K. Golden and D. Weld, `Representing sensing actions: The middle ground revisited ', Knowledge Representation and Reasoning, 174--185, (1996).  Home/Search   Document Details and Download   Summary   Related Articles   Check

....; r ; Delta ; p g then the choice between a and b will be based on full lookahead to the end of the program, i.e. the program will be executed essentially in the old way. 1.2 Sensing actions This on line style of execution is well suited to programs containing sensing actions. As described in [9,13,20], sensing actions are actions that can be taken by the agent or robot to obtain information about the state of certain fluents, rather than to change them. The motivation for sensing actions involves applications where because the initial state of the world is incompletely specified or because of ....

K. Golden and D. Weld. Representing sensing actions: the middle ground revisited. In Proc. KR-96, 1996.

....and sensing actions were first investigated in [45] in the context of the Situation Calculus, and in [57] this approach was combined with the solution to the Frame Problem provided by successor state axioms. Other approaches to planning with sensing exist using special purpose logics, e.g. [17, 12]. The basic idea of [45, 57] is to represent state knowledge by a binary situation situation relation K(s, s # ) meaning that as far as the robot knows in situation s it could as well be in situation s # . Hence, every given fact about any such s # is considered possible by the robot. Having ....

Keith Golden and Daniel Weld. Representing sensing actions: The middle ground revisited. In L. C. Aiello, J. Doyle, and S. Shapiro, editors, Proceedings of the International Conference on Principles of Knowledge Representation and Reasoning (KR), pages 174--185, Cambridge, MA, November 1996. Morgan Kaufmann.

....at www.cl.inf.tu dresden.de ozan papers.html . 7 Discussion In being correct with respect to full k except for queries including loops, fluent calculus, as well as FLUX, is more expressive than most existing systems for reasoning about actions and planning with sensing actions, such as [4, 1, 8, 3]. These approaches use restricted notions of incomplete states, which do not allow for handling any kind of disjunctive information or reasoning about cases as required, for example, if an action is described in k to have conditional e#ects depending on whether some unknown fluent is true or ....

K. Golden and D. Weld. Representing sensing actions: The middle ground revisited. In L. C. Aiello, J. Doyle, and S. Shapiro, ed.s, Proc. of KR, 174--185, 1996.

....questions arise when one wants to remove this assumption: how to reason about the knowledge of agents and what is a plan in the presence of incomplete information. The first question led to the development of several approaches to reasoning about effects of sensing (or knowledge producing) actions [10, 15, 16, 22, 24, 27]. The second question led to the notions of conditional plans and conformant plans. The former contains sensing actions and conditionals such as the well known if then else construct and the latter is a sequence of actions which leads to the goal regardless of the value of the unknown fluents in ....

K. Golden and D. Weld. Representing sensing actions: the middle ground revisited. In KR 96, pages 174--185, 1996.

....sensing [14] The expressiveness of general theories for conditional planning, on the other hand, raises the challenge to evolve inference algorithms that e#ciently deal with the modality of knowledge. Most existing planning methods are tailored to restricted classes of planning problems, e.g. [9, 7, 2, 16, 11]. In particular, none of these systems can solve planning problems where knowledge follows implicitly : A well known example is to determine acidity of a chemical solution by sensing the color of a Litmus strip [18] The only existing system with a general solution to the Frame Problem for ....

....In particular, the system has to backtrack over the attempt to test solution A (which renders unusable the only available Litmus paper) 5 Related Work A distinguishing feature of our system is its expressiveness in comparison to most existing systems for planning with knowledge and sensing. In [9] an implementation is described for which a semantics is given based on the general Situation Calculus solution to the Frame Problem for knowledge of [20] However, the implementation is based on the notion of an incomplete state as a triple of true, false, and unknown propositional fluents. The ....

K. Golden and D. Weld. Representing sensing actions: The middle ground revisited. In L. C. Aiello, J. Doyle, and S. Shapiro, ed.'s, Proc. of KR, p. 174--185, Cambridge, 1996.

....and sensing actions were rst investigated in [45] in the context of the Situation Calculus, and in [57] this approach was combined with the solution to the Frame Problem provided by successor state axioms. Other approaches to planning with sensing exist using special purpose logics, e.g. [17, 12]. The basic idea of [45, 57] is to represent state knowledge by a binary situation situation relation K(s; s 0 ) meaning that as far as the robot knows in situation s it could as well be in situation s 0 . Hence, every given fact about any such s 0 is considered possible by the robot. Having ....

