From 38963c7fdfa57aa9b046c9dd6a3d2e9f962f4303 Mon Sep 17 00:00:00 2001 From: Enrico Scala Date: Thu, 30 Jan 2020 15:36:26 +0100 Subject: [PATCH 1/2] A few edits on the description of ENHSP --- docs/reference/planners/ENHSP/main.md | 19 +++++++++++++------ 1 file changed, 13 insertions(+), 6 deletions(-) diff --git a/docs/reference/planners/ENHSP/main.md b/docs/reference/planners/ENHSP/main.md index 2e89126..829c2c8 100644 --- a/docs/reference/planners/ENHSP/main.md +++ b/docs/reference/planners/ENHSP/main.md @@ -17,14 +17,14 @@ Paper: [Interval-Based Relaxation for General Numeric Planning](https://pdfs.sem ENHSP, which stands for Expressive Numeric Heuristic Planner is a forward heuristic search planner. It supports the following -1. Classical Planning +1. Satisficing and Optimal Numeric Planning 2. Numeric Planning with linear and non-linear expressions 3. Planning with discretised autonomous processes and events 4. Global constraints, which are the analogous of always constraints of PDDL -5. discrete events -6. universal and existential quantification in formulas (action precondition, constraints and goals) -7. negative preconditions -8. novel heuristics based on landmarks and search techniques for planning with autonomous processes +5. Discrete events +6. Universal and existential quantification in formulas (action precondition, constraints and goals) +7. Negative preconditions +8. Novel heuristics based on landmarks and search techniques for planning with autonomous processes ## Support ENHSP requires `:typing` in all domains it runs on, which makes it incompatible with [Eviscerator](https://www.github.com/nergmada/eviscerator). No support table has been generated. @@ -33,4 +33,11 @@ ENHSP requires `:typing` in all domains it runs on, which makes it incompatible For more information on how to download, compile and use ENHSP, visit its [GitLab page here](https://gitlab.com/enricos83/ENHSP-Public). ## Additional Notes -ENHSP has support for non-linear continuous effects assuming they are monotonic. \ No newline at end of file +- ENHSP grounding is not optimised; this can become a bottleneck when your actions involve many many parameters +- ENHSP does not support durative actions +- Optimal planning is only supported for what is called simple numeric planning, i.e., only increase/decrease effects of constants, and only linear numeric conditions. Objective function can be linear combination of numeric fluents. If no objective function is given, the planner minimises the number of actions. + +## Other Relevant Papers ENHSP is built on +[Heuristics for Numeric Planning via Subgoaling](https://www.ijcai.org/Proceedings/16/Papers/457.pdf) [Scala, E., Haslum, P. and Thiébaux, S.] +[Landmarks for Numeric Planning Problems](https://www.ijcai.org/Proceedings/2017/0612.pdf) [Scala, E. Haslum, P. Thiebaux, S. and Magazzeni D.] +[Effect-Abstraction Based Relaxation for Linear Numeric Planning](https://www.ijcai.org/Proceedings/2018/0665.pdf) [Li, D. Scala, E. Haslum, P. and Bogomolov, S] \ No newline at end of file From 19bb0f3aa771f29950bd7cf61dabda412749ae8b Mon Sep 17 00:00:00 2001 From: Enrico Scala Date: Thu, 6 Feb 2020 16:50:16 +0100 Subject: [PATCH 2/2] A few edits in ENHSP description --- docs/reference/planners/ENHSP/main.md | 21 +++++++++++---------- 1 file changed, 11 insertions(+), 10 deletions(-) diff --git a/docs/reference/planners/ENHSP/main.md b/docs/reference/planners/ENHSP/main.md index 9203ee5..e6a71a5 100644 --- a/docs/reference/planners/ENHSP/main.md +++ b/docs/reference/planners/ENHSP/main.md @@ -15,27 +15,28 @@ Year Published: 2016 Paper: {% paper_link Interval-Based Relaxation for General Numeric Planning planners/enhsp 1 %} [Scala, E. Haslum, P. Thiebaux, S. Ramirez, M.] -ENHSP, which stands for Expressive Numeric Heuristic Planner is a forward heuristic search planner. It supports the following +ENHSP, which stands for Expressive Numeric Heuristic Planner is a forward heuristic search planner. It supports PDDL2.1 and discretised PDDL+, in particular: -1. Satisficing and Optimal Numeric Planning -2. Numeric Planning with linear and non-linear expressions +1. Satisficing and Optimal Simple Numeric Planning (see below) +2. Satisficing Numeric Planning with linear and non-linear expressions 3. Planning with discretised autonomous processes and events 4. Global constraints, which are the analogous of always constraints of PDDL -5. Discrete events -6. Universal and existential quantification in formulas (action precondition, constraints and goals) -7. Negative preconditions -8. Novel heuristics based on landmarks and search techniques for planning with autonomous processes +5. Good support for general quantifier-free formulas +6. Universal and existential quantification in formulas (action precondition, constraints and goals). This is hardly experimental. +7. A variety of heuristics (e.g., hadd, aibr, haddabs, landmarks) and search mechanisms (e.g., wastar, idastar, dfsbnb) ## Support ENHSP requires `:typing` in all domains it runs on, which makes it incompatible with [Eviscerator](https://www.github.com/nergmada/eviscerator). No support table has been generated. ## Downloading and Compiling ENHSP -For more information on how to download, compile and use ENHSP, visit its [GitLab page here](https://gitlab.com/enricos83/ENHSP-Public). +For more information on how to download, compile and use ENHSP, visit its [GitLab page here](https://gitlab.com/enricos83/ENHSP-Public). The planner is mantained by Enrico Scala (enricos83@gmail.com for any question). ## Additional Notes -- ENHSP grounding is not optimised; this can become a bottleneck when your actions involve many many parameters +- ENHSP grounding is not optimised; this can become a bottleneck when your action schema induce several even trivial groundings. - ENHSP does not support durative actions -- Optimal planning is only supported for what is called simple numeric planning, i.e., only increase/decrease effects of constants, and only linear numeric conditions. Objective function can be linear combination of numeric fluents. If no objective function is given, the planner minimises the number of actions. +- Conditional effects are supported but are highly experimental +- Optimal planning is only supported for what is called simple numeric planning (including classical planning), i.e., only increase/decrease effects of constants, and only linear numeric conditions. Objective function can be linear combination of numeric fluents. If no objective function is given, the planner minimises the number of actions. +- ENHSP is designed to experiment on numeric reasoning. If your problem is a purely classical planning problem, it is better to use a classical planner like FD, or even FF. ## Other Relevant Papers ENHSP is built on [Heuristics for Numeric Planning via Subgoaling](https://www.ijcai.org/Proceedings/16/Papers/457.pdf) [Scala, E., Haslum, P. and Thiébaux, S.]