Issue467

Title Assertion in LandmarkFactoryRpgSasp fails for unsolvable tasks
Priority bug Status chatting
Superseder Nosy List erez, florian, jendrik, malte
Assigned To Keywords
Optional summary
Related to issue937.

Created on 2014-09-17.12:36:33 by florian, last changed by malte.

Summary
Related to issue937.
Messages
msg9027 (view) Author: malte Date: 2019-10-26.21:55:20
When dealing with this, we have to be careful with landmarks that are true in the initial state, because these do not need achievers. Landmarks that are true in the initial state are permitted and may potentially become useful in future revisions of the landmarks code (that implement the "sensible orderings" we discussed).

Given that this error occurs inside the lm_rhw landmark factory but is based on a symptom that might equally trip up other landmark generators, we should also check the other landmark factories for similar problems.
msg9026 (view) Author: malte Date: 2019-10-26.21:43:53
I have added a link to issue937 in the summary, which is another problem that arises when the landmark code detects unsolvability.

Like Erez suggested five short years ago, I think the most direct way to address this is to give the landmark graphs a way to mark a problem as unsolvable and then skip parts of the processing that can misbehave on unsolvable problems.

For unsolvable problems, we can skip all further processing and perhaps discard unnecessary information like the landmark nodes etc. Instead, we could have the heuristic always return infinity.

Once we have such a facility, it could also be used to address issue937.
msg7861 (view) Author: malte Date: 2018-10-03.18:43:24
I think it would be good to address this one in the not-too-far future because
it is showing up in some of our experiments (see issue837), and failed
assertions aren't great. :-)
msg3516 (view) Author: malte Date: 2014-09-23.19:36:19
I think it should remain open. I'm not sure how soon we'll be able to work on
the landmark code, but I think it should be one of our main priorities.
msg3514 (view) Author: florian Date: 2014-09-23.19:32:38
I added the comment in the master branch. Should we set this issue to deferred,
resolved or leave it open?
msg3428 (view) Author: erez Date: 2014-09-17.15:59:00
Sounds like the landmarks are detecting that the problem is unsolvable, but not 
expecting that. As part of the refactoring it might be good to add an option for 
detecting unsolvability.
msg3427 (view) Author: malte Date: 2014-09-17.14:28:44
Please add a comment to this assertion explaining where it fails. It may not be
a good idea to try fixing this locally. The landmark code needs some more global
revisions.
msg3418 (view) Author: florian Date: 2014-09-17.12:36:33
The following assertion fails for the unsolvable tasks that are created if the
translator detects unsolvability.
To reproduce, search with "astar(lmcount(lm_rhw()))" on mystery/prob07.pddl in
debug mode.

landmarks/landmark_factory_rpg_sasp.cc:109: int
LandmarkFactoryRpgSasp::min_cost_for_landmark(LandmarkNode*,
std::vector<std::vector<int> >&): Assertion `min_cost <
numeric_limits<int>::max()' failed.
History
Date User Action Args
2019-10-26 21:55:20maltesetmessages: + msg9027
2019-10-26 21:43:53maltesetmessages: + msg9026
2019-10-26 21:30:26maltesetsummary: Related to issue937.
2018-10-03 18:47:29jendriksetnosy: + jendrik
2018-10-03 18:43:24maltesetmessages: + msg7861
2014-09-23 19:36:19maltesetmessages: + msg3516
2014-09-23 19:32:38floriansetmessages: + msg3514
2014-09-17 15:59:00erezsetnosy: + erez
messages: + msg3428
2014-09-17 14:28:44maltesetstatus: unread -> chatting
messages: + msg3427
2014-09-17 12:36:56floriansetnosy: + malte
2014-09-17 12:36:33floriancreate