A. Barret and D. S. Weld. Partial-order planning: Evaluating possible efficiency gains. Artificial Intelligence, 67(1):71--112, 1994.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....0 10 20 30 40 50 60 70 80 90 100 of problems Figure 6: NODES: STRIPS world Tyre World Results We have also conducted experiments in the Tyre World domain created by Russell [15] as introduced above. The tyre change problem as formulated could be termed laboriously serialisable [1], since there are very few orders which permit each goal to be achieved and then preserved whilst the remaining goals are achieved. Barrett and Weld quote a figure of 6 hours for solution time for early experiments in the domain with this problem [1, pp 99] while the best combination of their own ....

....effect of sub goals [9] and introduced a taxonomy of subgoals and their relative complexity which differentiated between independent , serialisable and non Gamma serialisable sub goals. This has recently been refined by Barrett and Weld to include trivially and laboriously serialisable sub goals [1]. Barrett and Weld conclude that solving laboriously serialisable goal problems is intractable based on their assumption that good serialisations cannot be predicted before planning commences. Yet we challenge this conclusion since we have shown that it is possible to identify type (i) and type ....

A. Barrett and D. S. Weld. Partial-Order planning: evaluating possible efficiency gains. Artificial Intelligence, 67:71--112, 1994.

....of efficiency within general planners can be addressed by the use of a knowledge rich domain capture coupled with compilation tools which operationalise implicit knowledge, and that this can greatly improve performance. Barrett and Weld argue that domain theory compilation can give great benefits [1] while McCluskey and Porteous [7, 9] present a method of encoding domain specifications plus a compilation process which can greatly reduce the search explosion in toy worlds. We hope to build on this previous work by extending current tools to input a more expressive representation language, ....

A. Barrett and D. S. Weld. Partial-Order planning: evaluating possible efficiency gains. Artificial Intelligence, 67, 1994.

....therefore backtracking is unnecessary. The second assumption not only helps guarantee that one can find a meta level path to the solution without backtracking, it also guarantees that the search is systematic[16] Thus, if the other two assumptions hold then for any problem solver (e.g. SNLP[1]) which is systematic, one can identify which of its search heuristics to modify in time linear with respect to the size of the supplied solution. 5.4.2 Computational Complexity of Compute Modifications There are two computationally expensive activities in the Compute Modifications component. ....

A. Barrett and D. Weld. Partial-order planning: Evaluating possible efficiency gains. Artificial Intelligence, 67(1), 1994.

.... the time required to encode the same problem) 4 Experimental results We empirically compared LPSP to the PAS approach on problem instances from three well known planning domains: the blocks world and logistics domains [Kautz and Selman, 1996; Ernst et al. 1997] and the artificial D1S1 domain [Barrett and Weld, 1994] . For the PAS approach, the performance depends on three components: i) the SAT encoding of the planning instances, ii) polynomial simplification algorithms which are applied to the SAT formula before general SAT solvers are invoked, and (iii) the SAT algorithm used to solve the simplified ....

A. Barrett and D. Weld. Partial order planning: Evaluating possible efficiency gains. Artificial Intelligence, 67(1):71--112, 1994.

....of this algorithm. iii) The generalized algorithm facilitates the separation of important ideas underlying individual algorithms from brand names , and thus provides a rational basis for understanding the tradeoffs offered by various planners. We will 2 The work of Barrett and Weld [2] as well as Minton et al. 26,27] are certainly steps in the right direction. However, they do not tell the full story since the comparison there was between a specific partial order and total order planner. The comparison between different partial order planners itself is still largely ....

....not be Auxiliary Constraints Monotonic (Auxiliary) Constraints Non Monotonic (Auxiliary) Constraints Interval Preservation Constraints Point Truth Constraints Fig. 5. The relation between the various types of auxiliary constraints satisfied with respect to a different mapping, A = tl [2], t2 [6] since [3] o3 deletes p) Similarly, a point truth constraint 9 t is said to be satisfied by a ground operator sequence under A , if and only if (i) either c is true in the initial state, and is preserved by every action of occurring before A (t) or (ii) c is made ....

[Article contains additional citation context not shown here]

A. Barrett and D. Weld. Partial Order Planning: Evaluating Possible Efficiency Gains. Artificial Intelligence, Vol. 67, No. 1, 1994.

....and on some new taken from the AIPS 98 planning competition. Our planner has proved to be faster in all of the cases, finding also in most (but not all) of the cases shorter solutions. 1 Introduction It is broadly accepted that plan space planning is more efficient than state space planning [1, 10]. Recent results have shown that the performance of planners like GRAPHPLAN and SATPLAN seems to be even better [2, 8] However, experience suggests that state space planners with appropriate heuristic functions outperform all the others. For example, by using domain dependent heuristics, very ....

