search

KCL-Planning / ROSPlan

The ROSPlan framework provides a generic method for task planning in a ROS system.
http://kcl-planning.github.io/ROSPlan
BSD 2-Clause "Simplified" License
357 stars 158 forks source link

Trouble with update of ROSPLAN #142

Open Samcheok opened 6 years ago

Samcheok commented 6 years ago

Hello,

We notice that there is a difference between ROSPLAN version (old) downloaded in Oct. 1st and that (new version) downloaded in Oct 15. Thus so, for the same plan, Esterel plan graph generated by old version is different from that generated by new version. The only change we are aware of is VALVisitorOperator.cpp file in rosplan_knowledge_base/src.

Here is a sample: recent.pdf old.pdf

0.0003: (fm_hub_to_rack f_lift1 r_st1 h_st) [8.0000] 0.0012: (fm_hub_to_rack f_lift2 r_st3 h_st) [4.0000] 0.0022: (fm_hub_to_rack f_lift3 r_st2 h_st) [4.0000] 4.0015: (fr_loading_to_forklift f_lift2 fr2 r_st3) [1.0000] 4.0025: (fr_loading_to_forklift f_lift3 fr3 r_st2) [1.0000] 5.0017: (fm_rack_to_hub f_lift2 r_st3 h_st) [4.0000] 5.0027: (fm_rack_to_hub f_lift3 r_st2 h_st) [4.0000] 8.0005: (fr_loading_to_forklift f_lift1 fr1 r_st1) [1.0000] 9.0007: (fm_rack_to_hub f_lift1 r_st1 h_st) [8.0000] 17.001: (fr_unloading_to_uld_from_fk f_lift1 h_st fr1 uld1 u_0 u_1 a) [1.0000] 18.002: (fr_unloading_to_uld_from_fk f_lift2 h_st fr2 uld1 u_1 u_2 a) [1.0000] 19.003: (fr_unloading_to_uld_from_fk f_lift3 h_st fr3 uld1 u_2 u_3 a) [1.0000]

Actually, old version generates correct graph for the plan.

Any comments would be appreciated very much.

Kyeongseok

m312z commented 6 years ago

Hello Kyeonseok,

Thanks for this! I think I understand the error.

Right now ROSPlan is handling all actions as durative actions. However, this has a negative effect on instant actions, which Gerard (@gerardcanal) spotted in the last pull request. See this for more information: https://github.com/KCL-Planning/ROSPlan/pull/132#discussion_r224075445

The PDDL to Esterel plan parser will need a thorough upgrade to deal with instant actions.

  1. Those actions should be encoded as a single node, with both conditions and effects attached.
  2. This will necessitate a change in the dispatcher, to ensure that "action enabled" and "action achieved" feedback messages do not trigger incorrect nodes.
  3. Finally, the checking for mutual exclusions should be updated to consider actions which do not have invariants between pairs of nodes.
  4. At the same time, this algorithm can be made more efficient.

I'll add this issue to the project board.

Michael

gerardcanal commented 6 years ago

Hi Kyeongseok, In your case, I would suggest that you either undo the change specified by @m312z in your local copy or switch to branch kinetic, which doesn't have that change and should be more stable.

Is your domain using PDDL durative-actions or you have standard PDDL actions?

Thanks Gerard

m312z commented 6 years ago

Hello Gerard,

Looking at the plan, it looks like it could be a mix of both. If it is durative actions only, then I am wrong and there is another bug at play.

Michael

jqs7328 commented 6 years ago

Dear Michael and Gerard,

First of all, thank you for your quick responses.

I'm working with Kyeongseok posted this issue on the same project.

Our domain deals with only durative actions - last August we noticed that instant actions did not work on both dispatcher and esterel plan of ROSPLAN, and remodeled our domain with durative actions, only.

Thanks!

Ji-Ae Shin