M.E. Pollack. The uses of plans. Artificial Intelligence, 57, 1992.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....2 Activation and Attention Shift In the commitment based approach to attention shift, planning phases in which interruptions are allowed alternate with phases in which interruptions are not allowed regardless of what external events or stimuli take place. The architectures IRMA [10][25] and PRS [19] 18] provide a filtering mechanism that enables the agent to ignore external events. This filter can be overridden by external events in particular conditions. 26] in the case of IRMA and [23] in the case of PRS report experimentation with the commitment based approach using two ....

M.E. Pollack. The Uses of Plans. Artificial Intelligence, 57:43-68, 1992

....1 Introduction Much work has been done in Artificial Intelligence aimed at solving several problems concerned with the construction of autonomous agents. Part of this work has to do with the architecture of artificial reactive agents but adopts an engineering point of view (e.g. 3] 8] [15]) Some of it is more concerned with the definition of formal logics suited to represent the mental states of the agents and their reasoning capabilities (e.g. 4] 9] 17] Only a few exceptions try to learn useful guidelines from theories and experiments of cognitive psychology and cognitive ....

M.E. Pollack, "The uses of plans", Artificial Intelligence, Vol. 57, pp. 43-68, 1992

....merely that the agent should avoid moving the files unnecessarily, then we want the original solution, with some background preference to minimize unnecessary changes. Such background preferences could be expressed in terms of a utility function over world states [ 30 ] a measure of plan quality [ 28, 36 ] , or an explicit notion of harm [ 35 ] Note that even if we decide to forbid moving the files from tex, there are still other actions, such as deleting all the files in important papers, or sending threatening email to president whitehouse.gov that haven t been excluded. This is a general ....

Martha Pollack. The uses of plans. Artificial Intelligence, 57(1), 1992.

....and achieving its goals. A number of position papers and books argue persuasively for the advantages of the plans as not programs model [Agre and Chapman, 1990; Bratman, 1987; Bratman, Israel, and Pollack, 1988; Grosz and Sidner, 1988; Hayes Roth, 1985; 1992; Miller, Galanter, and Pribram, 1960; Pollack, 1990; 1992; Suchman, 1987] Several implemented systems actually use some form of plans that are not programs to control their cognitive or physical behavior [Chrisman and Simmons, 1991; Firby, 1992; Gat, 1992; Johnson and Hayes Roth, 1987; Newell, Shaw, and Simon, 1958; Tommelein, Hayes Roth, and Levitt, ....

Pollack, M. The uses of plans. Artificial Intelligence , 57, 1-42, 1992.

....under, a common recipe with respect to their joint goal. At any instant in time the recipe is likely to be partial either temporally partial in that the exact ordering between some elements may not be specified or structurally partial in that not all of the actions have been fully elaborated [60]. Developing or refining the recipe is a complex activity because of the inter dependencies which exist between the agents actions. Because of this complexity a number of different planning paradigms, each with their own benefits and drawbacks, have been devised [12, 21, 25, 30] for example, ....

M. E. Pollack, The Uses of Plans, Artificial Intelligence 57 (1992) 43-68.

....planning, the reader is advised to see Allen, Hendler and Tate [2] 5.3 Multi agent Planning Multi agent planning in DAI research has evolved along a number of different dimensions. One dimension focuses on agents that cooperate to solve subproblems of a given problem and integrate the results [21,20,30,24,114,36]. Other lines of inquiry concerns planning for task allocation, so that effective execution will result [141,95] centralized planning for multi agent execution [113,67] or centralized planning to avoid execution time action conflicts [50,51] Another area of research has concentrated on multiple ....

M. Pollack. The uses of plans. Artificial Intelligence, 57(1):43--68, 1992.

....very expensive. Simon introduced the idea of satisficing [Simon, 1981] arguing that a rational agent does not always have the resources to determine what the optimal action is, and instead should attempt only to make good enough, to satisfice. Some current work on planning (for example [Pollack, 1992]) is about the tradeoff between getting around to acting, and spending enough time thinking. Such resource bounded reasoning leads to suboptimal behavior. In our work we do not consider the tradeoff between acting and planning time. We acknowledge the computational cost of finding the optimal ....

