Firby, Issues and architectures for planning and execution. Proc. Workshop Innovative Approaches Planning, Scheduling, and (1990).  Home/Search   Document Not in Database   Summary   Related Articles   Check

....reasoning (planning) We are left with the problem of synchronizing the high level reasoner with the lower levels controlling the robot. Hanks and Firby accomplish this by interfacing a fairly traditional high level planner with low level reactive behaviors used as primitive actions [Firby, 1987, Hanks and Firby, 1990] The Phoenix project [Cohen et al. 1988, Howe et al. 1990] uses an agent architecture that integrates planning and acting by implementing a two level system: the higher level cognitive component modifies the agent s behavior at irregular intervals, while a reflexive system provides ....

....conditions. One conse quence of this more general formulation is that projection may be very expensive. General probabilistic reasoning is known to be intractable [Cooper, 1987] A small but growing body of work on probabilistic temporal reasoning [Berzuini et al. 1989, Dean and Kanazawa, 1988, Hanks, 1990, Haddawy, 1990] holds out some hope, however, of doing restricted forms of probabilistic projection efficiently. Efficiently should not be taken to mean with negligible cost, however. One possibility would be to define an anytime decision procedure for doing probabilistic prediction and then ....

[Article contains additional citation context not shown here]

Steve Hanks and 1. James Firby. Issues and architectures for planning and execution. In Proceedings of the DARPA Workshop on Innovative Approaches to Planning, Scheduling, and Control. DAIPA, 1990.

.... As noted in the introduction, detection, classification, and recovery from sensing failures in mobile robots has been addressed by (Noreils Chatila 1995) Ferrell 1993) and (Payton et al. 1992) Other noteworthy efforts are those by (Weller, Groen, Hertzberger 1989) Velde Carignan 1984) (Hanks Firby 1990), and (Chavez Murphy 1993) Weller, Groen, Hertzberger 1989) and (Velde Carignan 1984) deal with sensor errors in general. Weller, Groen, Hertzberger 1989) create modules for each sensor containing tests to verify the input based on local expert knowledge. Environmental conditions ....

....Environmental conditions determine whether a test can be performed or not. The partitioning of problem space by symptom is based on these modules. The approach taken in this paper follows (Weller, Groen, Hertzberger 1989) testing corroborating sensors before using them for error classification. (Hanks Firby 1990) propose a planning architecture suitable for mobile robots. As with (Noreils Chatila 1995) a plan failure triggers exception handling. The system recovers by either choosing another method randomly whose pre conditions are currently satisfied (similar in concept to logical sensors (Henderson ....

Hanks, S., and Firby, R. 1990. Issues and architectures for planning and execution. In Proceedings of a Workshop on Innovative Approaches to Planning, Scheduling and Control, 59--70.

....for recovery, mainly because there is no obvious method for adjusting the invalid data. Further, the method should be expanded to include modifications or removal of sensors themselves if they are malfunctioning. 2. 4 Firby and Hank s hybrid deliberative reactive architecture Firby and Hanks [4] have presented a planning architecture that addresses exception handling. This architecture joins Firby s reaction action packages (RAPs) 2] with Hanks deliberation system. The action component handles the execution of plans and addresses exception handling for plan failures. Two types of plan ....

Hanks, S., Firby, R. J., "Issues and Architectures for Planning and Execution." Proceedings of a Workshop on Innovative Approaches to Planning, Scheduling and Control, San Diego, CA, November 5-8, 1990.

....at the beginning of the planning epoch. As the world changes, this model becomes increasingly out of date. Currently, we ignore this drift, and hope that an out of date plan is better than no plan at all. 13] This is in fact what many researchers refer to as integrated planning and reacting [9, 10, 12, 14], but quite evidently is not fully integrated. 4. Integrated view We propose a view of planning and reacting where planning receives equal billing and is in fact fully integrated with the reactor, as illustrated in Figure 3. Here there is a continual exchange of information between the planner ....

