PDDL papers

IPC-1998

The paper by McDermott et al. is the original definition of PDDL, as used for IPC-1 in 1998.

IPC-2000

This paper defines the subset of PDDL that was used for IPC-2 in 2000. Many features of the original PDDL that had not been used at the planning competition were not included (e.g. domain axioms, safety constraints, hierarchical actions, numerical fluents). All further extensions of PDDL add to this subset, rather than the original PDDL.

IPC-2002 (aka IPC-3) and related

The first paper introduces PDDL 2.1, which extends the version of PDDL used in the previous competitions by numeric state variables and an explicit model of concurrency (durative actions, with either discrete or continuous effects). The second paper further extends PDDL 2.1 into a language called PDDL+, introducing autonomous processes that are triggered either by actions of the planning agent or by effects of other autonomous processes. The other papers are "opinion papers", written as a response to PDDL 2.1.

IPC-2004 (aka IPC-4) and related

The papers by Edelkamp and Hoffmann introduce PDDL 2.2, which extends PDDL 2.1 by derived predicates (state variables which are computed as a function of other state variables) and by timed initial literals (propositions that are "autonomously" set to true or false at a certain time point in a temporal plan). The papers by Thiébaux et al. provide motivation, formal semantics and analysis of expressivity for derived predicates.

IPC-2006 (aka IPC-5) and related

These papers describe PDDL 3.0, which extends PDDL 2.2 by trajectory constraints and preferences. A trajectory constraint is a constraint on the set of valid plans; they are expressed in a temporal logic. Preferences allow expressing soft constraints (soft trajectory constraints, soft preconditions and soft goals), which are constraints that need not be satisfied by a plan, but lead to a decrease in plan quality if they are not.

IPC-2008 and related

The first paper introduces an extension of the STRIPS formalism to include functional state variables, which map to objects of the planning problem instead of the set {true, false}. Functional state variables often allow modeling a domain more naturally. The main language extension in PDDL 3.1, object fluents, are modeled after Functional Strips. The SAS+ planning formalism (see second paper) can be seen as a special case.

PddlResources (last edited 2011-04-19 11:47:54 by MalteHelmert)