Blum, A. and Furst, M. (1997). Fast planning through graph analysis. Arti cial Intelligence, 90:281 - 300.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....sequence is a goal state. Planning is sometimes refered to as domain independent planning, since the domain, as described by the available operators and their effects, is a part of the problem instance. Even though the problem has been much studied since the beginnings of artificial intelligence [13, 11, 6, 27, 29, 30, 19, 26, 22, 23, 21, 20, 24, 2], the most common formalisation is still the same as was used in the STRIPS system [11] The STRIPS formalism requires operators to have conjunctive preconditions and deterministic context independent effects. A later development, ADL [23] extends STRIPS to incorporate context dependent effects ....

Avrim L. Blum and Merrick L. Furst. Fast planning through graph analysis. Artificial Intelligence, 90, 1997.

....[24] ASPEN [5] or TALplanner [15] are knowledge intensive , relying on user provided problem decompositions, evaluation functions or search constraints 1 . 2 Action Model and Assumptions The action model we use is propositional STRIPS with extensions for time and resources. As in graphplan [3] and many other planners, the action set is enriched with a no op for each atom p which has p as its only precondition and effect. Apart from having a variable duration, a no op is viewed and treated like a regular action. 2.1 Time When planning with time each action a has a duration, dur(a) ....

A.L. Blum and M.L. Furst. Fast planning through graph analysis. Artificial Intelligence, 90(1--2):281 -- 300, 1997.

....of domain dependent search control for the HSTS planner) Planners like O Plan [20] and ASPEN [4] are also very expressive but inherently domaindependent. 2 Action Model and Assumptions The action model we use is propositional STRIPS with extensions for time and resources. Like in graphplan [2] and many other planners, the action set is enriched with a no op for each atom p which has p as its only precondition and effect. Apart from the fact that its duration is variable, a no op is viewed and treated like regular actions. 2.1 Time When planning with time each action has a duration, ....

A.L. Blum and M.L. Furst. Fast planning through graph analysis. Artificial Intelligence, 90(1--2):281 -- 300, 1997.

....practice. Tests show that the on line planner gives a major increase in robustness and reliability while not being significantly slower than perfect algorithms. 1 Introduction In planning research most projects using high level planning are theoretical planners for the STRIPS [6] domain [10] [1] [2] Most of the planning research being done on real robots[8] deals with the lower levels of planning. The restriction to the real robot domain [5] is important because assumptions made for theoretical planners such as Atomic Time 1 , Deterministic Effects 2 , Omniscience 3 and Sole ....

Avrim L. Blum, Merrick L. Furst. Fast Planning Through Graph Analysis. Artificial Intelligence, 90:281-300, 1997.

....specification of concurrent interacting actions and employ a nonmonotonic override mechanism to deduce the effects of a set of actions with conflicting effects. Finally, a number of contemporary planners can handle concurrent noninteracting actions to a certain degree examples include Graphplan (Blum Furst, 1995), and IPP (Koehler, 1998) which extends Graphplan to handle resource constraints, and more recently OBDD based planners such as MBP (Cimatti, Giunchiglia, Giunchiglia, Traverso, 1997) and UMOP (Jensen Veloso, 2000) while Knoblock (1994) provides a good discussion of the issue of ....

....could presumably be made to fit within their model, this seems not to be their main motivation. In fact, the planning algorithms they discuss deal with the issue of ensuring that parallel actions do not have negative synergistic effects, and explicitly 2. Moreover, other planning algorithms (e.g. Blum Furst, 1995; Kautz Selman, 1996) should prove amenable to extension to planning with concurrent interacting actions using similar ideas. 108 Planning with Concurrent Interacting Actions exclude the possibility of positive synergy. In our work, we abstract away from the temporal component and focus ....

[Article contains additional citation context not shown here]

Blum, A. L., & Furst, M. L. (1995). Fast planning through graph analysis. In Proceedings of the Fourteenth International Joint Conference on Artificial Intelligence, pp. 1636-- 1642 Montreal.

....of interest. While state based search techniques use knowledge of a specific initial state and a specific goal set to constrain the search process, forward search does not exploit knowledge of the goal set, nor does backward search exploit knowledge of the initial state. The Graphplan algorithm [14] can be viewed as a planning method that integrates the propagation of forward reachability constraints with backward goal informed search. We describe this approach in Section 5. Furthermore, work on partial order planning (POP) can be viewed as a slightly different approach to this form of ....

....graph, a structure computed prior to problem solving that caches reachability relationships among propo71 sitions. The graph can be consulted during the planning process in deciding which actions to insert into the plan and how to resolve threats. The Graphplan algorithm of Blum and Furst [14] attempts to blend considerations of both forward and backward reachability in a deterministic planning context. One of the difficulties with regression is that we may regress the goal region through a sequence of operators only to find ourselves in a region that cannot be reached from the initial ....

Avrim L. Blum and Merrick L. Furst. Fast planning through graph analysis. In Proceedings of the Fourteenth International Joint Conference on Artificial Intelligence, pages 1636--1642, Montreal, CA, 1995.