Martha E. Pollack. The uses of plans. Artificial Intelligence, 57:43--68, 1992.

....between conjunctive goals, 2) the emotional agent to irrationally give up the pursuit of available goals (e.g. run away in fear) 5.2. Agent s Architectures Examples of agent s architectures that emphasize the importance of goal selection generation are PRS [Georgeff Ingrand 87] and IRMA [Pollack 92] In PRS, once active goals have been chosen by match against current beliefs in the system, a tree of potentially applicable plans are attached to them for concurrent execution. This can be simulated in our architecture by using steps 1,2,5,6 of Cognitive (i.e. removing the plan generation ....

Martha Pollack. The Use of Plans. Artificial Intelligence, (57):1-42, 1992..

....they come from turning requests from other agents into goals, but goals also come from reaction to a given situation. Thus our view bridges the gap between plan generation (goal generation by subgoal regression [Fikes Nilsson 71] and agent architecture (goal generation by specific modules [Pollack 92] We claim that a planning reactive agent built on such a control mechanism has the following properties: autonomy: the agent can generate (sub)goals from higher goals by goal regression, but can also generate goals without final goals coming from outside the agent; adaptability: the same ....

.... objects; 2) a proof of convergence: a particular action, even not obvious at first sight, will contribute to the desired goal given a correct planner [Chapman 87] In unpredictable environments, though, building plans is rather considered as concocting recipes towards particular (final) goals [Pollack 92] and executing them is rather considered as following methods that might fail because of unpredictable events [Eztioni 91] But, in our control mechanism, we will see that plan generation must be a goal oriented activity itself. Therefore the previous duality of view (plan generation plan ....

[Article contains additional citation context not shown here]

Martha Pollack. The Uses of Plans. Artificial Intelligence, (57):1-42, 1992..

....in achieving its goals. A number of position papers and books argue persuasively for the advantages of the plans are not programs model [Agre and Chapman, 1990; Bratman, 1987; Bratman, Israel, and Pollack, 1988; Grosz and Sidner, 1988; Hayes Roth, 1985; 1992; Miller, Galanter, and Pribram, 1960; Pollack, 1990; 1992; Suchman, 1987] Several implemented systems actually use some form of plans that are not programs to control their behavior [Chrisman and Simmons, 1991; Firby, 1992; Gat, 1992; Johnson and Hayes Roth, 1987; Newell, Shaw, and Simon, 1958; Tommelein, Hayes Roth, and Levitt, 1992] Many of the ....

Pollack, M. The uses of plans. Artificial Intelligence , 57, 1-42, 1992.

....topic in the theory of artificial intelligence: At the bottomline, knowledge and reasoning are valuable insofar as they enable a creature to accomplish its goals through appropriate action. Dav90] page 395. An overview of the various aspects of planning can be found in [GL87] According to [Pol92] in AI planning is considered to be the process of formulating a program of action This research is partially supported by ESPRIT III BRA project No.6156 DRUMS II . y also at the Katholieke Universiteit Nijmegen. to achieve some specified goal. Obviously, it is only possible for an agent ....

M.E. Pollack. The uses of plans. Artificial Intelligence, 57:43--68, 1992.

....Intelligence (AI) A review of the classical STRIPS planner is presented in section 2. It is based on Fikes and Nilsson s 1971 paper that describes STRIPS, complemented with information gathered from Russell and Norvig s comprehensive text on AI [7] and from Pollack s review about plans in AI [5]. A comparison to a different approach to the planning problem is presented in section 3; it is based on Pollack s work on partial order planning [6, 5] 2 A survey of STRIPS STRIPS (STanford Research Institute Problem Solver) 2] is a pioneering planning program developed around 1970 at SRI ....

.... STRIPS, complemented with information gathered from Russell and Norvig s comprehensive text on AI [7] and from Pollack s review about plans in AI [5] A comparison to a different approach to the planning problem is presented in section 3; it is based on Pollack s work on partial order planning [6, 5]. 2 A survey of STRIPS STRIPS (STanford Research Institute Problem Solver) 2] is a pioneering planning program developed around 1970 at SRI International, formerly known as Stanford Research Institute. The authors called it back then a problem solver , but planner is a more accurate ....