S. Hanks and R. J. Firby. Issues and architectures for planning and execution. In K. Sycara, editor, Innovative Approaches to Planning, Scheduling, and Control. Morgan Kaufmann, 1990.

....1987; Beetz and McDermott, 1992; Firby, 1992; Georgeff, 1991; Hayes Roth, 1985; Laird et al. 1987; Simmons, 1991a ] All these systems address the issue of integrating reactivity and reasoning capabilities, even if each of them uses quite different techniques. RAP [ Firby, 1987; Firby, 1992; Hanks and Firby, 1990 ] is used in a partitioned architecture where a strategic planner interacts with a completely reactive system (RAP Executor) through a shared world model and plan representation. One of the advantages of this architecture is the fact that computational costs are reduced with respect to uniform ....

S. Hanks and R. J. Firby. Issues and Architectures for Planning and Execution. In Proceedings of the Workshop on Innovative Approaches to Planning, Scheduling and Control, pages 59--70, 1990.

.... referred to as reactive, situated, or behaviour based typically make all necessary control decisions at run time on the basis of limited amounts of information (usually only that which is currently availablefrom their sensors) limited internal state, and with a minimal amount of inference [HF90] 11 Whereas traditional planners are required to produce optimal or correct actions, non deliberative architectures are designed to produce robust actions. Typically built from relatively simple control mechanisms for example, finite state machines (FSMs) or domain specific ....

....the agent s optimal problem solving strategy, information action (if any) and base level motor action. 2.5 Intelligent Agency The Issues An autonomous agent operating in a complex environment is constantly faced with the problem of deciding what action to take next. As Hanks and Firby [HF90] point out, formulating this problem precisely can be very difficult since it necessitates consideration of a number of informational categories which are often difficult to ascertain for example, the benefits and costs to the agent of executing particular actions sequences; or which have ....

[Article contains additional citation context not shown here]

Steve Hanks and R. James Firby. Issues and architectures for planning and execution. In Proceedings DARPA Workshop on Innovative Approaches to Planning, Scheduling and Control, pages 59--70. Morgan Kaufmann: San Mateo, CA, 1990.

.... modules, on the ground that these two essentially different problems call for essentially different representation and reasoning mechanisms (e.g. Arkin, 1990; Gat, 1991 ] However, in order to ensure integration, some authors have insisted on the importance of a shared plan representation [ Hanks and Firby, 1990 ] i.e. a representation that is both adequate for specifying complex reactive controls, and for being reasoned about by higher level deliberative processes. The proposal I sketch below is meant to illustrate this point. Sketch of an integration proposal I outline here a proposal for a ....

Hanks, S. and Firby, R. J. 1990. Issues and architectures for planning and execution. In Workshop on Innovative Approaches to Planning, Sheduling and Control, San Diego, CA.

....This plan can be either hand coded, or can be generated by a planner for each given task. In our experiments, we used a simple goal regression planner to automatically generate navigation plans. The architecture in Figure 1 is an instance of a two level type of architecture ubiquitous in robotics [6]. However, two important aspects differentiate our approach from the most of the others. First, behaviors are typically assumed to be implemented by executable procedures that directly control robot s effectors, and arbitration between competing behaviors is implemented by giving complete control ....

S. Hanks and R. J. Firby. Issues and architectures for planning and execution. In Workshop on Innovative Approaches to Planning, Sheduling and Control, San Diego, CA, 1990.

....connect action, locally controlled by the execution process, to the overall goals of the agent, globally analyzed by the planning process. The heart of this connection is the plan: the connection will be successful if the execution layer can make good use of the plan. As noted by Hanks and Firby [47], an essential aspect of the integration problem is the definition of a shared plan representation, that is, a representation that can be reasoned about and generated by the high level processes, and can be effectively used by the low level processes to control execution. The definition of a ....

