Haslum, P., and Geffner, H. 2000. Admissible heuristics for optimal planning. In International Conference on AI Planning and Scheduling, 2000.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....exit distance under h ; the road map diameter in the AIPS 2002 instances varies around 1 to 6. Let us focus on the other heuristic planners. Those that are state of the art in runtime all use relaxed plan heuristics in one way or the other. There are 33 also planners based on other heuristics [41,42], but these optimal planners can not compete with the runtime performance of their sub optimal relaxed plan based counterparts. Except FF, the most successful heuristic planners in the last 3 years (speci cally in the AIPS 2000 and AIPS 2002 competitions) have been HSP2 [9] STAN4 [19] Mips ....

P. Haslum, H. Gener, Admissible heuristics for optimal planning, in: Chien et al. [57], pp. 140-149.

....approach to solving this problem (assuming that a solution exists) consists of searching the state space to find a path between the initial and final states. Several planning systems adopting this approach in conjunction with the use of (automatically extracted or user provided) heuristics [5, 4, 1] have shown notable results in terms of efficiency on various planning competition problems. Although a state space search algorithm is conceptually simple, it is not obvious how the data structures and control mechanisms required for the specification and execution of such an algorithm can be ....

....y) and Unstack(x; y) The two PAC events can be seen as ground instances of operators, except for the absence of the Delete list and of other secondary effects, which can be deduced from the current context . In other words, the agent can be seen as having solved a relaxed problem (cf. [4, 5]) in which non relevant facts are ignored and (possible) negative interations between actions are taken care of only in a later phase of the cognitive planning process. For example, achieving On(B,A) from OnTable(A,B) also implies given a two block world context Clear(B) Not(OnTable(B) ....

[Article contains additional citation context not shown here]

P. Haslum H. Geffner. Admissible heuristics for optimal planning. In Proceedings of the 5th Internat. Conf. of AI Planning Systems (AIPS 2000.

....action costs in the heuristics. In [39] a multi criteria heuristics is presented to obtain near optimal plans, considering multiple criteria apart from plan length alone. However, the described heuristics is not fully admissible, i.e. does only guarantee optimal plans under certain restrictions [21]. In fact, most heuristic state space planners are not able to guarantee optimality. A powerful approach is given in [35] where planning with resources is described as a structural constraint satisfaction problem (SCSP) The problem is solved by local search combined with global control. ....

P. Haslum and H. Geffner. Admissible Heuristics for Optimal Planning. In S. Chien, S. Kambhampati, and C. A. Knoblock, editors, Proceedings of the Fifth International Conference on Artificial Intelligence Planning and Scheduling (AIPS'00), pages 140--149, Breckenridge, Colorado, USA, April 2000. AAAI Press.

....approach to solving this problem (assuming that a solution exists) consists of searching the state space to find a path between the initial and final states. Several planning systems adopting this approach in conjunction with the use of (automatically extracted or user provided) heuristics [5, 4, 1] have shown notable results in terms of efficiency on various planning competition problems. Although a state space search algorithm is conceptually simple, it is not obvious how the data structures and control mechanisms required for the specification and execution of such an algorithm can be ....

....#) and Unstack(## #) The two PAC events can be seen as ground instances of operators, except for the absence of the Delete list and of other secondary effects, which can be deduced from the current context . In other words, the agent can be seen as having solved a relaxed problem (cf. [4, 5]) in which non relevant facts are ignored and (possible) negative interations between actions are taken care of only in a later phase of the cognitive planning process. 3 The connectionist planning schema The planning mechanism described above has been implemented using the representational ....

[Article contains additional citation context not shown here]

P. Haslum H. Geffner. Admissible heuristics for optimal planning. In Proceedings of the 5th Internat. Conf. of AI Planning Systems (AIPS 2000.

....which stay within a given limit (see [8] In the last years, it has been widely recognized that plan length alone is only one criterion to be optimized in planning. Several attempts have been made to extend heuristic search planners to allow for special heuristics respecting action costs, e.g. [9, 14]. A powerful approach is given in [20] where planning with resources is described as a structural constraint satisfaction problem (SCSP) The problem is solved by local search combined with global control. However, 20] promotes the inclusion of domain dependent knowledge; the general problem ....