Keith Golden and Daniel Weld. Representing sensing actions: The middle ground revisited. In L. C. Aiello, J. Doyle, and S. Shapiro, editors, Proceedings of the International Conference on Principles of Knowledge Representation and Reasoning (KR), pages 174-185, Cambridge, MA, November 1996. Morgan Kaufmann.

....test is only guaranteed to provide knowledge about the existence of power at the electrical outlet under one test outcome. While researchers have extended theories of action to include the notion of sensing or knowledge producing actions (e.g. Scherl Levesque 1993; Baral Tran 1998; Golden Weld 1996; Funge 1998) and have characterized the effect of sensing actions on an agent s state of knowledge, and even how to plan (e.g. Stone 1998; Golden Weld 1996) and to project (e.g. De Giacomo Levesque 1999b) in certain cases, with sensing actions, they have not addressed the problem of ....

.... theories of action to include the notion of sensing or knowledge producing actions (e.g. Scherl Levesque 1993; Baral Tran 1998; Golden Weld 1996; Funge 1998) and have characterized the effect of sensing actions on an agent s state of knowledge, and even how to plan (e.g. Stone 1998; Golden Weld 1996)) and to project (e.g. De Giacomo Levesque 1999b) in certain cases, with sensing actions, they have not addressed the problem of how to reason in a partially observable environment 1 . More generally, they have not examined the problem of how sensing actions can be coupled with knowledge of ....

Golden, K., and Weld, D. 1996. Representing sensing actions: The middle ground revisited. In Proc. Fifth Intnl. Conf. on Principles of Knowledge Representation and Reasoning (KR'96).

....planning under one of these approximations (0 approximation) is (under certain assumptions) less complex. Since the main idea behind 0 approximation is similar to the ideas used in the design of the existing planners developed for planning with incompleteness [EHW 92, KOG92, GB94, GEW96, GW96, PC96, SW98, WAS98, Rin99] we believe that the complexity results will shed additional light into these planners and also guide the development of future planners. Our complexity analysis will be based on an extension [BS98, BS00] of the action description language A proposed in 1991 by Gelfond ....

....skeptical reasoning. 2 We are very optimistic about the practicality of 0 approximation. One of the main reasons for this optimism is the similarity between the ideas of 0 approximation [BS00] and semi heuristic ideas underlying practically useful planners UWL, SADL, etc. described in [GB94, GW96] for example, GW96] states that: In UWL (and in SADL) individual literals have truth values expressed in three valued logic: T , F , U (unknown) Similarly, Goldman and Boddy in [GB94] use a single model to represent both the world and the planners knowledge about the world; this is similar ....

[Article contains additional citation context not shown here]

K. Golden and D. Weld. Representing sensing actions: the middle ground revisited. In KR 96, pages 174--185, 1996.

.... sections of this paper we consider control modules with special sensing actions and formulate their correctness with respect to action theories that allow such actions [Moo85, SL93, LTM97] A plan based on such a theory could be a conditional plan [Lev96] and may achieve knowledge goals [GEW96, GW96] We also present sufficiency conditions for correctness of individual control rules and show how it can be used to construct control modules with sensing actions. Special sensing actions also allow us to construct conditional plans from incomplete states where the robot does not have ....

....number of states in the worst case will be 3 n where n is the number of fluents if we use the approximate theory of sensing in [BS97] It will be much larger 2 2 n if the more general theories of sensing in [Moo85, SL93, LTM97] is used. It seems to us that the conditional planner in [GW96] uses the theory of sensing in [BS97] 8 Our Agent theory Reactivity and Deliberativity In the last two sections we formalized the correctness of control modules, gave sufficiency conditions that guarantees correctness, and gave an algorithm that uses the sufficiency conditions to construct ....

[Article contains additional citation context not shown here]

K. Golden and D. Weld. Representing sensing actions: the middle ground revisited. In KR 96, pages 174--185, 1996. 53

....it is appropriate or not. Researchers from different areas have started to adress these problems by constructing information gathering systems that automatically query multiple, relevant information sources (Etzioni and Weld, 1994; Arens et al. 1996b; Levy et al. 1996; Kwok and Weld, 1996; Golden et al. 1996; Friedman and Weld, 1997; Singh, 1998) The WebPlan project 1 (funded by the German Science Foundation, DFG) which started in August 1998 aims at developing techniques for building domain specific search assistants for the Internet. The goal is to support the whole search process by looking at ....

