📄 http:^^ai.eecs.umich.edu^people^durfee^circa.html
字号:
Date: Tue, 14 Jan 1997 23:29:46 GMT
Server: NCSA/1.5.1
Last-modified: Fri, 06 Sep 1996 13:33:17 GMT
Content-type: text/html
Content-length: 12302
<HTML><HEAD><title>CIRCA</title></HEAD><BODY><H1>CIRCA...</H1>is an architecture for combining artificial intelligence andreal-time systems to achieve intelligent real-time control,particularly of physical (robotic, manufacturing, aviation)systems. The research is conducted jointly between the<!WA0><A HREF="http://ai.eecs.umich.edu/">Artificial Intelligence Laboratory</A> and the <!WA1><A HREF="http://www.eecs.umich.edu:80/RTCL/">RealTime Computing Laboratory</A> <P><H1>Personnel</H1><!WA2><A HREF="http://ai.eecs.umich.edu/people/durfee/durfee.html">EdmundH. Durfee, Associate Professor of EECS</A> <P><!WA3><A HREF="http://www.eecs.umich.edu:80/RTCL/kgshin/index.html">KangG. Shin, Professor of EECS</A> <P><!WA4><A HREF="http://ai.eecs.umich.edu/people/marbles/homepage.html">EllaAtkins, Graduate Student</A> <P><!WA5><A HREF="http://www-personal.umich.edu/~mcvey/">ChipMcVey, Graduate Student</A> <P><H2>Alumni</H2><!WA6><A HREF="http://www.cs.umd.edu/users/musliner/index.html">DavidJ. Musliner</A>, PhD, now at Honeywell Technology Center <P><A>Eric Miller, MS, now at Loral.</A><H2>Some Publications</H2>E. M. Atkins, E. H. Durfee, K. G. Shin, <!WA7><A HREF="http://ai.eecs.umich.edu/people/marbles/prob_paper.html">Plan Development Using Local Probabilistic Models</A>, Proceedings of the Twelfth Conference on Uncertainty in Artificial Intelligence, pp. 49-56, August 1996.<p>E. M. Atkins, E. H. Durfee, K. G. Shin, <!WA8><A HREF="http://ai.eecs.umich.edu/people/marbles/papers/unhandled.abs.ps">Expecting the Unexpected: Detecting and Reacting to Unplanned-for World States</A>, Proceedings of AAAI-96, pp. ????, August 1996. (student abstract)<p>E. M. Atkins, E. H. Durfee, K. G. Shin, <!WA9><A HREF="http://ai.eecs.umich.edu/people/marbles/papers/action.workshop.ps">Detecting and Reacting to Unplanned-for World States</A>, AAAI-96 Workshop on Theories of Action, Planning, and Robot Control: Bridging the Gap, pp. 7-14, August 1996.<p>E. M. Atkins, E. H. Durfee, K. G. Shin, <!WA10><A HREF="http://ai.eecs.umich.edu/people/marbles/papers/temporal.workshop.ps">Building a Plan with Real-Time Execution Guarantees</A>, AAAI-96 Workshop on Structural Issues in Planning and Temporal Reasoning, pp. 1-6, August 1996.<p>E. M. Atkins, E. H. Durfee, K. G. Shin, <!WA11><A HREF="http://ai.eecs.umich.edu/people/marbles/papers/plan.exec.symp.ps">Detecting and Reacting to Unplanned-for World States</A>, to appear in AAAI-96 Fall Symposium on Plan Execution: Problems and Issues Technical Report, November 1996.<p>E. M. Atkins, E. H. Durfee, K. G. Shin, <!WA12><A HREF="http://ai.eecs.umich.edu/people/marbles/papers/flex.comp.symp.ps">Achieving Fully-Automated Aircraft Flight with Limited Resources</A>, to appear in AAAI-96 Fall Symposium on Flexible Computation in Intelligent Systems: Results, Issues, and Opportunities Technical Report, November 1996.<p>D. J. Musliner, J. A. Hendler, A. K. Agrawala, E. H. Durfee,J. K. Strosnider, and C. J. Paul,<!WA13><a HREF="http://www.cs.umd.edu/users/musliner/papers/rtai.ps">The Challenges of Real-Time AI</a>,<i>IEEE Computer</i>, Vol 28 #1, January 1995.Also appears as University of Maryland Technical Report CS-TR-3290(UMIACS-TR-94-69).<p>D. J. Musliner,<!WA14><a HREF="http://www.cs.umd.edu/users/musliner/papers/salsa.ps"> Using Abstraction and Nondeterminism to Plan Reaction Loops</a>,<i>Proc. National Conf. on AI</i>, pp. 1036-1041, Seattle WA, August 1994.<p>D. J. Musliner,<!WA15><a HREF="http://www.cs.umd.edu/users/musliner/papers/cirffss94.ps"> Predictive Sufficiency and the Use of Stored Internal State</a>,in <i>Proc. Conf. on Intelligent Robotics in Field, Factory, Service, and Space</i>, pp. 298-305, Houston TX, March 1994.<p>D. J. Musliner, E. H. Durfee, and K. G. Shin,<!WA16><a HREF="http://www.cs.umd.edu/users/musliner/papers/musliner-aij.ps"> World Modeling for the Dynamic Construction of Real-Time Control Plans</a>,to appear in <i>AI Journal</i>, 1994.