P. Haslum and H. Geffner. Admissible heuristics for optimal planning. AIPS'00, pp. 140--

....where Graphplan produces a k step plan with m actions, even when there is an n action serial plan (k n m) For serial Graphplan, each step contains exactly one action, and thus step optimality implies optimality with respect to number of actions. An exception is the work by Haslum and Geffner [17], which was done around the same time as ours; see Section 9 for a discussion. Tab le 4 Column titled Len shows the length of the found optimal plan (in number of actions) Column titled Est shows the heuristic value the distance from the initial state to the goal state. Column titled Time ....

....computed) Mutex marking and propagation in the planning graph directly correspond to the updating procedure of the lev function. In addition, upon convergence lev values directly correspond to the lev values extracted from a leveled planning graph. In fact, in a recent work, Haslum and Geffner [17] start from the DP formulation to derive a family of heuristics that are closely connected to the set level heuristic function h(S) developed in Section 4. The heuristics used in HSP and HSP r planners are also computed using a bottom up DP approach. The bottom line is that, unlike the ....

[Article contains additional citation context not shown here]

P. Haslum, H. Geffner, Admissible heuristics for optimal planning, in: Proc. AIPS-2000.

....plan in order to minimize the number of parallel steps. 3. Use a greedy technique to incrementally parallelize the plan. The first approach is basically infeasible, because we would need to consider all partial subsets of applicable actions exponentially increasing our branching factor [12, 5]. The second approach is extensively discussed in [14] in which reorderings and deorderings of sequential plans are done offline to minimize the number of parallel steps. The main drawback of this approach is that it does not minimize the number of parallel steps given the overall problem, but ....

.... Definition 2 (Interaction) Two actions ] do not interact with each other if 7 Ia b c ] dMe b f f b c g hP ] d d ij 7 a b c ] k dXelb f f b c g hP ] k d d d4m9n In other words, neither of the action deletes atoms from the precondition and effect lists of the other [5]. So, we consider a parallel step as one containing a set of actions which are pairwise independent. AltAlt p changes the branching strategy of AltAlt, and it has two main alternate steps, the selection of the most promising parallel node and the rearranging of the plan. We will discuss our ....

[Article contains additional citation context not shown here]

P. Haslum and H. Geffner. Admissible Heuristics for Optimal Planning. In Proc. AIPS-2000.

....of the 37 examples that we tried. Most important, while C PLAN runs out of time, CMBP and GPT run out of memory. This seems to point out that C PLAN range of applicability is different from the range of applicability of CMBP and GPT. Analogous results supporting this fact are reported in [19] where it is shown that for classical, highly parallel domains the planning as satisfiability approaches appear to do best. The fact that SAT based approaches and BDD based approaches have different range of applicability is also confirmed by previous work comparing BDD and DLL as SAT ....

Patrick Haslum and Hector Geffner. Admissible heuristics for optimal planning. In Proc. AIPS-2000, pages 140--149.

....Furthermore, we have discussed some optimizations. We believe that SAT based approaches to planning with incomplete information can be very e ective at least on problems with a high degree of parallelism. This belief is con rmed by a preliminary experimental analysis, by the results in [ Haslum and Ge ner, 2000 ] for the classical case, and also for the very positive results that SAT based approaches are having in formal veri cation. Acknowledgments A special thank to Paolo Ferraris for many fruitful discussions on the topic of this paper. Paolo has also participated to the design of the architecture ....

Patrick Haslum and Hector Gener. Admissible heuristics for optimal planning. In Proc. AIPS, pages 140-149, 2000.

....to a probabilistic version of SAT in order to handle some kinds of probabilistic planning. Recently Huang et al. 42] extended the Blackbox system itself to 1 One should note that for other domains direct search of the plan graph or heuristic state space search may be best. Haslum and Geffner [37] note that the SAT approaches tend to dominate for problems where there are many interactions between the goals and a large number of parallelizable actions in the solution. 9 learn domain specific control rules using inductive logic programming. 3.2 Planning with Nondeterministic Actions and ....

P. Haslum and H. Geffner. Admissible heuristics for optimal planning. In Proceedings of the Fifth International Conference on Artificial Intelligence Planning and Scheduling, pages 140--149. AAAI Press, 2000.

....as NASA s RAX [10] are based on the POP algorithms. Even for simpler planning domains, partial order planners search for and output partially ordered plans that offer a higher degree of execution flexibility. In contrast, none of the known state space planners can find parallel plans efficiently [8] , and CSP planners such as Graphplan only generate a very restricted class of parallel plans (see Section 5) The foregoing motivates the need for improving the efficiency of POP algorithms. We show in this paper that the insights and techniques responsible for the advances in plan synthesis ....

P. Haslum and H. Geffner. Admissible Heuristics for Optimal Planning. In Proc. AIPS-2000, 2000.

....In the second phase, HSPr performs a regression search guided by the computed heuristic function, just the same as Graphplan. Graphplan thus can be seen as a heuristic search planner: it employs an A regression search guided by the heuristic function encoded into the planning graph. As shown in [7] heuristic search planners can be modi ed to generate parallel plans like Graphplan, so parallelity is not a real di erence. Graphplan s feature memoization can be seen as an update of the heuristic function. In Graphplan the two phases are repeated, yielding an IDA search (iterative deepening ....

....[6] The future quest of heuristic search planning will be the derivation of better heuristic functions from such richer declarative languages. Possibly, approaches will include the ignorance of preconditions instead of delete lists or an extension of h max pair to h max triple or h max quadruple [7]. Moreover it is important to try bringing di erent approaches together. For example, maybe it s possible to combine heuristic search with logic based problem solvers and thus improve the performance of both. 11 ....

P. Haslum and H. Gener. Admissible heuristics for optimal planning. In Proceedings AIPS-2000. AAAI Press, 2000.

....correspond to the methods considered in this paper. More closely related is one of the options in the Medic system for encoding planning problems: a simple inference method which is referred to as simple data ow analysis (Ernst et al. 1997) This method is basically an instance of Reachable 1. Haslum and Ge ner (2000) present a parametrized class of admissible heuristics functions H k . There is an interesting and important relation between the heuristic function generation technique discussed in that paper and the parameterized class of reachability analysis algorithms discussed in this paper. When a ....

Haslum, P., & Gener, H. (2000). Admissible heuristics for optimal planning. In Proc. of the Fifth Intl. Conf. on AI Planning and Scheduling Systems, pp. 140-149.

....for linking our domain analysis and planning systems. Speci cally, ProbaPla [24] determines action sequence replaceability and nds a form of state invariants, while path analysis detects irrelevant actions (using the previously detected invariants in the process) Planners of the hsp family [14] can pro t from knowledge of action irrelevance and state invariants. We hope of course that DKEL will nd use outside of our systems, and become a useful tool for the integration of domain analysis in every planner. ....

Patrik Haslum and Hector Gener. Admissible heuristics for optimal planning. In Steve Chien, Subbarao Kambhampati, and Craig A. Knoblock, editors, Proceedings of the Conference on Articial Intelligence Planning & Scheduling, pages 140-149, April 2000.

....search has been shown to be a good framework for developing different kinds of planning algorithms. It has been most successful in non optimal sequential planning, e.g. hsp (Bonet Geffner 2001) and ff (Hoffman 2000) but has been applied also to optimal and parallel planning with good results (Haslum Geffner 2000). We continue this thread of research by developing a domain independent planning algorithm for domains with metric time and certain kinds of resources, based on regression search with a heuristic that is derived from the problem representation. The algorithm is optimal w.r.t. the overall ....

....optimal w.r.t. the overall execution time of the plan (the makespan) Main contributions are: First, we extend the heuristic search framework to a more realistic setting, involving actions that extend in time and are constrained by limited resources. The algorithm is developed the same way as in (Haslum Geffner 2000). The method by which the heuristic is derived in particular is the same, and this method is also used to derive estimators for resource consumption. Second, we aim to combine the extended expressivity with the performance of recent STRIPS planners. We present some results, setting a preliminary ....

[Article contains additional citation context not shown here]

Haslum, P., and Geffner, H. 2000. Admissible heuristics for optimal planning. In Proc. 5th International Conference on Artificial Intelligence Planning and Scheduling. AAAI Press.

....Recently, heuristic state space search has been shown to be a good framework for developing different kinds of planning algorithms. It has been most successful in non optimal sequential planning, e.g. hsp [4] and ff [10] but has been applied also to optimal and parallel planning with good results [8]. We continue this thread of research by developing a domain independent planning algorithm for domains with metric time and certain kinds of resources. The algorithm relies on regression search guided by a heuristic that estimates completion time and which is derived automatically from the ....

....This is actually a desirable feature 2 . The rule is optimality preserving in the sense that it generates some optimal plan. This, along with soundness, is all that is needed for optimality (provided an admissible search algorithm and heuristic are used) 3. 2 Heuristic As in previous work [8], we derive an admissible heuristic by introducing approximations in the recursive formulation of the optimal cost function. For any state s = E; F ) the optimal cost is H (s) t iff t is the least time t such that there is a plan P that achieves s at t. The optimal cost function, H , is the ....

P. Haslum and H. Geffner. Admissible heuristics for optimal planning. In Proc. 5th International Conference on Artificial Intelligence Planning and Scheduling. AAAI Press, 2000.

....Recently, heuristic state space search has been shown to be a good framework for developing different kinds of planning algorithms. It has been most successful in non optimal sequential planning, e.g. hsp [3] and ff [9] but has been applied also to optimal and parallel planning with good results [7]. We continue this thread of research by developing a domain independent planning algorithm for domains with metric time and certain kinds of resources. The algorithm relies on regression search guided by a heuristic that estimates completion time and which is derived automatically from the ....

....This is actually a desirable feature. 2 The rule is optimality preserving in the sense that it generates some optimal plans. This, along with soundness, is all that is needed for optimality (provided an admissible search algorithm and an admissible heuristic are used) 4. 3 Heuristic Like in [7], we derive an admissible planning heuristic by introducing approximations in the recursive formulation of the optimal cost function. For any state s = E; F ) the optimal cost is H (s) t iff t is the least time t such that there is a plan P that achieves s at t. The optimal cost function, H ....

[Article contains additional citation context not shown here]

P. Haslum and H. Geffner. Admissible heuristics for optimal planning. In Proc. 5th International Conference on Artificial Intelligence Planning and Scheduling. AAAI Press, 2000.

....bounds refer to approximate cost measures used for focusing and pruning the search. In AI Planning, admissible heuristics or lower bounds have received considerable attention recently, and most optimal planners currently use them, either explicitly or implicitly (e.g. by relying on a plan graph) [6, 19]. Branching, on the other hand, has received less attention and is seldom discussed explicitly. In this paper, we make the notion of branching in planning explicit and relate it to branching schemes used in combinatorial optimization. We argue that the directional branching schemes used in ....

....that are very competitive [6] The additive heuristic, however, is not a lower bound, and hence it s not useful to nd optimal plans. 3. 1 Heuristics h m A family of admissible heuristics (lower bounds) h m , for m = 1; 2; for sequential and parallel Strips planning is formulated in [19]. The heuristic h m approximates the cost of a set of atoms C by the cost of the most costly subset of size m in C. Thus, for m = 1, h m approximates the cost of a set of atoms by the cost of the most costly atom in the set, for m = 2, h m approximates the cost of the set by the cost of the ....

[Article contains additional citation context not shown here]

P. Haslum and H. Gener. Admissible heuristics for optimal planning. In Proc. of the Fifth International Conference on AI Planning Systems (AIPS-2000), pages 70-82, 2000.

No context found.

Haslum, P., and Geffner, H. 2000. Admissible heuristics for optimal planning. In International Conference on AI Planning and Scheduling, 2000.

No context found.

P. Haslum and H. Ge#ner, Admissible Heuristics for Optimal Planning, Proceedings of AIPS-00, 2000, pp. 140--149.

No context found.

P. Haslum and H. Geffner. Admissible Heuristics for Optimal Planning. In Proc. AIPS 2000.

No context found.

P. Haslum and H. Geffner. Admissible Heuristics for Optimal Planning. In Proc. AIPS-2000.

No context found.

Patrick Haslum and Hector Geffner. Admissible heuristics for optimal planning. In Chien et al. [15], pages 140--149.

No context found.

Parik Haslum and H ector Geffner. Admissible heuristics for optimal planning. In Proc. of the 5th Intl. Conf. on Artificial Intelligence Planning and Scheduling, pages 140--149. AAAI Press, 2000.

No context found.

P. Haslum and H. Ge#ner. Admissible heuristics for optimal planning. In Arti#cial Intelligence Planning and Scheduling #AIPS#, pages 140#149, 2000.

First 50 documents

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