High-Level Local Control Strategies for the Coordination of Large Groups of Mobile Autonomous Agents

点击翻译按钮获取中文摘要

Intellectual Merit: Remarkable advances have been made in recent years in the development of small,agile, relatively inexpensive mobile autonomous vehicles and sensors, ultimately intended for a wide variety ofpurposes such as search and rescue, exploration, environmental monitoring, distributed sensing, the cleaningand maintaining of structures, etc. The existence of such devices and the anticipated development of still moreadvanced versions raises compelling questions: Can large numbers of small autonomous vehicles/sensors besuccessfully deployed in the form of a swarm or flock as a search team, an exploration group, or somethingsimilar, to cooperatively carry out a prescribed task and to respond as a group to high-level managementcommands? Can such a group reliably function with a centralized supervisor, with limited communicationbetween members, with limited power capabilities and/or with the roles of its members not all the same andperhaps changing with time. In broad terms these are the issues which motivate this proposed research.In particular we propose two pronged assault on the issue of coordinating large groups of mobile autonomousagents/mobile sensors. First, our intention is to continue the work we are already doing with NSFsupport, on the applications of graph rigidity to the maintaining of vehicle formations and to the localizationof sensors in a large, possible mobile network. Second we propose to continue studying prototype synchronizationand formation forming problems such as flocking and rendezvous, which we considered in priorwork, with the objective of uncovering new principles of cooperative control and new means to analyze suchsystems. In the former category, we plan to take a hard look at how one might modify the rigidity conceptso as to make in applicable to leader-follower formations. We will also look carefully at just what is requiredglobal rigidity wise, to enable one to determine the position of a single sensor in a non-rigid network. Wewill address all aspects of the distance constrained sensor network localization problem, in an e.ort to devisea provably correct distributed localization algorithm, for a moderately general class of networks.We will explore synchronization problems and group pattern forming problems using simple high levelpoint models. Our aim here is not so much to solve the problems we will pose as it is to rely on them tosuggest interesting questions about cooperative control, which are potentially applicable to more realisticmodels in more realistic settings. We will look at rate of convergence issues. We will attempt to generalizeand formalize the concept of analytic synchronization with the goal of creating a means of analyzing a largeclass of asynchronous multi-agent motion control problems. We will be particularly interest in the structureof the switched dynamical systems which result due to the changing of neighbors in a mobile network ofvehicles. What we want to know is if there are inherent structural properties we can exploit {as we did in the.ocking problem} to obtain convergence results, so as to avoid the NP hardness associated with convergencequestions with typical switched dynamical systems.Broader Impacts: This project will advance discovery and understanding while promoting teaching, trainingand learning in at least two di.erent ways. First, graduate students involved in the project will beexpected to attend and present there results at technical meetings. Second, project graduate students willcontribute to and participate in a short course on the topic which we envision organizing and giving towardsthe end of the project. We cannot think of any especially unusual way in which this project might broadenparticipation of under represented groups, although we certainly would encourage such participation. Thisproject will enhance infrastructure for research and education in at least two ways. First we expect tocontinue our cross-disciplinary collaboration with our environmental biologist colleagues with whom we'vealready been working for several years. Second, we expect to continue to collaborate with our experimentalistcolleagues at Yale and UCSB who are building a mobile sensor network test beds for trying out variouscontrol algorithms. We hope to continue to contribute to broadening dissemination to enhance scienti.cand technological understanding by organizing and running boni .de interdisciplinary technical gatheringslike the Block Island Workshop on Cooperative Control we co-organized and which took take place in June,2003. We also expect to give overview talks on the subject such as the plenary presentation we gave lastJuly at the MTNS meeting in Belgium, and the Penner Lecture we gave last November to the engineeringfaculty at UCSD.