Optimising Partial-Order Plans Via Action Reinstantiation

IJCAI, pp. 4143-4151, 2020.

Cited by: 0|Bibtex|Views6|DOI:https://doi.org/10.24963/ijcai.2020/573
EI
Other Links: dblp.uni-trier.de|academic.microsoft.com
Weibo:
This paper presented a practical technique for plan optimisation through the modification of both ordering and variable binding constraints

Abstract:

This work investigates the problem of optimising a partial-order plan’s (POP) flexibility through the simultaneous transformation of its action ordering and variable binding constraints. While the former has been extensively studied through the notions of deordering and reordering, the latter has received much less attention. We show that...More

Code:

Data:

0
Introduction
  • This work investigates the modification of objects and resources in a plan to achieve higher flexibility w.r.t. action orderings.
  • The aim is to achieve a least commitment approach to planning [Weld, 1994] that maximises executiontime flexibility by delaying decisions regarding action ordering and resource utilisation for as long as possible.
  • There are soil and rock samples at waypoints W2 and W3, resp., and both rovers begin at W1.
Highlights
  • This work investigates the modification of objects and resources in a plan to achieve higher flexibility w.r.t. action orderings
  • While the optimisation of action ordering has been extensively studied through the notions of plan deordering and plan reordering, both from a theoretical [Backstrom, 1998; Aghighi and Backstrom, 2017] and practical [Kambhampati and Kedar, 1994; Muise et al, 2016; Siddiqui and Haslum, 2012; Say et al, 2016] perspective, the optimisation of resources that are used in the course of executing a plan has received much less attention
  • Complexity Results Deciding whether a partial-order plan has a reinstantiated de/reordering with fewer than k ordering constraints is NP-complete, and the optimisation problem of finding a minimum reinstantiated de/reordering cannot be approximated within a constant factor: Theorem 1
  • This paper presented a practical technique for plan optimisation through the modification of both ordering and variable binding constraints
  • Results show that in 52% of cases, MRR provides a flex increase of 20% over the explanation-based order generalisation baseline in 22m, with results varying by domain
  • Whether MRR is preferable to less costly methods such as explanation-based order generalisation depends on the application
Methods
  • The authors answer the question above by comparing optimised plans produced by the Loandra MAXSAT solver [Berg et al, 2019] using the MD, MR, MRD and MRR encodings.

    The authors compare the MAXSAT-based techniques with the explanation-based order generalisation (EOG) deordering technique of Kambhampati and Kedar [1994].
  • If P = O, θ, ≺ is a POP VP is a validation stucture of P iff VP ⊆ LP and there exists a total order ≺⊆≺ s.t. if op, q(t), oc, q(u) ∈ VP and op, q(t ), oc, q(u) ∈ LP op ≺ op
  • This validation structure serves as a deordering heuristic: Definition 19.
Results
  • POP has a reinstantiated de/reordering with fewer than k ordering constraints is NP-complete, and the optimisation problem of finding a minimum reinstantiated de/reordering cannot be approximated within a constant factor: Theorem 1.
  • Hardness is by reduction from the NP-complete [Backstrom, 1998] decision problem of minimum deordering, which asks whether P has a deorder with < k ordering constraints.
Conclusion
  • This paper presented a practical technique for plan optimisation through the modification of both ordering and variable binding constraints.
  • Results show that in 52% of cases, MRR provides a flex increase of 20% over the EOG baseline in 22m, with results varying by domain.
  • Whether MRR is preferable to less costly methods such as EOG depends on the application.
  • If offline preprocessing time is available and execution-time flexibilty is paramount, or the application domain is one where reinstantiation can quickly and consistently improve plan flexibility, reinstantiation is clearly worthwhile
Summary
  • Introduction:

    This work investigates the modification of objects and resources in a plan to achieve higher flexibility w.r.t. action orderings.
  • The aim is to achieve a least commitment approach to planning [Weld, 1994] that maximises executiontime flexibility by delaying decisions regarding action ordering and resource utilisation for as long as possible.
  • There are soil and rock samples at waypoints W2 and W3, resp., and both rovers begin at W1.
  • Methods:

    The authors answer the question above by comparing optimised plans produced by the Loandra MAXSAT solver [Berg et al, 2019] using the MD, MR, MRD and MRR encodings.

