AI planners

The planners we compare Autoplan with, what each one is for, who built it and where to read about it. Each has a page of its own with its papers, its source code and how it did on our benchmarks.

Automated planners are research software first. Most began as a paper from a university group and many were entered in the International Planning Competition. They differ in what they can model and how they search, so the right choice depends on your problem. For a feature-by-feature comparison, see Supported features. For the numbers, see Benchmarks.

Classical planners

Planner	From	Approach
LAMA and Fast Downward	Silvia Richter, Matthias Westphal, Malte Helmert	Heuristic search with landmarks, anytime

Numeric planners

Planner	From	Approach
Metric-FF	Jörg Hoffmann, University of Freiburg	Relaxed plans over numeric variables
ENHSP	Enrico Scala and colleagues, ANU and University of Brescia	Heuristic search; classical, numeric and PDDL+
Count Downward	Daniel Gnad and colleagues, Heidelberg, Linköping, Technion, Bar-Ilan	Numeric Fast Downward with abstraction heuristics
Panino	Giacomo Rosa and colleagues, University of Melbourne and RMIT	Best-first search with numeric novelty
Patty	Matteo Cardellini and colleagues, University of Genova	Planning as satisfiability, with SMT and patterns
Tyr	Dominik Drexler, Linköping University	Ground and lifted search

Temporal planners

Planner	From	Approach
LPG-td	Alfonso Gerevini, Alessandro Saetti, Ivan Serina, University of Brescia	Stochastic local search on action graphs
POPF	Amanda Coles, Andrew Coles, Maria Fox, Derek Long, University of Strathclyde	Forward search with partial orders and linear programming
OPTIC	J. Benton, Amanda Coles, Andrew Coles, Arizona State and King's College London	POPF extended with preferences and time-dependent costs
Temporal Fast Downward	Patrick Eyerich, Robert Mattmüller, Gabriele Röger, University of Freiburg	Search over time-stamped states

PDDL+ planners

Planner	From	Approach
UPMurphi	Giuseppe Della Penna and colleagues, University of L'Aquila	Explicit model checking, universal plans
DiNo	Wiktor Piotrowski and colleagues, King's College London	UPMurphi with a heuristic for hybrid domains

ENHSP also handles PDDL+. The groupings show what each planner is best known for; several handle more than one kind of problem.

The International Planning Competition

The International Planning Competition (IPC) has compared planners on shared benchmarks since 1998, and it is where much of the PDDL language took shape. The ICAPS competitions page lists the editions. Their published domains and problems remain a common test set for planning research.

IPC 2002 brought in temporal and metric domains, written in PDDL 2.1.
IPC 2023 classical and IPC 2023 numeric were reviewed in the AI Magazine report.
IPC 2026 numeric was organised by Luigi Bonassi, Joan Espasa Arxer, Francesco Percassi and Enrico Scala, and presented at ICAPS 2026 in Dublin. Count Downward, Panino, Patty and Tyr were all entries.

Our benchmarks use these published problems, but they are our own runs under our own limits, not competition results.