[Article contains additional citation context not shown here]

M. E. Pollack. The uses of plans. Artificial Intelligence, 57(1):43--68, 1992.

....for success of action goals and how these conditions can be achieved by communication with other agents. For example, see Appelt [1985] and Lochbaum [1994] However, to the extent that communicative goals are pursued on their own merit, a generation and filtering mechanism may be needed. Pollack s [1992] IRMA agent architecture provides an example of goal selection by filtering. Refine Communicative Goals into Multiple Coordinated Communicative Acts. An agent preparing to communicate may need to refine poorly specified goals into goals that are specific enough to act on, and coordinate the ....

M. Pollack. The uses of plans. Artificial Intelligence. 57:43-68, 1992.

....limited applicability. However, it lays the foundation on which to build a theory of reactive recognition. We present simple algorithms for means end plan recognition that make use of plans, similar to those used in reactive planning systems, to determine what means must be observed in 1 Pollack [16] writes: Yet, most research on plan recognition has taken place in isolation from the AI planning renaissance. Could a marriage of these two research projects bear any fruit 2 Analogously, we shall refer to the usage of plans to guide the planning process as means end plan execution. order ....

M. E. Pollack. The uses of plans. Artificial Intelligence, 57(1):43--68, 1992.

....has to arrive at an assessment of the cost involved in achieving it. Second, we insist that the task of evaluating an option should be computationally realizable; and in particular, our work here is developed within a theoretical framework first articulated in [3] and then further elaborated in [4, 19], that models resource bounded agents as operating always against the background of some current set of intentions, or plans, which helps to frame their subsequent reasoning. In contrast to the standard theory of rational choice, where actions are evaluated in isolation, we therefore develop a ....

Martha Pollack. The uses of plans. Artificial Intelligence, 57:43--68, 1992.

....make suboptimal decisions, lest they spend too long thinking, and fail to settle on a particular action in time for the execution of that action to be useful. It is therefore sometimes necessary for agents in dynamic environments to trade potential decision quality for decision timeliness [31]. The recent AI literature includes several approaches to managing this trade off with meta level control processes, which focus the agent s object level deliberation [6, 2, 15, 35, 30] Meta level control processes are intended to facilitate satisficing behavior [38] i.e. behavior that is ....

....messages. Intelligent agents in dynamic environments are sometimes better off disregarding certain messages. Despite the fact that this may lead to locally suboptimal behavior, it may be necessary to achieve behavior that is adequate overall. We have discussed this point at length elsewhere [2, 31, 32], and have proposed a mechanism, filtering to deal with it. We are currently exploring ways to combine a filtering mechanism with a traditional scheduling algorithm such as TDM. Other scheduling schemes that we are investigating are the Dynamic Priority algorithm (DP) of Nassehi et al. 26] and a ....

M. E. Pollack. The uses of plans. Artificial Intelligence, 57:43--68, 1992.

....work. Section 4 presents the experiments and discuss the results, which show the types of conditions under which filtering leads to improved performance, and the types of conditions that require restrictions on filtering. Finally, in Section 5, we summarize our results, and 1 Elsewhere Pollack [1992] discusses this claim in detail. The idea of trading decision quality for timeliness dates back to Simon s critique of decision theory and his development of the notion of satisficing [Simon 1955,Simon 1957] Good [1983] similarly distinguishes between the idealized notion of deliberation that is ....

Martha E. Pollack. The uses of plans. Artificial Intelligence, 57:43--68, 1992.

No context found.

M.E. Pollack. The uses of plans. Artificial Intelligence, 57, 1992.

No context found.

Pollack, M., "The Use of Plans", Artificial Intelligence, 43-68, 1992.

No context found.

M. E. Pollack. The uses of plans. Artificial Intelligence, 57:43--68, 1992.

No context found.

M.E. Pollack. The uses of plans. Artificial Intelligence, 57(1), 1992, pp. 43-68.

No context found.

M. Pollack, The uses of plans, Artificial Intelligence, 1992, 57(1): 43-68.

No context found.

M.E. Pollack. The uses of plans. Artificial Intelligence, 57(1), pp. 43-68, 1992.

No context found.

Martha E. Pollack. The uses of plans. Artificial Intelligence, 57(1):43-- 68, 1992.

No context found.

43-68.

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