Barett A., Weld D.S.: Partial order planning: Evaluating possible efficiency gains. J. Artificial Intelligence, 67 (1994) 71-112

....restrictions I, A and O in each and every experiment. Since Sasplan must perform these checks in a real world planning situation, it is reasonable that the tests are performed also in the experimental setting. 5.0. 1 The D 1 S 1 Domain The D 1 S 1 domain was invented by Barrett and Weld [7]. A D 1 S 1 domain of size n, n 2, consists of 2n propositional atoms I 1 ; I n and G 1 ; G n together with n operators A 1 ; A n . In Strips style notation, the operators are defined as follows: action A 1 : precond fI 1 g : add fG 1 g) action A i : precond ....

....with a state variable with domain f0; 1g and modifiying the operators accordingly. It is easily verifiable that the resulting SAS instance satisfies restricitons I, A and O. Blum and Furst [8] showed that Graphplan is competitive with the best performance reported on the D 1 S 1 domain [7]. In Table 5, we can see that Sasplan outperforms Graphplan on this domain. 5.0.2 The Tunnel Example The tunnel example is a toy domain that has been used in control theory. It assumes a tunnel (see Figure 9) divided into n sections such that the light 27 can be switched on and off ....

A. Barrett and D. S. Weld, Partial-order planning: Evaluating possible efficiency gains, Artif. Intell. 67(1) (1994) 71--112.

....have fallen out of favor in the AI planning community. This is due to the fact that there are alternate spaces in which searching for plans is generally more effective. Partial order planners that search in the space of partially ordered plans have been shown to possess a number of advantages [9, 41]. And more recently planners that search over GRAPHPLAN graphs [13] or over models of propositional theories representing the space of plans [32] have been shown to be quite effective. Nevertheless, as we will demonstrate, the combination of domain specific search control information, expressed ....

A. Barrett and D.S. Weld. Partial-order planning: evaluating possible efficiency gains. Artificial Intelligence, 67(1):71--112, 1994.

....1.1 The Problems of Knowledge Sparse Planning Research into classical planning in Artificial Intelligence has for decades concentrated on theoretical issues of planning algorithms. Recent work has concentrated on, for example, the relative performance of total order vs partial order planners [3, 57, 44], the inherent computational complexity of plan generation [6, 23] extending the expressiveness of the classical 1 model [25] and general, theoretical frameworks for planning engines [31] This research has been dominated by the use of the literal or proposition as the basic level of ....

....of the classical generative planner and to compare the efficiency tradeoffs between linear and partial order planners. Initial results with systematic causal link partial order planners suggested that they were more efficient than linear planners, in part as a result of reducing redundancy [39, 3]. But these results have been called into question, as fixed planning strategies can give wildly varying relative performance over a number of different planning domains. In a similar vein, some researchers have concluded that we are asking the wrong question: rather than ponder over which is ....

[Article contains additional citation context not shown here]

A. Barrett and D. S. Weld. Partial-Order planning: evaluating possible efficiency gains. Artificial Intelligence, 67:71--112, 1994.

....main varieties state space planners that search in the space of world states, and partial order (plan space) planners that search in the space of partial plans. Several recent studies demonstrate that searching in the space of plans provides a more flexible and efficient framework for planning [1,34]. Despite their many perceived advantages, plan space planners are not a panacea for computational intractability of domain independent planning. In particular, it is widely realized [21,47,40,26] that effective search control is critically important for getting efficient planning capabilities out ....

....plan space planner, that is both clean and elegant. We also show that the framework is capable of significantly improving the performance of a plan space planner. First, we will describe SNLP EBL [29,28] a system that learns search control rules for SNLP, a causal link partial order planner [30,1]. Learning is initiated whenever the planner detects a failure or crosses the depth limit. In either case, SNLP EBL explains the failure by isolating a minimal subset of the constraints on the partial plan 2 Fail Fail Fail Success Planner EBL Control Rules Problem Solution Fig. 1. EBL in ....

[Article contains additional citation context not shown here]

A. Barrett and D.S. Weld. Partial Order Planning: Evaluating Possible Efficiency Gains. University of Washington, Technical Report 92-05-01, 1992

....again. While an inference engine is used to prove goals, a diagnostic engine is provided to diagnose faults. Repairing in FLIPPER is pre writing scripts for various purposes, however, we expect to utilize a planner to automatically generate necessary primitive actions to rectify diagnosed faults. 3 A Goal Directed Approach to Fault Diagnosis Guarded Horn Clauses (GHCs) are used to express model rules. A GHC rule R is expressed as: H IF G B, where H , G , and B are the head, guard and body respectively. Goals are divided into diagnosable ones, which model writers think are relevant to the ....

