Cost-Optimal Planning in IPC 2018: Symbolic Search and Pattern Databases vs. Portfolio Planning

Anonymous

Keywords: Symbolic Search, Planning Pattern Database, Portfolio Planning

Abstract:

The optimal track is one the most exiting events in the International
Planning Competition (IPC).

In this position paper we argue that —despite of not winning
the competition— symbolic search and pattern databases
were likely the most influential planning approaches in the
latest IPC in 2018, and, in continuation to the precursor IPC
in 2014, should be considered as candidates for the current
state-of-the-art.

Five of the Top 6 planners in the 2018 competition, namely
Complementary (1 and 2), Planning-PDBs, Symbolic-Bidirectional,
and Scorpion are based on these technologies.
These planners use the same technology across all domains
and plan in one state space.

The winner of IPC 2018 with an ~ 1% lead in problems being
solved, however, is a so-called portfolio planner, consisting of
a selection of many different planners, one of which is chosen
in a classifier that was trained on a manually selected set of
benchmark instances. In about half of its successful runs, it
called the winner of the previous IPC, which in turn is based
on symbolic search. We argue on whether or not to exclude
portfolios from the IPC is possible and wanted.

Track: short (15 minutes)