Probabilistically Reusing Plans in Deterministic Planning

Inspired by probabilistic path planning, we contribute a planning approach that probabilistically balances heuristics and past plans as guidance to planning search. Our ERRT-PLAN algorithm generates multiple search branches probabilistically choosing to extend them towards the current goal or towards actions or goals of a past given plan. We have defined domains to show where current techniques could be trapped. Also, we show experimental results with a variety of domains, where we show the strengths of ERRT-PLAN. Introduction and Related Work As the complexity of planning is realized, researchers create a variety of planning approaches to try to increase the scalability horizon of their planning search. We view three different classes of domain-independent approaches to automated planning, as we sketch in Figure 1.1 Given a domainD, and a new problem P , planners can extract domainindependent heuristics to guide their search. Such “Planning from scratch” has shown to be very efficient in large classes of problems, as is the case of forward-chaining search using a combination of a well-informed heuristic, coming from a relaxed planning graph (Hoffmann & Nebel 2001). “Planning with learning” can in addition rely on training experience as compiled into domain-dependent heuristics. And “Planning with reuse” can use specific past solution plans. In this work, we focus on planning with reuse, as we assume that a similar past solution plan may be available in several situations where the new planning problem naturally just deviates from a past problem, as for example may be the case when there is a need for replanning during plan execution. Thus, according to the three classes of domain-independent guidance defined in Figure 1, the closest related works to the one presented in this paper are reuse approaches. They could We leave out from this classification all domain-dependent planning approaches, as Hierarchical Task Networks (Nau et al. 2003) or approaches based on manually defined domain-dependent heuristics (Bacchus & Kabanza 2000). Planning from scratch P D ? Domain Independent

[1]  Jörg Hoffmann,et al.  The Metric-FF Planning System: Translating ''Ignoring Delete Lists'' to Numeric State Variables , 2003, J. Artif. Intell. Res..

[2]  Dana S. Nau,et al.  SHOP2: An HTN Planning System , 2003, J. Artif. Intell. Res..

[3]  Joelle Pineau,et al.  RRT-Plan: A Randomized Algorithm for STRIPS Planning , 2006, ICAPS.

[4]  Daniel Borrajo,et al.  Using Cases Utility for Heuristic Planning Improvement , 2007, ICCBR.

[5]  Ivan Serina,et al.  Plan Stability: Replanning versus Plan Repair , 2006, ICAPS.

[6]  Bernhard Nebel,et al.  The FF Planning System: Fast Plan Generation Through Heuristic Search , 2011, J. Artif. Intell. Res..

[7]  Manuela M. Veloso,et al.  Real-time randomized path planning for robot navigation , 2002, IEEE/RSJ International Conference on Intelligent Robots and Systems.

[8]  Malte Helmert,et al.  How Good is Almost Perfect? , 2008, AAAI.

[9]  Steven M. LaValle,et al.  Randomized Kinodynamic Planning , 1999, Proceedings 1999 IEEE International Conference on Robotics and Automation (Cat. No.99CH36288C).

[10]  Manuela M. Veloso,et al.  Linkability: Examining Causal Link Commitments in Partial-order Planning , 1994, AIPS.

[11]  Jaime G. Carbonell,et al.  Derivational Analogy in PRODIGY: Automating Case Acquisition, Storage, and Utilization , 1993, Machine Learning.

[12]  James A. Hendler,et al.  A Validation-Structure-Based Theory of Plan Modification and Reuse , 1992, Artif. Intell..

[13]  Ivan Serina,et al.  Kernel functions for case-based planning , 2010, Artif. Intell..

[14]  Bernhard Nebel,et al.  Plan Reuse Versus Plan Generation: A Theoretical and Empirical Analysis , 1995, Artif. Intell..

[15]  Fahiem Bacchus,et al.  Using temporal logics to express search control knowledge for planning , 2000, Artif. Intell..