For IPC-5 we have defined some significant extensions to PDDL2.2 (the language of the 2004 competition, IPC-4), which are described in "A. Gerevini, P. Haslum, D. Long, A. Saetti, Y, Dimopoulos, Deterministic planning in the fifth international planning competition: PDDL3 and experimental evaluation of the planners, Artificial Intelligence, volume 173(5-6), pp. 619-668, 2009". The BNF grammar of the new language (PDDL3.0) is given in another related document (Postscript version, PDF version). PDDL3.0 is also described in a recent paper presented at the "Workshop on Preferences and Soft Constraint in Planning (ICAPS 2006)", which contains some examples from the benchmarks used in IPC-5.

The new PDDL3.0 language has been developed in collaboration with Derek Long. The main new features of the language are "soft goals", which are desired goals that a valid plan does not have to necessarily achieve, and "state trajectory constraints", which are constraints on the structure of the plans. Trajectory constraints can be either hard or soft. Hard trajectory constraints can be used to express control knowledge or restrictions on the valid plans for a planning domain and/or for a specific planning problem. Soft goals and trajectory constraints can be used to express preferences that affect the plan quality, without restricting the set of the valid plans.

In PDDL3, a soft goal or constraint is a "preference" that has a "penalty weight" associated with its violation in the plan. The specification of the plan metric for a plan includes these violation penalty weights that have to be minimized. In general, not all the specified preferences (possibly with different penalty weights) can be satisfied, and identifying the best subset of preferences that can be achieved is an extra difficulty to deal with in the planning process. Moreover, if there is a CPU-time limit for planning, in order to produce good quality solutions, the planner has also the identify a good (possibly suboptimal) set of soft goals/constraints that can be achieved within the available amount of computational resources.

The papers on PDDL3 mentioned above contain many examples of soft goals as well as of hard and soft trajectory constraints.


PDDL2.1 (more precisely, the first three levels of PDDL2.1) forms the basis for the IPC-4 language, called PDDL2.2, developed by Stefan Edelkamp and Joerg Hoffmann. Like PDDL2.1, PDDL2.2 is divided into three levels corresponding to ADL, numeric, and durational planning. On top of the PDDL2.1 language features, two new constructs are added:


PDDL -- the Planning Domain Definition Language -- is the standard language for the encoding of planning domains. The original version of the language was developed by Drew McDermott, with the help of the 1998 Planning Competition committee. Fahiem Bacchus selected a subset of the original language as the language for the 2000 competition. Maria Fox and Derek Long then extended the competition language to allow for numerical variables, and concurrent execution of durational actions (original PDDL already featured a form of numerical variables, which was however never put to use). The 2002 competition language, PDDL2.1, had the following features:

PDDL2.1 is separated into different levels of expressivity. Level 1 is ADL planning, level 2 adds numeric constructs, level 3 adds durational actions. (There is also a fourth level featuring continuous effects, but this has not been put to use in IPC-3, and will not be put to use in IPC-4.)