• Planners

Competing planners

There were 29 planners registered, of which 13 were withdrawn. On this page, we provide information about the remaining 16 planners, categorized by the tracks in which they competed.

Note: Some of the planners listed below are intimately tied to the particular competition format and will not work (or not as well as possible) in a different experimental setting, e.g. when varying the timeout or memory limit. Also, we made some small modifications to some of the planners, in particular to their build processes, to get them to compile on the evaluation platform. For these reasons, we recommend that you don't use the planners as provided here for scientific experiments. Instead, we suggest that you contact the authors of the planners to provide you with a suitable planner version.

Participants are encouraged to add links to their planners' homepages here. Also, feel free to annotate your name with a link to your homepage with the wiki's edit functionality.

Note to participants: Please do not replace the planner sources given here with improved/fixed versions, because we want to provide the actual planner versions used at the competition for reference purposes. Feel free to upload post-competition planner versions in addition to the ones provided here.

Sequential Satisficing Track

1. C^3 (c3) (PDF) (source)

   • Team members: Miguel Ramírez, Nir Lipovetzky, Héctor Geffner

2. DAE-1 (dae1) (PDF) (source)

3. DAE-2 (dae2) (PDF) (source)

   • Team members: Jacques Bibai, Pierre Savéant, Marc Schoenauer, Vincent Vidal

4. DTG-Plan (dtg-plan) (PDF) (source)

   • Team members: Ruoyun Huang, Yixin Chen, Weixiong Zhang
   • Note: This is an anytime planner.

5. FF(h_a) (ffha) (PDF) (source)

6. FF(h_sa) (ffsa) (PDF) (source)

   • Team members: Emil Keyder, Héctor Geffner

7. LAMA (lama) (PDF) (source)

   • Team members: Silvia Richter, Matthias Westphal

   • Note: For experiments with LAMA, please download the latest version.

   • Note: This is an anytime planner.

8. Plan-A (plan-a) (PDF) (source)

   • Team members: Qiang Lv, Yixin Chen, Ruoyun Huang

9. SGPLAN6 (sgplan6) (PDF) (source)

   • Team members: Chih-Wei Hsu, Benjamin Wah

   • Version with some bug fixes uploaded after IPC: (source)

10. Upwards (upwards) (PDF) (source)

    • Team members: Andrew Coles, Amanda Smith
    • Note: Originally intended for the sequential optimization track. It participated in this track noncompetitively.

baseline planner: FF with a precomputation step to throw away (ignore) action costs (source)

Notes:

[1] Uses multi-threading.

[2] Participated noncompetitively.

Entries in parentheses denote PDDL features which are supported by the planner, but inputs not using the feature are preferred. For example, FF(h_a) prefers typed representations to untyped representations.

Sequential optimization track

1. CFDP (cfdp) (PDF) (source)

   • Team members: Stéphane Grandcolas, Cyril Pain-Barre

2. co-plan (co-plan) (PDF) (source)

   • Team members: Nathan Robinson, Charles Gretton, Duc Nghia Pham

3. CPT3 (cpt3) (no planner abstract) (source)

   • Team members: Vincent Vidal
   • Note: It is likely that the problems when compiling the planner on 64-bit machines for the temporal satisficing track (see below) also apply to this version, although we did not test this specifically.

4. Gamer (gamer) (PDF) (source)

   • Team members: Stefan Edelkamp, Peter Kissmann

5. hsp^*_0 (hsps0) (PDF) (source)

6. hsp^*_f (hspsf) (PDF) (source)

   • Team members: Patrik Haslum

7. Mips XXL (mips-xxl) (PDF) (source)

   • Team members: Stefan Edelkamp, Shahid Jabbar
   • Note: Performs external search (using hard disk). This needs to be taken into account when performing experiments.

8. Plan-A (plan-a) (PDF) (source)

   • Team members: Qiang Lv, Yixin Chen, Ruoyun Huang
   • Note: This planner is not guaranteed to produce globally optimal solutions, but solutions which are optimal with respect to a certain planning horizon. It participated noncompetitively.

9. Upwards (upwards) (PDF) (source)

   • Team members: Andrew Coles, Amanda Smith
   • Note: Ignores action costs and is hence not optimal in the sense of this track. It participated noncompetitively.

baseline planner: Uniform cost search, implemented as A* search with a zero heuristic (source)

Notes:

[1] Negative literals are supported only for the "=" predicate.