<p>D. J. Musliner, K. G. Shin, and E. H. Durfee,<!WA17><a HREF="http://www.cs.umd.edu/users/musliner/papers/airtc94.ps"> Automating the Design of Real-Time Reactive Systems</a>,in <i>Proc. Symposium on AI in Real-Time Control</i>, 1994.<p>D. J. Musliner, E. H. Durfee, and K. G. Shin,<!WA18><a HREF="http://www.cs.umd.edu/users/musliner/papers/tsmc.ps">CIRCA: A Cooperative Intelligent Real-Time Control Architecture</a><i>IEEE Transactions on Systems, Man, and Cybernetics</i>, Vol 23 #6, 1993. </a><p>D. J. Musliner, <!WA19><a HREF="http://www.cs.umd.edu/users/musliner/papers/musliner-diss.ps">CIRCA: The Cooperative Intelligent Real-Time Control Architecture</a><i>Ph.D. Thesis</i>, The University of Michigan, Ann Arbor, MI, 1993. </a><p>D. J. Musliner, E. H. Durfee, and K. G. Shin,<!WA20><a HREF="http://www.cs.umd.edu/users/musliner/papers/musliner-sigman.ps"> Integrating Intelligence and Real-Time Control into Manufacturing Systems</a>,<i>Working Notes of the SIGMAN Workshop on IntelligentManufacturing Technology</i>, July 1993.<p>D. J. Musliner, E. H. Durfee, and K. G. Shin,<!WA21><a HREF="http://www.cs.umd.edu/users/musliner/papers/musliner-rtoss.ps"> Any-Dimension Algorithms</a>,in <i>Proc. Workshop on Real-Time Operating Systems and Software</i>, May 1992.<p>D. J. Musliner, E. H. Durfee, and K. G. Shin,<!WA22><a HREF="http://www.cs.umd.edu/users/musliner/papers/musliner-ss92.ps"> Reasoning About Bounded Reactivity to Achieve Real-Time Guarantees</a>, in<i> Proc. AAAI Spring Symposium on Selective Perception</i>, March 1992.<p>D. J. Musliner, E. H. Durfee, and K. G. Shin,<!WA23><a HREF="http://www.cs.umd.edu/users/musliner/papers/caia91.ps"> Execution Monitoring and Recovery Planning with Time</a>, in<i> Proc. Conf. on Artificial Intelligence Applications</i>, February 1991.<p>Edmund H. Durfee and Victor R. Lesser. Incremental Planning to Control a Time-Constrained, Blackboard-Based Problem Solver. <I>IEEE Transactions on Aerospace and Electronic Systems</I>, special issue on space telerobotics, 24(5):647-662, September 1988.<p>Victor R. Lesser, Jasmina Pavlin, and Edmund H. Durfee. Approximate Processing in Real-Time Problem Solving. <I>AI Magazine</I>, Vol. 9, No. 1, pages 49--61, Spring 1988.<p>Edmund H. Durfee. Towards Intelligent Real-Time Cooperative Systems. In <I>AAAI Spring Symposium on Planning in Uncertain, Unpredictable, or Changing Environments</I> , pages 29--33, Stanford, CA, March 1990.<p><H2>Short Description</H2>As Artificial Intelligence (AI) techniques become mature, there has beengrowing interest in applying these techniques to controlling complexreal-world systems which involve hard deadlines. In such systems, thecontroller is required to respond to certain inputs within rigiddeadlines, or the system may fail catastrophically. For example, theMars Rover project requires a partially- or fully-autonomous vehiclethat can perform unsupervised navigation in hazardous conditions, whereerrors could mean the loss of multi-million dollar research anddevelopment efforts. Controlling the Rover includes real-time taskssuch as obstacle avoidance and emergency reactions to unexpected terrainhazards. If the system ever fails to meet the deadlines associated withthese control tasks, it may suffer damage tantamount to mission failure.Since the number of possible domain situations is too large to be fullyenumerated, and the consequences of failure are so severe, testing aloneis insufficient to guarantee the required real-time performance. Thesecontrol problems require systems which can be proven to meet thehard deadlines which the environment imposes. <p>Hard-wired control schemes using fixed algorithms are amenable to suchperformance analysis, but cannot address high-level problems such asreasoning about goals, resource restrictions, and recovery fromunexpected failures. Unfortunately, many of the AI techniques andheuristics developed to solve these high-level problems are not suitedto analyses that would provide guaranteed response times. For example,systems that learn are able to form new chains of inferences, resultingin changing performance characteristics that may defy worst-casebounding. Even when AI techniques