DiNo
DiNo, the Discretised Nonlinear Heuristic Planner, adds heuristic guidance to UPMurphi's approach to PDDL+. It was built for hybrid domains with non-linear dynamics.
- Authors
- Wiktor Piotrowski, Maria Fox, Derek Long, Daniele Magazzeni, Fabio Mercorio
- Institutions
- At publication: King's College London, University of Milan-Bicocca
- Kind of planning
- PDDL+, linear and non-linear
- Source
- github.com/KCL-Planning/DiNo
How it works
DiNo is built on UPMurphi and keeps its Discretise and Validate method. It adds a heuristic, the Staged Relaxed Planning Graph+, which guides enforced hill-climbing through the discretised space, with deferred heuristic evaluation for completeness.
What it handles
As for UPMurphi: PDDL+ processes and events and non-linear continuous change, which Autoplan does not yet handle. Domains must be typed, and the only metric is minimising total time. See the feature comparison.
Results in our benchmarks
| Track | Problems | DiNo | Autoplan |
|---|---|---|---|
| IPC 2023 numeric | 400 | 5 | 383 |
| IPC 2026 numeric | 320 | 2 | 270 |
| IPC 2002 temporal | 436 | 0 | 338 |
| IPC 2004 temporal | 976 | 0 | 715 |
| IPC 2006 temporal | 450 | 0 | 308 |
| IPC 2008 temporal | 390 | 0 | 348 |
| IPC 2011 temporal | 240 | 0 | 218 |
| IPC 2014 temporal | 200 | 0 | 181 |
| IPC 2018 temporal | 90 | 0 | 86 |
DiNo f95f07d: 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.
DiNo was built for hybrid systems with non-linear dynamics, not for speed on IPC benchmarks. Low numbers here say little about the problems it was made for.
Paper
- Wiktor Piotrowski, Maria Fox, Derek Long, Daniele Magazzeni, Fabio Mercorio. Heuristic Planning for PDDL+ Domains. IJCAI 2016, 3213-3218.