PDDL Problems mismatch with PutTaskExtended File?

[CatherineWong]

Hi there --

I'm trying to replicate using the FFPlanner on the dataset to just make sure we can recover the original PDDL actions using the domain definition.

I'm a little confused because the data described at https://github.com/askforalfred/alfred/tree/master/data (which lists the data/train/task_type-object-movableReceptacle-receptacle-sceneNum/trial_ID/problem_x.pddl that I assume is the PDDL initial state) seems to contain a bunch of predicates that aren't in that ground truth file here: https://github.com/askforalfred/alfred/blob/master/gen/planner/domains/PutTaskExtended_domain.pddl, like canContain, so currently you can't actually run the FFPlanner on those problem files + the provided domain file and (I think) get a PDDL plan?

[CatherineWong]

Happy to provide an example file, but I think I'm looking at the right ones?

I'm additionally confused because the canContain predicate does appear in this somewhat different alfworld domain, but this one has some other changes: https://github.com/alfworld/alfworld/blob/master/alfworld/data/alfred.pddl

[MohitShridhar]

@CatherineWong, the PDDL planner needs two things: state and domain descriptions. The domain description, i.e. PutTaskExtended_domain.pddl describes the transition dynamics of the world. But you also need a state description, i.e. problem_x.pddl, that describes the current state of the environment. This state is generate here with the state_to_pddl function.

Hope this helps!

[MohitShridhar]

Oh yeah, sorry, the ALFWorld domain was modified to deal with Textworld issues. The domain file is different.

[MohitShridhar]

Closing due to inactivity.