    The authors compare the MAXSAT-based techniques with the explanation-based order generalisation (EOG) deordering technique of Kambhampati and Kedar [1994].
  • If P = O, θ, ≺ is a POP VP is a validation stucture of P iff VP ⊆ LP and there exists a total order ≺⊆≺ s.t. if op, q(t), oc, q(u) ∈ VP and op, q(t ), oc, q(u) ∈ LP op ≺ op
  • This validation structure serves as a deordering heuristic: Definition 19.
  • Results:

    POP has a reinstantiated de/reordering with fewer than k ordering constraints is NP-complete, and the optimisation problem of finding a minimum reinstantiated de/reordering cannot be approximated within a constant factor: Theorem 1.
  • Hardness is by reduction from the NP-complete [Backstrom, 1998] decision problem of minimum deordering, which asks whether P has a deorder with < k ordering constraints.
  • Conclusion:

    This paper presented a practical technique for plan optimisation through the modification of both ordering and variable binding constraints.
  • Results show that in 52% of cases, MRR provides a flex increase of 20% over the EOG baseline in 22m, with results varying by domain.
  • Whether MRR is preferable to less costly methods such as EOG depends on the application.
  • If offline preprocessing time is available and execution-time flexibilty is paramount, or the application domain is one where reinstantiation can quickly and consistently improve plan flexibility, reinstantiation is clearly worthwhile
Tables
  • Table1: For each domain, fEOG is the mean flex produced by EOG, and for each encoder, C is the % of plans for which a (satisficing or optimal) solution was found, T is the mean run time in minutes, and ∆EOG is the mean % flex increase over EOG. All flex values are computed from plans for which every encoder found a solution, and all ∆EOG values are statistically significant with p < 0.01 as calculated from a single-tailed, paired t-test. Empty cells indicate no data, or no significant difference
  • Table2: Plan count (nP ) and coverage (C) by encoding size (in number of clauses) for MR, MRD and MRR. Note that the encoding size is a log scale, each size range being twice that of the previous
Download tables as Excel
Reference
  • [Aghighi and Backstrom, 2017] Meysam Aghighi and Christer Backstrom. Plan reordering and parallel execution - A parameterized complexity view. In Proc. of AAAI, pages 3540–3546, 2017.
    Google ScholarLocate open access versionFindings
  • [Backstrom, 1998] Christer Backstrom. Computational aspects of reordering plans. Journal of Artificial Intelligence Research, 9:99–137, 1998.
    Google ScholarLocate open access versionFindings
  • [Bercher and Olz, 2020] Pascal Bercher and Conny Olz. POP ≡ POCL, right? Complexity results for partial order (causal link) makespan minimization. In Proc. of AAAI, 2020.
    Google ScholarLocate open access versionFindings
  • [Berg et al., 2019] Jeremias Berg, Emir Demirovic, and Peter J. Stuckey. Core-boosted linear search for incomplete MaxSAT. In Proc. of the 16th International Conference on the Integration of Constraint Programming, AI, and Operations Research, pages 39–56, 2019.
    Google ScholarLocate open access versionFindings
  • [Brachman and Levesque, 2004] Ronald Brachman and Hector Levesque. Knowledge Representation and Reasoning. Morgan Kaufmann Publishers, USA, 2004.
    Google ScholarFindings
  • [Fox and Long, 1999] Maria Fox and Derek Long. The detection and exploitation of symmetry in planning problems. In Proc. of IJCAI, pages 956–961, 1999.
    Google ScholarLocate open access versionFindings
  • [Joslin and Roy, 1997] David Joslin and Amitabha Roy. Exploiting symmetry in lifted CSPs. In Proc.of AAAI, page 197–202, 1997.
    Google ScholarLocate open access versionFindings
  • [Kambhampati and Kedar, 1994] Subbarao Kambhampati and Smadar Kedar. A unified framework for explanationbased generalization of partially ordered and partially instantiated plans. Artificial Intelligence, 67(1):29–70, 1994.
    Google ScholarLocate open access versionFindings
