus Fritzsche, Mikhail Gruntov, Alexander Shleyfman, Daniel Gnad. Domain-Abstraction Heuristics for Simple Numeric Planning. Proceedings of SoCS 19(1), 224-228, 2026.