.... there is no swap space in lv POST [lv] on [vg] PhysVol pv] forAll ( pv] logPart [lv] # bpv] logPart [lv] RULES [VolGroup vg] onlyOnePV IF [PhysVol pv] belongsTo [vg] PhysVol p] forAll ( p] belongsTo [vg] pv) VolGroup vg] noneLV IF [LogVol lv] forAll ( lv] on [vg] 4. 3 A Heuristic Partial Order Planning Algorithm In the heuristic planning algorithm below, we use the following functions and notations: pre(A) returns pre conditions of an action A; post(A) returns post conditions of an action A; g(G) returns 1 if goal G can be achieved by an access method; ....

A. Barrett and D.S. Weld, Partial order planning: Evaluating possible efficiency gains, Artificial Intelligence, Vol. 67, No. 1, 71-112, 1994.

....can be performed by a large number of alternative operations, all using different resource combinations. Resource changes are costly and may deteriorate the accuracy of machining (or inspection) operations. The order of performing tasks is very important. Typically, the tasks are nonserializable (Barrett and Weld 1994): they cannot be achieved without interleaving subplans for other tasks. Beyond ordering constraints, there can be also resource binding constraints that require two or more tasks to be performed with the same resources (quite a common requirement in inspection planning) The constraints stem from ....

Barrett, A., and Weld, D. S. 1994. Partial-Order Planning: Evaluating Possible Efficiency Gains. Artificial Intelligence 67:71-112.

....advocated by others. Advocates of the delayed commitment approach argue that unnecessary backtracking can be eliminated by postponing choices until as late as possible. In particular, Yang and Chan cite experimental results (based on an extension to the SNLP planner implemented by Barret and Weld [3]) showing that delayed commitment can improve planning performance substantially in certain cases [36] A key issue for determining whether to use eager or delayed variable binding is the density of the solution space: the military problems used for the experiments reported here have dense ....

A. Barrett and D. Weld. Partial order planning: Evaluating possible efficiency gains. Technical Report 92-05-01, Department of Computer Science and Engineering, University of Washington, 1992.

.... required to encode the same problem into SAT) 4 Experimental Results We empirically compared LPSP to the PAS approach on problem instances from three well known planning domains: the blocks world and logistics domains [Kautz and Selman, 1996; Ernst et al. 1997] and the artificial D1S1 domain [Barrett and Weld, 1994] . 4 For the PAS approach, the performance depends on three components: i) the SAT encoding of the planning instances, ii) polynomial simplification algorithms which are applied to the SAT formula before general SAT solvers are invoked, and (iii) the SAT algorithm used to solve the simplified ....

A. Barrett and D. Weld. Partial order planning: Evaluating possible efficiency gains. Artificial Intelligence, 67(1):71--112, 1994.

No context found.

A. Barret and D. S. Weld. Partial-order planning: Evaluating possible efficiency gains. Artificial Intelligence, 67(1):71--112, 1994.

No context found.

A. Barrett and D. Weld. Partial Order Planning: Evaluating Possible Efficiency Gains. Artificial Intelligence, Vol. 67, No. 1, 1994.

No context found.

A. Barrett and D.S. Weld. Partial-order planning: evaluating possible efficiency gains. Artificial Intelligence, 67(1):71--112, 1994.

No context found.

A. Barrett and D. Weld. Partial order planning: evaluating possible efficiency gains. Artificial Intelligence, 67:71--112, 1994.

No context found.

A. Barrett and D.S. Weld. Partial Order Planning: Evaluating Possible Efficiency Gains. Artificial Intelligence, Vol. 67, No.1, 1994.

No context found.

A. Barrett and D. Weld. Partial order planning: Evaluating possible efficiency gains. Technical Report 92-05-01, Department of Computer Science and Engineering, University of Washington, Seattle, WA, June 1992.

No context found.

A. Barrett and D. Weld. Partial order planning: evaluating possible efficiency gains. Artificial Intelligence, 67:71--112, 1994.

No context found.

A. Barrett and D. Weld. Partial order planning: evaluating possible efficiency gains. Artificial Intelligence, 67:71--112, 1994. 12

No context found.

A. Barrett and D. Weld. Partial order planning: Evaluating possible efficiency gains. Technical Report 92-05-01, Department of Computer Science and Engineering, University of Washington, Seattle, WA, June 1992.

No context found.

A. Barret and D. S. Weld. Partial-order planning: evaluating possible efficiency gains. Artificial Intelligence, 67: (71-112), 1994.

No context found.

Barrett, A. & Weld, D. S., "Partial-Order Planning: Evaluating Possible Efficiency Gains", Technical Report 92-05-01 Expanded Version, Department of Computer Science & Engineering, University of Washington, Seattle, WA 98195, February 19, 1993.

First 50 documents  Next 50

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