Panino
Panino, the PArtitioned NumerIc NOvelty solver, brings novelty search to numeric planning. It won the overall agile and satisficing awards of the IPC 2026 numeric track.
- Authors
- Giacomo Rosa, Jean Honorio, Nir Lipovetzky, Sebastian Sardiña
- Institutions
- University of Melbourne, RMIT University
- Kind of planning
- Classical and numeric
- Source
- IPC 2026 repository, built on JPDDLPlus
How it works
Panino runs a greedy best-first search whose main feature, in its abstract's words, is "a partitioned numeric novelty measure that adapts classical novelty to features induced by numeric and propositional subgoals". States that reach something new score well, and the novelty table is partitioned by a subgoaling heuristic. A companion paper develops the theory, a notion of width for numeric problems.
The competition had two configurations: Panino for agile planning, and Panino-anytime for satisficing, which restarts with tighter cost bounds.
What it handles
Classical PDDL with conditional effects and action costs, some forms of disjunctive and quantified conditions, numeric conditions and effects, and plan metrics. It does not handle durative actions. See the feature comparison.
In the competitions
At IPC 2026, Panino won the overall agile track and Panino-anytime the overall satisficing track, according to the published results.
Results in our benchmarks
| Track | Problems | Panino | Autoplan |
|---|---|---|---|
| IPC 2023 numeric | 400 | 290 | 383 |
| IPC 2026 numeric | 320 | 217 | 270 |
Panino 67c8c08: 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.
The Benchmarks page has a second row, Panino-LC, on the classical tracks. That is our name, not its authors': it is Panino run on lower-cased input through our own wrapper, because PDDL is case-insensitive but Panino's parser is not. It is otherwise identical to the Panino row.
| Track | Problems | Panino-LC | Autoplan |
|---|---|---|---|
| IPC 1998–2018 classical | 2,742 | 1,316 | 2,450 |
| IPC 2023 classical | 140 | 17 | 109 |
Panino-LC 67c8c08: 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
- Giacomo Rosa, Jean Honorio, Nir Lipovetzky, Sebastian Sardiña. IPC 2026 Numeric Track: The Panino Solver. Planner abstract, IPC 2026 numeric track.
- Giacomo Rosa, Jean Honorio, Nir Lipovetzky, Enrico Scala, Sebastian Sardiña. A Notion of Width for Numeric Planning Problems. ICAPS 2026 HSDIP Workshop.