....the Query Interpretation phase is to construct a planning problem. Executing a solution plan of this problem performs the actual retrieval operations. Therefore, the purpose of the query has to be identified and the query has to be transformed into a set of information goals (Etzioni et al. 1992; Golden et al. 1996). For this purpose, a lot of domain knowledge (the domain ontology and the knowledge about the information sources) has to be considered. Besides the domain specific operators, a set of domain independed information goals and information seeking actions have to be defined. 5 Planning for ....

[Article contains additional citation context not shown here]

Golden, K. and Weld, D. (1996). Representing sensing actions: The middleground revisited. In Proceedings of KR-96.

....composition, non determinism, recursion, and sensing. As our examples show, the correspondence between databases and robotics is simple: the e ects of actions are modeled by database updates, and sensing actions are modeled by database queries. Unlike other theories of sensing actions, e.g. [10, 6], our treatment does not require an explicit theory of knowledge. To introduce the logical framework underlying our approach, we rst present an overview of T R. Then, we de ne a specialized theory of actions for open worlds. We present how to specify elementary actions and complex actions, and ....

K. Golden and D. Weld, `Representing sensing actions: the middle ground revisited', in KR-96, (1996).

....ELSE flip lock;push door; Thus sensing actions are very important for planning in presence of incomplete information. In the past, sensing actions have been formalized in [Moo79, Moo85, SL93, Haa86, LTM97] and planning in presence of incomplete information has been studied in [EHW 92, GEW96, GW96, Lev96, KOG92] To motivate our work we now briefly review the earlier formalizations of sensing actions. 1.1 Moore s formalization To the best of our knowledge sensing actions were first formalized by Moore in his dissertation [Moo79] and in some of his later papers, for example, Moo85] ....

K. Golden and D. Weld. Representing sensing actions: the middle ground revisited. In KR 96, pages 174--185, 1996.

.... actions which were earlier used in the EVAR system in [Sch95] and formulate their correctness with respect to action theories that allow knowledge producing actions [Moo85,SL93,LTM97] A plan based on such a theory could be a conditional plan [Lev96] and may achieve knowledge goals [GEW96,GW96] We also present sufficiency conditions for correctness of individual control rules and show how it can be used to construct control modules with sensing actions. Special sensing actions also allow us to construct conditional plans from incomplete states where the robot does not have ....

....that the number of states in the worst case will be 3 n where n is the number of fluents if we use the approximate theory of sensing in [BS97a] It will be a much larger 2 2 n if the more general theories of sensing in [Moo85,SL93,LTM97] is used. It seems the conditional planner in [GW96] uses the theory of sensing in [BS97a] 12 We will need a modification of the planners in [EHW 92] for constructing minimal cost conditional plans. 37 8 From theory to practice: Our robot in the AAAI 96 contest We participated 13 in the AAAI 96 robot navigation contest [KNH97] Our ....

[Article contains additional citation context not shown here]

K. Golden and D. Weld. Representing sensing actions: the middle ground revisited. In KR 96, pages 174--185, 1996. 48

....verifier of control modules, an interpreter for the language of hierarchical control modules, and a compiler that would 64 efficiently generate control modules for classes of action theories. We hope to be able to use the conditional planner (with appropriate modifications) developed in [GW96] 12 Conclusion We will conclude our paper with two quotes by Dijkstra and Hoare, respectively, on programming methodologies and relate our results in this paper to their quotes. In [Dij72] Dijkstra says: The only effective way to raise the confidence level of a program significantly is to ....

K. Golden and D. Weld. Representing sensing actions: the middle ground revisited. In KR 96, pages 174--185, 1996. 68

....; r ; Delta ; p g then the choice between a and b will be based on full lookahead to the end of the program, i.e. the program will be executed essentially in the old way. 1.2 Sensing actions This on line style of execution is well suited to programs containing sensing actions. As described in [9,13,20], sensing actions are actions that can be taken by the agent or robot to obtain information about the state of certain fluents, rather than to change them. The motivation for sensing actions involves applications where because the initial state of the world is incompletely specified or because of ....