....call for different types of tools, and pose different requirements on a plan representation. For example, many path planners generate paths with sharp turns, which are not adequate for execution; and many robot execution systems are based on plans written in elaborated reactive languages (e.g. [47, 31]) which are too complex to be generated by current planners. Fuzzy logic has an extremely attractive feature in this respect: its ability to represent both the symbolical and the numerical aspects of reasoning. Fuzzy logic can be embedded in a full logical formalism, endowed with a symbolic ....

S. Hanks and R. J. Firby. Issues and architectures for planning and execution. In Procs. of the DARPA Workshop on Innovative Approaches to Planning, Sheduling and Control, San Diego, CA, 1990.

....the executor will suspend problematic execution threads, request the planner to modify the plan, continue execution of nonproblematic actions while waiting for the planner, implement lower level behaviors while waiting, and restart execution when a modified plan is received. Hanks and Firby [11] discuss the issues involved in integrating planning and execution, including issues not addressed in this paper, such as reasoning about the utility of deliberation. They do not present an implemented formalism beyond RAP. McDermott [15] describes RPL, a Lisp like programming language for writing ....

Steve Hanks and R. James Firby. Issues and architectures for planning and execution. In Katia P. Sycara, editor, Proceedings of the Workshop on Innovative Approaches to Planning, Scheduling and Control, pages 59--70. Morgan Kaufmann Publishers Inc., San Mateo, CA, November 1990.

....long loop. Also, the RAP system cannot think ahead. Both these limitations point to the need for an extra layer of control that places constraints on RAP behaviour prior to execution; in other words, neither urgency or uncertainty obviate the need for more deliberative decision making. Hence, in (Hanks Firby, 1990) the RAP system is extended by considering the extra, deliberative layer of control needed in an autonomous agent. The addition of planning ahead and reasoning abilities generates two new design problems: how to deliberate, and how to coordinate deliberation and reactive execution. This problem is ....

Hanks, S. & Firby, R. J. (1990). Issues and architectures for planning and execution.

....The question of direct feedback when some request by the AI can not be fulfilled by the real time system is not addressed. Also, MARUTI assumes a single processor architecture, which greatly reduces the complexity of the issues. The general issues of deliberating versus acting are discussed in [5]. They note that neither thinking ahead nor acting at the last moment should be pursued to the exclusion of the other. The high level scheduler must be aware of this issue when it submits process groups to the low level scheduler. If the process groups are submitted too far in advance, the state ....

Steve Hanks and R. James Firby. Issues and architectures for planning and execution. In Proceedings of the Workshop on Innovative Approaches to Planning, Scheduling and Control, pages 59--70, November 1990.

.... real time AI focuses either on restricted AI techniques that have predictable performance characteristics [4, 19, 23] or on reactive systems that retain little of the power of traditional AI [1, 5] Several researchers are investigating systems which combine reactive and traditional AI methods [2, 14, 31, 35]. These approaches have concentrated on retaining both reactive and unpredictable mechanisms, but do not address the guarantees required by hard real time tasks. To combine unrestricted AI techniques with the ability to make hard performance guarantees, we propose a Cooperative Intelligent ....

....very similar to CIRCA s guaranteed TAP schedules. They too are investigating the mechanisms by which a planning system can generate real time task requirements. However, by restricting the system to a single processor, they exacerbate the complex issues of trading off action and deliberation [14, 29]. A refinement of the single processor approach uses iterative improvement algorithms to guarantee that the intelligent system can be interrupted at any time and will still yield a solution, possibly with reduced precision or confidence [4, 19] Such anytime algorithms cannot provide 4 Sets of ....

[Article contains additional citation context not shown here]

S. Hanks and R. J. Firby, "Issues and Architectures for Planning and Execution," in Proc. Workshop on Innovative Approaches to Planning, Scheduling and Control, pp. 59--70, November 1990.