  • [Kambhampati and Nau, 1996] Subbarao Kambhampati and Dana S. Nau. On the nature and role of modal truth criteria in planning. Artifical Intelligence, 82(1-2):129–155, 1996.
    Google ScholarLocate open access versionFindings
  • [Korhonen et al., 2017] Tuukka Korhonen, Jeremias Berg, Paul Saikko, and Matti Jarvisalo. MaxPre: An extended MaxSAT preprocessor. In Proc. of the 20th Int. Conference on Theory and Applications of Satisfiability Testing, volume 10491 of LNCS, pages 449–456.
    Google ScholarLocate open access versionFindings
  • [Lifschitz, 1987] Vladimir Lifschitz. On the semantics of STRIPS. In Michael Georgeff and Amy Lansky, editors, Reasoning about Actions and Plans, pages 1–9. Morgan Kaufmann, San Mateo, CA, 1987.
    Google ScholarLocate open access versionFindings
  • [Lipovetzky and Geffner, 2017] Nir Lipovetzky and Hector Geffner. Best-first width search: Exploration and exploitation in classical planning. In Proc. of AAAI, pages 3590– 3596, 2017.
    Google ScholarLocate open access versionFindings
  • [Muise et al., 2016] Christian Muise, J. Christopher Beck, and Sheila A. McIlraith. Optimal partial-order plan relaxation via MaxSAT. Journal of Artificial Intelligence Research, 57:113 – 149, 2016.
    Google ScholarLocate open access versionFindings
  • [Nguyen and Kambhampati, 2001] XuanLong Nguyen and Subbarao Kambhampati. Reviving partial order planning. In Proc. of IJCAI, pages 459–464, 2001.
    Google ScholarLocate open access versionFindings
  • [Olz and Bercher, 2019] Conny Olz and Pascal Bercher. Eliminating redundant actions in partially ordered plans A complexity analysis. In Proc. of ICAPS, pages 310–319, 2019.
    Google ScholarLocate open access versionFindings
  • [Pochter et al., 2011] Nir Pochter, Aviv Zohar, and Jeffrey S. Rosenschein. Exploiting problem symmetries in statebased planners. In Proc. of AAAI, 2011.
    Google ScholarLocate open access versionFindings
  • [Richter and Westphal, 2010] Silvia Richter and Matthias Westphal. The LAMA planner: Guiding cost-based anytime planning with landmarks. Journal of Artificial Intelligence Research, 39(1):127–177, 2010.
    Google ScholarLocate open access versionFindings
  • [Riddle et al., 2016] Pat Riddle, Jordan Douglas, Mike Barley, and Santiago Franco. Improving performance by reformulating PDDL into a bagged representation. In Proc. of the 8th Workshop on Heuristics and Search for Domainindependent Planning, pages 28–36, 2016.
    Google ScholarLocate open access versionFindings
  • [Rintanen, 2003] Jussi Rintanen. Symmetry reduction for SAT representations of transition systems. In Enrico Giunchiglia, Nicola Muscettola, and Dana S. Nau, editors, Proc. of ICAPS, pages 32–40, 2003.
    Google ScholarLocate open access versionFindings
  • [Rintanen, 2010] Jussi Rintanen. Heuristics for planning with SAT. In David Cohen, editor, Proc. of CP, pages 414–428, 2010.
    Google ScholarLocate open access versionFindings
  • [Say et al., 2016] Buser Say, Andre A. Cire, and J. Christopher Beck. Mathematical programming models for optimizing partial-order plan flexibility. In Proc. of ECAI, pages 1044–1052, 2016.
    Google ScholarLocate open access versionFindings
  • [Shlyakhter, 2007] Ilya Shlyakhter. Generating effective symmetry-breaking predicates for search problems. Discrete Applied Mathematics, 155(12):1539–1548, 2007.
    Google ScholarLocate open access versionFindings
  • [Siddiqui and Haslum, 2012] Fazlul Hasan Siddiqui and Patrik Haslum. Block-structured plan deordering. In Proc. of the 25th Australasian Joint Conference on Advances in Artificial Intelligence, pages 803–814, 2012.
    Google ScholarLocate open access versionFindings
  • [Waters et al., 2018] Max Waters, Bernhard Nebel, Lin Padgham, and Sebastian Sardina. Plan relaxation via action debinding and deordering. In Proc. of ICAPS, pages 278–287, 2018.
    Google ScholarLocate open access versionFindings
  • [Weld, 1994] Daniel S. Weld. An introduction to least commitment planning. AI Magazine, 15(4):27–61, 1994.
    Google ScholarLocate open access versionFindings
Full Text
Your rating :
0

 

Tags
Comments