K. Golden and D. Weld. Representing sensing actions: the middle ground revisited. In Proc. KR-96, 1996.

....the answering machine displays a count of new messages, and the thermostat display indicates whether the furnace is on or off) While our implemented system depends on this assumption, our overall approach does not. By using a more complex, information gathering planner such as PUCC1NI or XII [9, 10, 11], our NLIA would operate correctly even without notification of these events. Interpreting a user s commands is more complex if there are multiple plans being executed at the same time. In this case a user s command can affect not just the appliance, but also the agent s execution stack of ....

Golden, K.; Weld, D. Representing Sensing Actions: The Middle Ground Revisited. KR, 1996.

....and recover from ambiguity, omissions, and errors in these requests. At its core, the softbot could handle goals specified in an expressive subset of first order logic with modal operators for handling time and for distinguishing information gathering goals from those that requested state changes [Golden and Weld, 1996] . We labeled the softbot a goal oriented interface, because human requests specified what the user wanted; the softbot was responsible for deciding how and when to satisfy the request. Since most users are uncomfortable with logical notation, we provided a forms oriented interface front end. ....

K. Golden and D. Weld. Representing sensing actions: The middle ground revisited. KR96, pages 174--185, 1996.

....We want our softbots to perform useful tasks for us. Ideally, we just say what we want, and rely on the softbot to figure out how to accomplish it. For example, I might ask my softbot to find a colleague s phone number or to fetch and print 1 Portions of this thesis were previously published in [34, 25]. 2 For readers unfamiliar with UNIX, a brief description of UNIX commands can be found in Section 1.4.2. 3 some papers from some online repository. Goals like these require the softbot to integrate multiple resources. There is no single command that will find and print papers from an online ....

K. Golden and D. Weld. Representing sensing actions: The middle ground revisited. In Proc. 5th Int. Conf. on Principles of Knowledge Representation and Reasoning, pages 174--185, 1996.

....for something to eat. However, most planners that deal with incomplete information don t behave much like Pinocchio. Most planners, by adopting some form of knowledge preconditions (Moore 1985) require the agent to know, a priori, that an action will have some desired result. As we argued in (Golden Weld 1996), and will briefly discuss here, these knowledge preconditions are representational handcuffs, which make action representations more awkward and limit the utility of our planners; our action language, sadl, eliminates them. In this paper, we show how a simple generalization of causal links allows ....

....quantified, and the quantifier takes the widest possible scope. For example, satisfy(in.dir (f , tex) T) means Ensure that there s at least one file in directory tex, and satisfy(in.dir (myfile, tex) tv) means Find out whether or not myfile is in tex. The initially annotation, introduced in (Golden Weld 1996), is similar to satisfy, but it refers to the time when the goal is given to the agent, not to the time when the goal is achieved. initially(P , tv) means that by the time the agent has finished executing the plan, 3 sadl stands for Sensory Action Description Language. it should know whether P ....

[Article contains additional citation context not shown here]

Golden, K., and Weld, D. 1996. Representing sensing actions: The middle ground revisited. In Proc. 5th Int. Conf. Principles of Knowledge Representation and Reasoning, 174--185.

No context found.

K. Golden and D. Weld, `Representing sensing actions: The middle ground revisited ', Knowledge Representation and Reasoning, 174--185, (1996).

No context found.

Golden, K. and Weld, D., Representing sensing actions: the middle ground revisited. Proc. KR'96, Morgan Kaufmann, 1996, pp. 174--185.

No context found.

K. Golden and D. Weld. Representing sensing actions: the middle ground revisited. In KR 96, pages 174--185, 1996.

No context found.

K. Golden and D.S. Weld. Representing sensing actions: the middle ground revisited. In Proc. of KR'96, 174--185.

No context found.

Golden, K. and D. Weld: 1996, `Representing Sensing Actions: The Middle Ground Revisited'. In: L. C. Aiello, J. Doyle, and S. C. Shapiro #eds.#: Principles of Knowledge Representation and Reasoning: Proceedings of the Fifth International Conference. Cambridge, MA, pp. 174#185.

No context found.

K. Golden and D.S. Weld. Representing sensing actions: the middle ground revisited. In Proc. of KR'96, 174--185.

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