....system s operation. 6.2 Planned Reactions, Proven Safety As noted earlier, many reactive AI systems have been composed of manually engineered reaction plans. Some systems have been designed with higher level reasoning processes that select which of the available reactive elements are active [3, 6, 13, 16, 30]. Other reactive systems are designed, like CIRCA, to automatically generate reaction plans from primitive component descriptions [29] Performing this type of reaction planning is similar to classical planning in the sense that it is done before the plan is executed, and usually involves ....

S. Hanks and R. J. Firby, Issues and architectures for planning and execution, in: Proc. Workshop on Innovative Approaches to Planning, Scheduling, and Control (1990) 59--70.

....theory of agency in and of itself: intelligent agents can predict at least some aspects of the execution time world, and can therefore avoid obviously short sighted behaviors painting oneself into a corner, for example. So recently the notion of an agent architecture has come into vogue [Hanks and Firby 1990], Ingrand and Georgeff 1990] Bresina and Drummond 1990] a system that can both make predictions about, and act in the world, as appropriate. At the same time researchers turned their attention to more realistic planning domains, another issue arose, that of validation. What constitutes a ....

....of researchers in the field nowadays: Exogenous Events We mentioned above that perhaps the most limiting assumption of the Blocksworld was the fact that no exogenous, or unplanned, events could occur. Relaxing this assumption makes the process of predicting the effects of plans more difficult [Hanks 1990b] and also introduces the concept of reacting to unplanned events as they occur at execution time [Agre and Chapman 1987] Firby 1989] It also makes the passage of time more important the longer one waits, the more the world changes which brings up a number of issues having to do with the ....

[Article contains additional citation context not shown here]

Steve Hanks and R. James Firby. Issues and architectures for planning and execution. In Workshop on Innovative Approaches to Planning, Scheduling, and Control, November 1990.

No context found.

Firby, Issues and architectures for planning and execution. Proc. Workshop Innovative Approaches Planning, Scheduling, and (1990).

No context found.

S. Hanks and R. J. Firby. Issues and architectures for planning and execution. In Workshop on Innovative APproaches to Planning, Scheduling and Control, San Diego, CA, 1990.

No context found.

S. Hanks and R. J. Firby. Issues and architectures for planning and execution. In Proceedings of the Workshop on Innovative Approaches to Planning, Scheduling and Control, pages 59--70, San Diego, California, November 1990.

No context found.

S. Hanks and R. J. Firby. Issues and Architectures for Planning and Execution. In Proceedings of the Workshop on Innovative Approaches to Planning, Scheduling and Control, pages 59--70, 1990.

No context found.

Hanks 1990 Hanks, S., and R. J., "Issues and Architectures for Planning and Execution," in Sycara, K. (ed.), Workshop on Innovative Approaches to Planning, Scheduling, and Control, Defense Advanced Research Projects Agency, November, 1990, pp. 59-70, Morgan Kaufmann, San Mateo, CA, 1990.

No context found.

Hanks, S., and Firby, J. R., "Issues and Architectures for Planning and Execution", Proc. of Conf. on Innovative Approaches to Planning, Scheduling and Control, San Diego, California, pp. 59-70, November, 1990.

No context found.

HANKS90 Hanks, S., Firby, R. J., Issues and Architectures for Planning and Execution, DARPA Workshop on Innovative Approaches to Planning, Scheduling and Control, San Mateo, CA, November 1990.

No context found.

HANKS90 Hanks, S., Firby, R. J., Issues and Architectures for Planning and Execution, DARPA Workshop on Innovative Approaches to Planning, Scheduling and Control, San Mateo, CA, November 1990.

No context found.

Hanks, S., & Firby, R. J. (1990). Issues and architectures for planning and execution. In Proceedings of a Workshop on Innovative Approaches to Planning, Scheduling and Control, San Diego, CA: DARPA.

No context found.

Steve Hanks and R. James Firby. Issues and architectures for planning and execution. In DARPA Workshop on Innovative Approaches to Planning, Scheduling and Control, pages 59-70, 1990.

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