Fourteenth National Conference onArtificial Intelligence (AAAI-97)

Ninth Conference on Innovative Applications ofArtificial Intelligence (IAAI-97)

July 27-31, 1997

Sponsored by the American Association for Artificial Intelligence

Featuring: An opening keynote address by James Allen (University of Rochester) § Threedays of technical paper presentations by top scientists in the field § A three-day paralleltrack of invited talks on topics ranging from the emergence of spacecraft autonomy, to AI

and the changing concept of computation § Innovative Application Award winners § Invitedtalks and presentations on emerging AI applications and technologies § The Mobile Robot Com-petition, where robots will challenge each other with their latest earth-bound skills at vacu-uming rooms, finding TV remote controls, and serving hors d’ouevres, and their latest extra-terrestrial skills at discovering life on Mars § The Hall of Champions, where many of the bestcomputer players of classic games of strategy including chess, checkers, bridge, scrabble, andbackgammon will play a series of challenge matches against some of the best human players§ A continuing education program, the Tutorial Forum, where one ticket allows attendanceat two days of courses on the latest developments in sub-fields of AI and related technologies§ Joint Exhibition § Student Abstract and Poster program, and associated SIGART/AAAI Doc-toral Consortium program § Workshops on a variety of research topics (invitation only).

Cou

rtes

y, P

rovi

denc

e / W

arw

ick

Con

vent

ion

& V

isito

rs B

urea

u

AAAI-97: A PreviewAs in years past, AAAI-97 continues to servethree major roles in the community:• Satisfying the needs of current and future

technical specialists through 13 workshopsand 50 technical sessions, the StudentPoster Display, and the SIGART/AAAIDoctoral Consortium

• Providing continued professional educationthrough the Tutorial Forum and keynoteand invited speakers, including reports onhighlights from recent UAI, ML, KR andKDD conferences

• Educating the public (while stimulating on-going research projects) through the Hall ofChampions and Robot Competition.The Hall of Champions is new to AAAI-97.

At the Hall will be demonstrations of the bestcomputer players of a variety of classic games ofstrategy. AAAI-97 attendees and the publicwill be able to compete with these programs,discuss technical and social issues with theirauthors, and view them playing a series of chal-lenge matches against some of the best humanplayers around.

The popular Tutorial Forum, introduced atAAAI-96, also continues this year, with admis-sion allowing participation in up to four con-secutive tutorials during two full days. Thisyear, among other topics will be “Belief Net-works and Decision-Theoretic Reasoning forAI,” “Mobile Robot Control Architectures,”“Physics-Based Modeling for Vision and Virtu-al Human Animation,” and “Data Mining.”

We will also be continuing the popular Stu-dent Abstract and Poster program, an opportu-nity for students to present and discuss theirwork during its early stages, meet peers with re-lated interests, and introduce themselves tomore senior members of the field. It presents awonderful opportunity for all of us to get ac-quainted with some of the up-and-coming tal-ent and their new research ideas. Also for stu-dents is the SIGART/AAAI DoctoralConsortium program—a small, focused gather-ing that allows selected students to presenttheir work to a faculty panel and their peers fordiscussion and practical advice.

In sum, the AI community continues its goalof enhancing the atmosphere of excitement,innovation, controversy, and intellectual en-gagement at its annual conference. Please joinus by taking part in AAAI-97!

- Ben Kuipers and Bonnie WebberProgram Cochairs, AAAI-97

IAAI-97: A PreviewThe Ninth Annual Conference on InnovativeApplications of Artificial Intelligence (IAAI-97), held Monday – Wednesday, July 28-30,continues the IAAI tradition of case studies ofdeployed applications with measurable benefitswhose value depends on the use of AI technol-ogy. In addition, IAAI-97, for the first time,augments these case studies with papers that ad-dress emerging areas of AI technology or appli-cations. It also includes a series of invited talksorganized around this same theme of emergingareas. IAAI is organized as an independent pro-gram within the National Conference, and theschedules are coordinated, to allow attendees tomove freely between IAAI and National Con-ference sessions. IAAI and the National Con-ference are jointly sponsoring several invitedtalks that fit the theme of both programs.

A key goal of IAAI-97 is to promote a dia-log between basic and applied AI. Basic AI re-search will benefit by learning about challengesof real-world domains and difficulties and suc-cesses in applying AI techniques to real busi-ness problems. AI applications developers willbenefit from learning about new AI techniquesthat will enable the next generation of applica-tions. IAAI-97 will address the full range of AItechniques including knowledge-based sys-tems, natural language, and vision.

IAAI-97 showcases 11 winners of the “In-novative Application Award.” These applica-tions are the most impressive AI applicationsof the past year. They are selected for theirdemonstrated business value and their techni-cal innovation. The papers are case studies thatprovide a valuable guide to designing, building,managing, and deploying systems incorporat-ing AI technologies. These applications pro-vide clear evidence of the impact and valuethat AI technology has in today’s world. Orga-nizations honored this year are leading busi-nesses and government agencies from the US,Europe, and Asia. They include Air Productsand Chemicals, Inc.; Brightware, Inc.;Carnegie Mellon University; Clarity SystemsPte Ltd, Singapore; Fannie Mae; HewlettPackard Company; Hyundai Engineering &Construction Company, Ltd, Korea; HyundaiInformation Technology Company, Ltd, Korea;IBM Corporation; Information Technology In-stitute, Singapore; ISoft, France; Korea Ad-vanced Institute of Science and Technology;New York State Education Department; Ox-ford Health Plans; The Parallaz Corporation;Siena College; SISCOG – Sistemas CognitivosLda., Portugal; and Union Pacific Railroad.

This year completes the changes institutedlast year to revitalize IAAI. A new track,Emerging Applications and Technologies, has

been added to “bridge the gap” between AI re-search and AI applications development. Pa-pers in this track describe efforts whose goal isthe engineering of AI applications, and whichinform AI researchers about the utility ofspecific AI techniques for and constraints im-posed by applications domains and/or AI appli-cations developers about tools, techniques, ormethods that will enable the next generationof new and more powerful applications.

IAAI-97 also includes a series of invitedtalks that address key areas in which we expectto see AI applied in the near future. Invitedtalks will address (1) technical background—what AI techniques are applicable to this area?(2) current breakthroughs—what has hap-pened recently to make applications feasible inthis area? (3) emerging applications—what aresome sample applications that are beginning toappear? (4) barriers—what are the remainingtechnical and institutional roadblocks to wide-spread use of AI in this area? (5) strategies forprogress—what can and should the AI commu-nity do to overcome these barriers? We inviteyou to contribute to the dialog between basicand applied AI by joining us for IAAI-97.

Ted E. Senator, IAAI–97 Program ChairBruce G. Buchanan, IAAI–97 Program Cochair

2 AAAI–97 & IAAI–97 Preview

AAAI-97 Program Cochairs:Benjamin J. Kuipers, University of Texas atAustin; Bonnie Webber, University of Pennsylvania

AAAI-97 Associate Chair: Ramesh Patil, University of Southern California / Information Sciences Institute

IAAI-97 Conference Chair: Ted E. Senator

IAAI-97 Conference Cochair: Bruce Buchanan

Hall of Champions Chair: Matthew L. Ginsberg,CIRL/University of Oregon

Robot Building Laboratory Chair: David Miller,KISS Institute for Practical Robotics

Robot Competition Chair: Ronald C. Arkin,Georgia Institute of Technology

Student Abstract and Poster Chair: Polly K. Pook,Massachusetts Institute of Technology

Tutorial Cochairs: Bart Selman, ATT Laboratories

Brian C. Williams, NASA Ames Research Center

Workshop Chair: Raymond C. Mooney, University of Texas at Austin

SIGART/AAAI-97 Doctoral Consortium Organizers: Vibhu O. Mittal, University of Pittsburgh & Loren G. Terveen, AT&T Research

Keynote Address: AI Growing Up: The Changes and Opportunities

James F. Allen, University of Rochester9:00 – 10:00 AM, Tuesday, July 29

If we draw an analogy between the development of AI and the stagesof human development, I would argue that AI research so far has beena varied and exploratory childhood-with false starts, great leaps offaith, no clearly defined goals, and even some antisocial behavior. Yet,like most people, we have somehow survived and have learned some ofthe basic facts of life. But we are now moving onto adolescence, wherewe must learn much of our self-discipline and acquire a set of life-longhabits. The reason that I think we are at this profound transition pointis that it is now possible to build simple intelligent artifacts, from sim-ple robots, to reasoning systems that analyze and predict phenomena,to simple natural language dialog systems. The presence of such arti-facts will enable us, in fact require us, to develop a new paradigm of re-search that combines theoretical work with a significant empiricalcomponent. I will draw from my own and other’s work in natural lan-guage, especially work aimed at defining conversationally proficient in-telligent agents, to illustrate why we are at such a critical point. I’llthen lay out some choices we have to make and explore what I believeare our excellent prospects for the future.

James Allen is the John Dessauer Professor of Com-puter Science at the University of Rochester, where hehas been on the faculty since 1979, after receiving hisPh.D in computer science from the University ofToronto. He was one of the first recipients (1984) ofthe Presidential Young Investigator award from NSFand is a Fellow of the American Association for Ar-

tificial Intelligence (AAAI). He was editor-in-chief of ComputationalLinguistics from 1983-1993 and is the author of Natural Language Un-derstanding, published by Benjamin Cummings in 1987, with a secondedition published in 1995, and Reasoning About Plans, published byMorgan Kaufmann in 1991. He was general chair of the Second Inter-national Conference on Principles of Knowledge Representation heldin 1991 and was chairman of the Computer Science Department atRochester from 1987-1990. From 1992-1996, he was Director of thecognitive science program at the University of Rochester. He is cur-rently codirector of the Center for the Sciences of Language at theUniversity of Rochester. He has received numerous grants for his re-search, from NSF, ARPA, ONR, AFOSR, Rome Labs, and other agen-cies.

His research is generally aimed at defining a computational model ofan intelligent, conversational agent, and many projects address the in-terface between knowledge representation and reasoning and languagecomprehension. He has published numerous research articles in the ar-eas of natural language understanding, knowledge representation, tem-poral reasoning, planning and plan recognition.

AAAI-97 Opening ReceptionThe AAAI-97 opening reception will be held Tuesday, July 29 from6:00 – 7:00 PM in the Rhode Island Convention Center. Attendees willhave an opportunity to view the exhibits and perhaps challenge one ofthe computer players of a variety of classic games of strategy in the Hallof Champions. The AAAI-97 Student Abstract Poster Session will beheld simultaneously, as will the poster session for the SIGART/AAAI-97 Doctoral Consortium. A variety of hors d’oeuvres, some served byrobots from the Sixth Annual Mobile Robot Competition, and a no-host bar will be available. Admittance to the reception is free toAAAI-97 registrants. A $20.00 per person fee will be charged forspouses, children, and other nontechnical conference registrants.

AAAI-97/IAAI-97 Invited PresentationsTuesday – Thursday, July 29-31

The Mobile Robot Competition: Martians, Remotes, Hors d’Ouevres, and Cleaning up theMess Afterwards

Ronald C. Arkin, Georgia Institute of Technology and R. James Firby, University of Chicago2:00 – 3:00 PM, Tuesday, July 29

This year’s competition, the sixth annual held atAAAI, continues to expand upon the legacy of thosewhich preceded it. In this talk we first review theevent’s history and goals. This year, however, marks asignificant departure from the past. We survey the fourdifferent events which make up this year’s competition(find life on Mars, find the remote control, home vac-uuming, and hors d’ouevres anyone?). Their signi-ficance to the AI and robotics communities lies alongseveral lines: addressing opportunities in the explo-ration of Mars that are inspired by NASA’s recentlaunch of the Pathfinder mission and its Sojourner ro-bot, coupled with the teasing scientific possibility oflife on that planet; developing assistive robotic tech-

nology for the disabled; mainstreaming service robot applications; andheightening human-robot interaction by having robots serve food tothe AAAI conference attendees at this year’s conference reception.This talk aspires to provide the research and intellectual backdrop thathighlights the various aspects of AI, robotics, and computer visionwhich are challenged by each event, and to describe which specific re-search approaches are being used by the various competitors, that youmay have only read about but not actually seen in action.

Keynote Address / Invited Presentations

Complete and up-to-date information on the AAAI–97 conference can befound on the world wide web at www.aaai.org/Conferences/National/1997/

Embodied Intelligent Agents:Issues and Trends in Robotics

George A. Bekey, University of Southern California2:00 – 3:00 PM, Wednesday, July 30

We survey some of the changes in robotics during thepast 20 years, from simple industrial manipulators to au-tonomous intelligent agents, with an emphasis on therole of AI. Included are developments in robot control,learning, locomotion and group behavior, and currenttrends such as miniaturization, humanoids, robotcolonies and human-robot interaction.

The Emergence of Spacecraft Autonomy(Joint Conferences Talk)

Richard J. Doyle, Jet Propulsion Laboratory9:00 – 10:00 AM, Wednesday, July 30

The challenge of space flight in NASA’s future is to en-able more frequent and more intensive space explo-ration missions at lower cost. Nowhere is this challengemore acute than among the planetary exploration mis-sions which JPL conducts for NASA. The launching ofa new era of solar system exploration—beyond recon-naissance—is being designed for the first time around

the concept of sustained intelligent presence on the space platformsthemselves. Artificial intelligence, spacecraft engineering, mission de-sign, software engineering and systems engineering all have a role to playin this vision, and all are being integrated in new work on spacecraft au-tonomy.

