POPF
POPF is a temporal planner that searches forward through states but keeps its plans only partially ordered. It handles durative actions, continuous numeric change and problems where actions must overlap.
- Authors
- Amanda Coles, Andrew Coles, Maria Fox, Derek Long
- Institution
- At publication: University of Strathclyde
- Kind of planning
- Temporal and numeric, with linear continuous change
- Source
- POPF on SourceForge
How it works
In the words of its paper, POPF "is built on the foundations of grounded forward search, in combination with linear programming to handle continuous linear numeric change". Rather than fix the order and timing of actions as it goes, it keeps them as sets of constraints, committing only as far as it must. It builds on the COLIN planner, including its temporal relaxed planning graph heuristic.
What it handles
Numeric fluents, durative actions, duration inequalities, required concurrency, continuous effects and timed initial literals. It does not handle disjunctive or quantified conditions, or conditional effects. See the feature comparison.
In the competitions
POPF2 was joint runner-up in the temporal satisficing track of IPC 2011, and, according to the competition's survey, the only planner to solve problems in all domains with required concurrency.
Results in our benchmarks
| Track | Problems | POPF | Autoplan |
|---|---|---|---|
| IPC 1998–2018 classical | 2,742 | 751 | 2,450 |
| IPC 2023 classical | 140 | 0 | 109 |
| IPC 2023 numeric | 400 | 56 | 383 |
| IPC 2026 numeric | 320 | 30 | 270 |
| IPC 2002 temporal | 436 | 132 | 338 |
| IPC 2004 temporal | 976 | 343 | 715 |
| IPC 2006 temporal | 450 | 141 | 308 |
| IPC 2008 temporal | 390 | 213 | 348 |
| IPC 2011 temporal | 240 | 110 | 218 |
| IPC 2014 temporal | 200 | 41 | 181 |
| IPC 2018 temporal | 90 | 36 | 86 |
POPF popf2-11jun2011: 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.
Paper
- Amanda Coles, Andrew Coles, Maria Fox, Derek Long. Forward-Chaining Partial-Order Planning. Proceedings of ICAPS 2010.