....concurrent. For example, different vehicles can be moved and different packages loaded and unloaded, relatively independent of one another. This kind of concurrency exists in many standard domains, and has motivated approaches such as partial order planning and Graphplan s use of parallel actions [18, 6]. However, few planners have considered the duration of concurrent actions. To create efficient plans, a planner must be able to plan a sequence of several short actions, like loading, driving and unloading a truck, in parallel with a long action, like flying an airplane between distant ....

....it is only more recently that the idea has re emerged and been applied to forward chaining [4] and SAT based [9] planners. Time and resource reasoning has been integrated in several HTN planners (e.g. SHOP [20] In classical planning, though many planners form parallel plans (e.g. Graphplan [6] and descendants) not so many treat operators with non unit durations and internal state. Several approaches in this direction, e.g. Deviser [27] Zeno [21] IxTeT [16] and TripTic [23] are based on partial ordered planning combined with temporal constraint reasoning. A more recent approach is ....

A. L. Blum and M. L. Furst, `Fast planning through graph analysis', Artificial Intelligence, 90(1--2), 281--300, (1997).

....considering the variable values or combinations of variable values that can or cannot be realized. This knowledge can be integrated into abstraction techniques by eliminating abstract states that contain unreachable variable combinations. The method we develop is based on the GRAPHPLAN algorithm [3]: the graph buildingphase of GRAPHPLAN can be viewed as performing an approximate reachability analysis that is used to prune subsequent goal regression search in a classical planning framework. However, we make certain modifications designed to deal with the DBN action representation. In ....

....If there is enough material to produce one of P4 or P5, then RdyP4 and RdyP5 might change, but the condition on P6 cannot, and again AsmP6 can have its value fixed. 3. 1 Reachability Analysis without Direct Correlations Our algorithm for structured reachability analysis in inspired by GRAPHPLAN [3], a classical planning algorithm that essentially performs a reachability analysis to construct a plan graph and then performs goal regression within that graph. Specifically, the graph building phase of GRAPHPLAN operates by alternating the construction of propositional levels and action levels. ....

Avrim L. Blum and Merrick L. Furst. Fast planning through graph analysis. In Proceedings of the Fourteenth International Joint Conference on Artificial Intelligence, pages 1636--1642, Montreal, 1995.

....examples is of 88 2 calculation times on a Sparc 10 machine Ex I X T E T I X T E T Gain 1 0.5 s 0.2 s 60 2 2.4 s 0.4 s 83 3 14 s 1 s 93 4 71 s 6.5 s 91 5 94.9 s 4.1 s 96 6 124 s 3.5 s 97 7 178.8 s 4. 3 s 98 We have compared the results we have obtained to those that Graphplan [1] obtains on the same examples. An exact comparison isn t possible, as I X T E T allows temporal planning, resource management, dynamic domains and gives partially ordered plans. We gave similar but reduced problems to Graphplan. Figure 8 shows a comparative graph of results obtained with I X T E T ....

M. Furst A. Blum. Fast Planning Through Graph Analysis. Artificial Intelligence, 90:281--300, 1997.

....considering the variable values or combinations of variable values that can or cannot be realized. This knowledge can be integrated into abstraction techniques by eliminating abstract states that contain unreachable variable combinations. The method we develop is based on the GRAPHPLAN algorithm [3]: the graph buildingphase of GRAPHPLAN can be viewed as performing an approximate reachability analysis that is used to prune subsequent goal regression search in a classical planning framework. However, we make certain modifications designed to deal with the DBN action representation. In ....

....If there is enough material to produce one of P4 or P5, then RdyP4 and RdyP5 might change, but the condition on P6 cannot, and again AsmP6 can have its value fixed. 3. 1 Reachability Analysis without Direct Correlations Our algorithm for structured reachability analysis in inspired by GRAPHPLAN [3], a classical planning algorithm that essentially performs a reachability analysis to construct a plan graph and then performs goal regression within that graph. Specifically, the graph building phase of GRAPHPLAN operates by alternating the construction of propositional levels and action levels. ....

Avrim L. Blum and Merrick L. Furst. Fast planning through graph analysis. In Proceedings of the Fourteenth International Joint Conference on Artificial Intelligence, pages 1636--1642, Montreal, 1995.

No context found.

Blum, A. and Furst, M. (1997). Fast planning through graph analysis. Arti cial Intelligence, 90:281 - 300.

No context found.

A. Blum and M. Furst, "Fast planning through graph analysis," in Proceedings of the Fourteenth International Joint Conference on Artificial Intelligence, Montreal, 1995, pp. 1636 -- 1642.

No context found.

A.L. Blum and M.L. Furst. Fast planning through graph analysis. Artificial Intelligence, 90:281--300, 1997. 7

No context found.

, 719--726. Blum, A. L., & Furst, M. L. (1995). Fast planning through graph analysis. In Proceedings of the Fourteenth International Joint Conference on Artificial Intelligence, pp. 1636--

No context found.

, 719#726. Blum, A. L., & Furst, M. L. #1995#. Fast planning through graph analysis. In Proceedings of the Fourteenth International Joint Conference on Arti#cial Intelligence, pp. 1636#

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC