David Ruby and Dennis Kibler. Learning subgoal sequences for planning. In Proceedings of the Eleventh International Joint Conferenceon Artificial Intelligence, pages 609--614, Detroit, MI, 1989.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....is inserted. Also information about how many resources that subplan generated, consumed and its precondition is remembered. For more information about Petri net reachability tree analysis see [28] The usage of subplans reduces dramatically the time needed to construct a plan if used wisely [29]. In our case subplans are generated as a side e ect using Petri net state space collapsing. If the achieved state is equivalent to goal, search is terminated at that particular reachability tree node, plan is added to a list of plans P , and the inspection of next transitions begins. In the ....

D. Ruby, D. Kibler. Learning Subgoal Sequences for Planning. Proceedings of IJCAI'89, Detroit, Michigan USA, 20-25 August 1989.

....proving. While allowing only certain literals at di erent levels of abstractions, the number of LL program clauses at every level is reduced, and thus proving complexity in the LL sequent calculus is reduced. Using abstraction techniques we may cut solution search space from b d kb d=k [61, 87], where b is the branching factor of a search, d is the depth of the search space and k is the ratio of abstraction space to base space in an hierarchy. In [61] it has been shown that while using optimum abstraction hierarchies it is possible to reduce the expected search time from O(n) to O(log ....

....about how many resources a particular subsolution generated, consumed and its precondition marking is remembered. For more information about Petri net reachability tree analysis see [82] The usage of subsolutions reduces dramatically the time needed to construct a proof if used wisely [87]. In our case subsolutions are generated as a side e ect using Petri net state space collapsing. If an achieved state is equivalent to goal, search is terminated at the given particular reachability tree node, the proof is inserted into the list of proofs (announce(proof) and the inspection of ....

D. Ruby, D. Kibler. Learning Subgoal Sequences for Planning. In Proceedings of the Eleventh International Joint Conference on Arti cial Intelligence (IJCAI'89), Detroit, Michigan, USA, 20-25 August, 1989.

....space in LL sequent calculus is reduced. 1 Introduction Abstraction techniques, constituting a subset of divide and conquer approaches, are widely viewed as methods for making intractable problems tractable. Using abstraction techniques we may cut solution search space from b d to kb d=k [9, 14], where b and d are respectively the branching factor and the depth of the search tree and k is the ratio of the abstraction space to the base space in an abstraction hierarchy. In [9] it is showed that when optimum abstraction hierarchies are used, it is possible to reduce the expected search ....

D. Ruby and D. Kibler. Learning subgoal sequences for planning. In Proceedings of the Eleventh International Joint Conference on Arti cial Intelligence (IJCAI'89), Detroit, Michigan, USA, 20-25 August, 1989.

....is inserted. Also information about how many resources that subplan generated, consumed and its precondition is remembered. For more information about Petri net reachability tree analysis see [24] The usage of subplans reduces dramatically the time needed to construct a plan if used wisely [25]. In our case subplans are generated as a side e ect using Petri net state space collapsing. If achieved state is equivalent to goal, search is terminated at that particular reachability tree node, plan is added to a list of plans P , and inspection of next transitions begins. In the case goal ....

D. Ruby, D. Kibler. Learning Subgoal Sequences for Planning. Proceedings of IJCAI'89, Detroit, Michigan USA, 20-25 August

....generate subplans (which are common parts shared by many plans) to be executed by default when timing constraints apply to the planning process or other problems are encountered. The usage of subplans could reduce dramatically the time needed to construct a plan for achieving some particular goal [33]. Usually, while solving di erent goals simultaneously, the planner should prefer executing plans, which have overlapping parts [40] in such way resource usage may be minimized. For example to buy seven di erent object from a store, it does not make sense to walk seven times to the same shop, if ....

D. Ruby, D. Kibler. Learning Subgoal Sequences for Planning. In: Proceedings of IJCAI'89, Detroit, Michigan USA, 20-25 August 1989, Vol. 1, pp. 609-614.

....actions not recommended by the policy after exhausting the (reachable) policy subspace. 7 Note that even if we debug a policy until it has a 1. 0 goal satisfaction probability, the set of critical scrs will (almost) never constitute a universal plan [23] 8 A related technique, SteppingStone [22], learns intermediate subgoals for a means ends problem solver. the critical choices in Figure 2 can be avoided by introducing the intermediate subproblem of going to the SE corner of room B. The proposed reactive system design did not constrain how the primitive reactive policies are ....

Ruby, D. & Kibler, D. 1989. Learning Subgoal Sequences for Planning. Proc. of the IJCAI Conf.

.... using so called intermediate goals (these are facts that the planner must make true before it can achieve an original goal) has been explored inside GAM and the results are reported in (Koehler Ho mann, 1998) Earlier work addressing the task of learning intermediate goals can be found in (Ruby Kibler, 1989), but this problem has not been in the focus of AI planning research since then. A third line of work addresses the interaction of GAM with a forward searching planning system. We have seen that GAM preserves the correctness of a planner, and that it preserves the completeness at least on ....

Ruby, D., & Kibler, D. (1989). Learning subgoal sequences for planning. In Sridharan (Sridharan, 1989), pp. 609-615.

.... A rst investigation using so called intermediate goals (these are facts that the planner must make true before it can achieve an original goal) has been explored inside GAM and the results are reported in [KH98] Earlier work addressing the task of learning intermediate goals can be found in [RK89], but this problem has not been in the focus of AI planning research since then. Currently, we are exploring a third line of work, which addresses the interaction of GAM with a planning system. We have seen that GAM preserves correctness of a planner, and that it preserves completeness, at least ....

D. Ruby and D. Kibler. Learning subgoal sequences for planning. In IJCAI-89

.... Rewriting Logic translation and have illuminated this methodology with several process calculi examples [4] Clavel has built an inductive theorem prover in Maude using the meta level strategy language [5, 6] An object is represented in Maude, and rules of inference, such as explicit induction [16] are defined at the meta level [6] Strategies for choosing and applying the rules of inference are thus meta metalevel rules. Maude s extremely efficient implementation enables the effective computations of the result of such meta metalevel rules on reasonably large examples in reasonable time. ....

....seems to be devoted to first order systems and, if I may be provocative, this is a curious state of affairs. First order logic is not expressive: consider how powerful resolution provers are, and how few applications they have. Adding set theory to first order logic yields an expressive system [14, 16], but formalizing concepts such as set comprehension, fx 2 A j P (x)g, and general union, S x2A B(x) requires a higher order syntax. Approaches based on first order syntax are not attractive we might have a clumsy language of combinators or be forced to define an auxiliary function every time ....

[Article contains additional citation context not shown here]

D. Ruby and D. Kibler. Learning subgoal sequences for planning. In Proceedings of IJCAI-89, pages 609--614, 1989.

....other macro learning algorithms and discuss its strengths and weaknesses. Most of the existing macro learning programs are based on the notion of subgoaling: the learner tries to acquire macros that achieve some subgoal without undoing previously satisfied subgoals (Korf, 1985; Laird et al. 1986; Ruby Kibler, 1989, 1992; Tadepalli, 1991; Tadepalli Natarajan, 1996) Micro Hillary, like MacLearn (Iba, 1985) does not assume subgoaling, but assumes the existence of a heuristic function. EASe (Ruby Kibler, 1992) combines subgoaling with a heuristic function to guide the search for the current subgoal. The ....

Ruby, D., & Kibler, D. (1989). Learning subgoal sequences for planning. In Proceedings of the Eleventh International Joint Conference on Artificial Intelligence, pp. 609--614.

....(FAF) is more efficient. They suggest that FAF is likely to be more efficient and can be computed in constant time. While our approach to island refinement is based on a proper language abstraction of the goal and assumptions, island search does not necessarily involve abstraction, see e.g. [16] on stepping stones. There is a variety of approaches to abstraction. Problem abstraction to guide problem solving and planning has, for instance, been addressed in [17, 6, 10] More specifically, abstraction in theorem proving is addressed, e.g. in Plaisted s classical paper [15] Hutter and ....

D. Ruby and D. Kibler. Learning subgoal sequences for planning. In Proceedings of IJCAI-89, pages 609--614, 1989.

....particular decomposition is appropriate for efficient planning. The previous methods that have been used for representing control knowledge include macro operators and control rules [ Minton, 1988 ] Explanation based learning (EBL) has been a popular method for speedup learning [ Minton, 1988; Ruby and Kibler, 1991 ] EBL, because of its dependence on a single example to generate a rule, leads to over specific rules. This results in proliferation of expensive to match rules, a problem described as utility problem [ Minton, 1988 ] Inductive logic programming (ILP) techniques, on the other hand, can ....

D. Ruby and D. Kibler. Learning subgoal sequences for planning. In Proceedings of AAAI-91. AAAI Press, 1991.

....on learning, and much of the recent experimental work on planning falls into this area. In this paradigm, one runs a planning system with and without a learning component, then examines differences in performance between the two variants. Allen and Langley (1990) Iba (1989) Minton (1990) Ruby and Kibler (1989), and Shavlik (1990) report evidence that a variety of learning components can improve the behavior of planning systems after sufficient experience in a given domain. In some cases, researchers have also found negative results; both Iba (1989) and Minton (1990) have shown that naive learning ....

Ruby, D., & Kibler, D. (1989). Learning subgoal sequences for planning. Proceedings of the Eleventh International Joint Conference on Artificial Intelligence (pp. 609--614). Detroit, MI: Morgan Kaufmann.

No context found.

David Ruby and Dennis Kibler. Learning subgoal sequences for planning. In Proceedings of the Eleventh International Joint Conferenceon Artificial Intelligence, pages 609--614, Detroit, MI, 1989.

No context found.

D. Ruby and D.F. Kibler, `Learning subgoal sequences for planning', in Proceedings of the 11th IJCAI, ed., N.S. Sridharan, pp. 609--614, (1989). International Joint Conference on AI, Morgan Kaufmann.

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