can be shown to have predictableresponse times, the variance in these response times is typically solarge that providing timeliness guarantees based on the worst-caseperformance would result in severe underutilization of the computationalresources during normal operations. <p>Thus we perceive an apparent conflict between the nature of AI and theneeds of real-world, real-time control systems. While AI methods arecharacterized by unpredictable or high-variance performance, real-timecontrol systems require constant, predictable performance. Mostresearch on ``real-time AI'' (RTAI) focuses either on restricted AItechniques that have predictable performance characteristics or onreactive systems that retain little of the power of traditional AI. <p>The AI Lab and the Real-Time Computing Lab are cooperating on anew branch of RTAI research here at the University of Michigan. Tocombine unrestricted AI techniques with the ability to make hardperformance guarantees, we are investigating a Cooperative IntelligentReal-Time Control Architecture (CIRCA). In this architecture, an AIsubsystem reasons about task-level problems that require its powerfulbut unpredictable reasoning methods, while a separate real-timesubsystem uses its predictable performance characteristics to deal withcontrol-level problems that require guaranteed response times. The keydifficulty with this approach is allowing the subsystems to interactwithout compromising their respective performance goals. We havedeveloped a scheduling module and a structured interface that allow theunconstrained AI subsystem to asynchronously direct the real-timesubsystem without violating any response-time guarantees. <p>Realistic intelligent control systems must recognize their resourcelimitations and make tradeoffs in the quality of their control outputs,or responses. Many systems recognize resource limitations and trade offthe precision, confidence, or timeliness of their responses. CIRCAextends this mechanism by allowing the system to explicitly trade offthe completeness of its responses. CIRCA's AI subsystem and schedulercooperatively reason about the real-time subsystem's executionresources, and choose a subset of responses that the real-timesubsystem will guarantee. By manipulating the responses that thereal-time subsystem is guaranteeing, the AI subsystem attempts to ensurethat the real-time subsystem will meet hard deadlines and also achievethe overall system goals. CIRCA also provides mechanisms to utilize thetime which becomes available when guaranteed mechanisms use less thantheir worst-case scheduled time allowance. <p>To achieve flexible control, CIRCA requires that the AI methods reasonabout the expected real-time demands of the environment and buildcontrol plans to guarantee meeting those demands. CIRCA does this usinga formal graph-based model of agent/environment interactions, exploringa space of states that the system could be in due to its own actions,due to external events, and due to the passage of time. In constructingcontrol plans, CIRCA determines what actions it must guarantee to takeand how often it will be able to take them to ensure that the systemdoes not enter a state where it could transition into failure (due tothe passage of time). Currently, CIRCA is able to develop such controlplans when possible, and when not possible CIRCA provides well-definedtransformations to the graph model (based predominantly on eliminatingor extending various types of transitions) that allow it tosystematically relax requirements until it can guarantee the performanceof a control plan. Our ongoing work is investigating how to choose fromamong candidate transformations to yield the best possible control plan.We are also investigating improved scheduling techniques for efficientlygenerating guaranteed control plans, using internal state in thereal-time subsystem to reduce costly sensory actions, and strategies fortransitioning among control plans. Application domains for CIRCAinclude manufacturing process control and mobile robotics. <p></BODY><HR><ADDRESS><B>Last Updated: </B><I>5/4/95</I></ADDRESS>⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -