B. Srivastava and S. Kambhampati. Synthesizing customized planners from specifications. Journal of Artificial Intelligence Research, 8:93--128, 1998.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....work on learning and reasoning with planning domains has come out of the PRODIGY project, but performance of the scale demonstrated by our approach has not been achieved. There has also been some more recent work on utilizing domain dependent control knowledge by Srivastava and Kambhampati [52] and by Kautz and Selman [34] Srivastava and Kambhampati present a scheme for taking domain specific information similar to that used by TLPLAN and using that information as input to a complex program synthesis system. The end result is an semiautomatically constructed planning system that is ....

....utilizes. Unlike TLPLAN however, their approach requires a complex program synthesis step to make use of this information (a customized planner must first be synthesized) In TLPLAN the control information is simply part of the planner s input. Furthermore, the empirical results presented in [52] show performance that is orders of magnitude inferior to TLPLAN. For example, their customized planners took about one minute each to solve the standard tire fixit problem, a 12 package logistics problem, and a 14 block problem. TLPLAN takes 0.06 seconds to solve the tire fixit problem, about 3 ....

B. Srivastava and S. Kambhampati. Synthesizing customized planners from specifications. Journal of Artificial Intelligence Research, 8:93--128, 1998.

....Cameron Bruce Fraser Dept. of Computer Science University of Waterloo Waterloo, Ontario Canada, N2L 3G1 cbfraser logos.uwaterloo.ca March 10, 2000 1 Introduction A number of works have shown that planning can be speeded up, often very significantly, by utilizing extra domain knowledge [KM81, BK00, BK96, KS98, SK98, DK99, Rei99, NCLMA99, vBC99]. The question that immediately arises is where does this extra information come from We have gained considerable experience with utilizing extra domain information in a planner through implementing numerous planning domains in the TLPLAN system [BK00] The TLPLAN system is a planning system ....

B. Srivastava and S. Kambhampati. Synthesizing customized planners from specifications. Journal of Artificial Intelligence Research, 8:93--128, 1998.

.... methods bring three main benefits to the field as they allow for the validation of planning domain models against the real world, the recognition of useful structural domain properties to improve planning efficiency (e.g. 1] and the design of customised planners for specific domain classes (e.g. [2]) The use of structural properties to speed up planning algorithms is not new in the literature; good examples are [3] 4] 5] 6] to mention but a few. However, neither of these approaches has proposed a generic framework to analyse domain properties independently of the used planning ....

....has proposed a generic framework to analyse domain properties independently of the used planning algorithms. Moreover, few approaches have addressed the problem of developing complete methodologies to construct domain models specially suited for planners. Recent works in the literature [1] [2] [7] have demonstrated the usefulness of these methods for improving current planner systems. The lack of a methodology for domain model construction is especially critical in the case of large multi agent domains (e.g. Urban Traffic Control [8] 9] Construction [10] Distributed Software Domains ....

[Article contains additional citation context not shown here]

B.Srivastava and S.Kambhampati "Synthesizing Customized Planners from Specifications" Journal of Artif. Intell. Research, Vol. 8, pp 93-128, 1998.

....efficient in any one given domain. The domain dependent approaches need to be (re)designed for each domain separately, but can be very efficient in the domain for which they are designed. One enticing alternative to these approaches is to automatically synthesize domain independent planners. [25] This theoretical consilience among related but separate fields suggests the striking versatility of the automated, domain specific design approach. Again, formalization of such systems is crucial for their widespread influence and application. 6 Conclusions and Future Work Our ultimate goal is ....

B Srivastava and S Kambhampati. Synthesizing Customized Planners from Specifications. Journal of Artificial Intelligence Research 8 (1998), 93-128.

....to speed up the process of finding a model. In their work they have reported speed ups of up to a factor of 7 gained by adding domain specific information (although they have not as yet obtained the exponential speedups reported by Bacchus and Kabanza) Finally, Srivastavan and Kambhampati [SK98] have presented a scheme where domain specific knowledge, represented in a declarative fashion, is used as input to an automated programming system. This knowledge is used by the system to construct a domain specific planner. These approaches share 1 Specifically, planning problems that require ....

.... to) add (at p to) add (at t to) del (at p from) del (at t from) Figure 1: Operators for the Logistics World Here we utilize the formalization of these constraints originally given by Bacchus and Kabanza [BK98] Both Kautz and Selman [KS98] and Srivastavan and Kambhampati [SK98] have described similar types of control information for this domain. We start out with a collection of auxiliary predicates. Given its current location, c loc, an object, obj, is in the wrong city if it has a goal location and that location is in a different city. 3 (def predicate ....

B. Srivastava and S. Kambhampati. Synthesizing customized planners from specifications. Journal of Artificial Intelligence Research, 8:93--128, 1998.

....I, results in a world state in which all goals G are satisfied. Different classical planning paradigms can be characterized by the means and representations to find solutions for a planning problem, e.g. the requirements and properties of the plan representation (see (Kambhampati et al. 1995; Srivastava and Kambhampati, 1998)) 2.3 Plan Representation CAPlan is a partial order planner that is based on the SNLP algorithm (McAllester and Rosenblitt, 1991; Barrett and Weld, 1994) It searches in the space of plans to find a solution plan for a given problem. An introduction to these ideas can also be found in (Weld, ....

....Intern. Conf. on AI Planning Systems (AIPS 96) pages 125 133. Kambhampati, S. Knoblock, C. and Yang, Q. 1995) Planning as refinement search: A unified framework for evaluating design tradeoffs in partial order planning. Artificial Intelligence, 76:167 238. Kambhampati, S. Mali, A. and Srivastava, B. 1998). Hybrid planning for partially hierarchical domains. In Proceedings of AAAI 98. Korf, R. 1987) Planning as search: A quantitative approach. Artificial Intelligence, 33:65 88. McAllester, D. and Rosenblitt, D. 1991) Systematic nonlinear planning. In Proceedings of AAAI 91, pages 634 639. ....

Srivastava, B. and Kambhampati, S. (1998). Synthesizing customized planners from specifications.

....initially concentrated on the synthesis of planners using state space refinement theories (FSS and BSS) Work on the synthesis of PSS planner is still in progress and we discuss our approach towards the end of this paper. Empirical evaluation shows that synthesized planners can be very efficient [18]. For example, in the blocks world domain where the goal was stack inversion, a KIDS synthesized planner could solve 14 blocks task in under a minute. This is unheard of in traditional planners. In the logistics domain, a task with 12 packages, 4 planes and 8 places was solved in under a minute. ....

....in under a minute. To put the performance results in perspective, we compared KIDS synthesized planners and the instantiations of UCP (which emulate a spectrum of classical planners, including the popular SNLP planner [12] by selecting the appropriate refinement) across many blocks world tasks [18]. In our experiments with state space planners for the blocks world domain, the best of the KIDS synthesized planners outperformed the best of the UCP instantiations when given the same domain specific informa tion. We hypothesize that this is because KIDS can profitably fold in the ....

[Article contains additional citation context not shown here]

B. Srivastava and S. Kambhampati. Synthesizing customized planners from specifications. Journal of AI Research, 1997 (to appear).

No context found.

B. Srivastava and S. Kambhampati. Synthesizing customized planners from specifications. Journal of Artificial Intelligence Research, 8:93--128, 1998.

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