In parcprinter:p01, when I run blind search, I get a plan with cost 169009,
which is the same as the optimal plan found in the IPC-2008 results.
However, when I run the following:
./downward --heuristic
"hlm,hff=lm_ff_syn(lm_rhw(reasonable_orders=true,cost_type=1,lm_cost_type=1))" -
-search "lazy_wastar(hff,hlm,preferred=(hff,hlm),w=5,bound=169009)" < output
I get a better plan, with cost 163011. The actual plan is:
blackfeeder-feed-letter sheet1 (8000)
blackcontainer-toime-letter sheet1 (2000)
blackprinter-simplex-letter sheet1 front image-1 (113013)
blackcontainer-fromime-letter sheet1 (2000)
endcap-move-letter sheet1 (2000)
htmoverblack-move-letter sheet1 (17999)
down-movedown-letter sheet1 (9999)
finisher1-stack-letter sheet1 dummy-sheet (8000)
This happens on the latest code from downward-ipc (before the changes I made to
the search code for adjusted action cost).
I tried to run VAL on the plan, but got a SEGFAULT.
|