On this page, we provide the original source code of competing planners, i.e. the actual code that was used during the competition. Moreover, here you will also find bug description and updated scource code of planners in which issues were identified after the competition.
Below, the original source code of planners, divided by tracks.
- Sequential Satisficing
- Sequential Optimal
- Sequential Multicore
- Sequential Agile
- Temporal Satisficing
This planner took part in Sequential Optimal track. Below the bug description and here the updated code.
- Unreachable conditional effects were handled incorrectly in src/search/conditional_effect_lm_cut_heuristic.cc. The patch just make the code ignore them.
- Our decision to switch off one of our components during runtime was based on a ratio that was computed by a division. Unfortunately, this was an integer division so the result was always 0 or 1. The patch fixes this by casting the input to a double first.
- In eager_search.cc we just added some debug output that notifies the user of the choice that is done (the choice itself is not influenced).
This planner took part in Sequential Optimal track.
- RIDA* needs to have "conditional-effects" as a explicitly listed requirement in the problem's domain pddl file. This probably affected performance in CaveDiving and Maintenance domains.
DPMPlan -- MIPlan -- NuCeLaR
The planners took part in Sequential Optimal/Satisficing/Multicore tracks.
- Planners store temporary files in src/parser subdirectory. If not correctly removed, such files might lead planners to solve a different problem (one provided before) rather than the current one. Take care to remove such files before every run!
- We believe that this bug affected one domain in Seq-Sat, two domains in Set-Mco and several domains in Seq-Opt.
This planner took part in Sequential Satisficing/Agile tracks. Below the bug description and here the updated code.
- A bug was found in the translator. In some cases, that appear in most instances of CaveDiving domain, conditional effects were not fully compiled away.
IBaCoP -- IBaCoP2
The planners took part in Sequential Satisficing/Agile/Multicore tracks. Fixed code and extended description will be available soon.
- Planners store temporary files in a (still unkown) subdirectory. If not correctly removed, such files might lead planners to solve a different problem (one provided before) rather than the current one.
- A memory-related bug in the features extraction phase.
- A routes-related bug (not sure this affected the planners on the IPC benchmarks).
BFS_f and SIW
- Planners were computing h^2 without exploiting it. This issue caused the affected planners not even start searching in the Agile track.