search

QuMuLab / pr2

FOND Planner
MIT License
4 stars 2 forks source link

Put in a stronger heuristic for the initial plan seed #3

Open haz opened 4 years ago

haz commented 4 years ago

The forest-new benchmark (problems p_10_7 and p_10_10) have initial determinized problems that are too hard for hadd / hff. This is likely also present in the sysadmin probabilistic benchmark.

haz commented 1 year ago

Another option here would be to (1) optionally take an initial plan as a seed; and (2) call a strong classical planner on a determinized version of the initial state and then use (1). Would also open the door to an investigation on the impact having a good first seed plan actually is.