[2] The planners have limited support for certain ADL features, but the restrictions are complicated. We are not even sure what they are so it's just safer to assume they don't take ADL, except if that is the only formulation of a domain that is available.

[3] External search algorithm (requires disk space).

[4] Uses multi-threading. Participated noncompetitively.

[5] Does not support action costs (assumes unit cost). Participated noncompetitively.

Entries in parentheses denote PDDL features which are supported by the planner, but inputs not using the feature are preferred.

Temporal satisficing track

• Note: Some parts of the baseline planner lack number support, so that it had to be faked for the three domains which require it. The baseline planner includes some special-purpose code for this to avoid implementation effort, but identical (or, in case of modeltrain, possibly better) results can be easily obtained with a fully general implementation because these steps are easy to automate.

1. CPT3 (cpt3) (no planner abstract) (source)

   • Team members: Vincent Vidal

   • Note: We encountered problems when compiling the planner on 64-bit machines. Specifically, compilation succeeded, but the planner failed on solvable instances (e.g. Peg Solitaire #1). Hence, we recommend compiling the planner on 32-bit machines, which worked for us. (Running the 32-bit planner on a 64-bit machine did not cause problems.)

2. DAE-1 (dae1) (PDF) (source)

3. DAE-2 (dae2) (PDF) (source)

   • Team members: Jacques Bibai, Pierre Savéant, Marc Schoenauer, Vincent Vidal

4. SGPLAN6 (sgplan6) (PDF) (source)

   • Team members: Chih-Wei Hsu, Benjamin Wah

   • Version with some bug fixes uploaded after IPC: (source)

5. Temporal Fast Downward (temporal-fast-downward; tfd) (PDF) (source)

   • Team members: Gabriele Röger, Patrick Eyerich, Robert Mattmüller
   • Note: This is an anytime planner.

   • Note: The competition version of TFD is slightly outdated. If you would like to use TFD for experiments or further development, you can obtain the latest version (refactored, bug fixes) from the authors (roeger@informatik.uni-freiburg.de, eyerich@informatik.uni-freiburg.de, mattmuel@informatik.uni-freiburg.de).

6. TLP-GP (tlp-gp) (PDF) (source)

   • Team members: Frédéric Maris, Pierre Régnier

baseline planner: Metric-FF with a precomputation step that throws away action durations and a post-processing step to annotate the produced plan with suitable timestamps based on a simple scheduling algorithm (source)

Entries in parentheses denote PDDL features which are supported by the planner, but inputs not using the feature are preferred.

Temporal optimization track

This track did not take place because all planners except for one (CPT3) were withdrawn. CPT3 was run in the temporal satisficing track instead. Note when evaluating the performance of CPT3 that it is intended as an optimal planner (under conservative temporal planning semantics).

Net benefit satisficing track

This track did not take place because all planners except for one were withdrawn.

Net benefit optimization track

1. Gamer (gamer) (PDF) (source)

   • Team members: Stefan Edelkamp, Peter Kissmann

2. hsp^*_p (hspsp) (PDF) (source)

   • Team members: Patrik Haslum

   • Note: the planner only fully supports monotonic numeric fluents, so that its plans in the transport domain are not guaranteed to be optimal. (This was only noticed after the competition.)

3. Mips XXL (mips-xxl) (PDF) (source)

   • Team members: Stefan Edelkamp, Shahid Jabbar
   • Note: performs external search (using hard disk) which needs to be taken into account when performing experiments

Notes:

[1] The planners have limited support for certain ADL features, but the restrictions are complicated. We are not even sure what they are so it's just safer to assume they don't take ADL, except if that is the only formulation of a domain that is available.

[2] "There are severe limitations on the support for numeric variables. But, the planners should complain (exit with failure) if the problem contains something they can't handle." Note: obviously, the checks performed by the planners are insufficient (as evidenced by the transport domain). In particular, fluent assignment effects are silently ignored, rather than causing an exit-with-failure. There may be other omissions as well.

[3] External search algorithm (requires disk space).

Entries in parentheses denote PDDL features which are supported by the planner, but inputs not using the feature are preferred.