What Does KR Have to Say to AI?David Etherington, CIRL/University of Oregon11:40 AM – 12:40 PM, Tuesday, July 29

In recent years, Knowledge Representation (KR) hasbecome more and more of a discipline unto itself, fo-cusing on artificial problems while other areas of AIhave tended to develop their own representations andalgorithms. This talk will consider what traditional KRhas to offer to AI.

Computer Game Players: What They Mean for Society and for Science (Panel)

Organizer: Matthew L. Ginsberg, CIRL/University of Oregon11:40 AM – 12:40 PM, Wednesday, July 30

Computer game players have come of age, exhibitingworld class performance in many — if not yet quite all— games of strategy. This panel discussion will focus onthe lessons to be drawn from these successes: How canthe rest of AI duplicate the successes of the game play-ers? Is the fact that humans are being displaced as theworld’s best game players an anomaly, or a harbinger of

things to come? The panelists will be drawn from the participants in theHall of Champions, and will include both program authors and humanchampions.

Advances in Uncertain ReasoningEric Horvitz, Microsoft Research4:30 – 6:00 PM, Wednesday, July 30

For thirteen years, the Conference on Uncertainty andArtificial Intelligence (UAI) has been a central meetingfor researchers from computer science, decision science,operations research, statistics, and psychology with in-terest in developing computational methods for grap-pling effectively with inescapable uncertainties in thereal world. I will discuss recent advances in uncertain

reasoning, highlighting key developments in representation, inference,and learning.

Machine Learning for Intelligent SystemsPat Langley, Daimler-Benz Research and Technology Center and Institute for the Study of Learning and Expertise3:10 – 4:10 PM, Tuesday, July 29

Recent research in machine learning has focused on su-pervised induction for simple classification or predictionand, in the process, has become disconnected from AI’soriginal goal of creating complete intelligent agents. Inthis talk, Langley reviews recent work on machinelearning for planning, natural language, and related top-ics that runs counter to this trend and thus holds inter-

est for the AI research community at large.

Prospects, Trends and Issues In Government Sup-port of AI: Views of the Funding Agencies (Panel)

Organizer: Mel Montemerlo, NASA2:10 – 3:10 PM, Thursday, July 31

Resource-Bounded Language ProcessingFernando Pereira, AT&T Bell Labs10:30 – 11:30 AM, Wednesday, July 30

Much of natural-language processing research in thepast thirty years assumed a ready supply of general andlinguistic knowledge, and limitless computational re-sources to use it in understanding and producing lan-guage. However, in practice accurate knowledge is hardto acquire and computational power is limited. Trying tokeep within that resource budget, approximate, often

statistical, knowledge sources and less profligate, often finite-state, pro-cessing models have been developed and applied with remarkable suc-cess to problems such as speech recognition, parsing and translation. Fur-thermore, the new approaches have close connections with recentdevelopments elsewhere in AI.

The Ascent of SoarPaul S. Rosenbloom, USC/Information Sciences Institute; John Laird,University of Michigan; and Jill Lehman, Carnegie Mellon University10:30 – 11:30 AM, Tuesday, July 29

For the past fifteen years we have beenevolving Soar from its origins as a prob-lem solving agent towards the grand chal-lenge of a human-like intelligent agent.Here we reflect on this past history, de-

scribe Soar’s present form, and speculate on the future path towards hu-man-like intelligent agents.

4 AAAI–97 / IAAI–97 Invited Talks & Panels

Preaching What We Practice: How AI is Changing the Concept of Computation

Lynn Andrea Stein, Massachusetts Institute of Technology1:00 – 2:00 PM, Thursday, July 31

AI is transforming computer science, replacing notionsof computation as calculation with computation as in-teraction, shifting focus from algorithmic I/O to sus-tained behavior patterns. Yet our pedagogy remains vir-tually unchanged. Stein will describe a radical in-troduction to computer science that teaches studentsthis model from the outset.

James Bond and Mike Ovitz: The Secret Life of Agents

Katia P. Sycara, Carnegie Mellon University9:00 – 10:00 AM, Thursday, July 31

As agents populate Cyberspace in their many guises androles, they coordinate and interact in different ways,spanning self-interested, as well as collaborative inter-actions. Agent coordination should be supported by anagent’s internal architecture and agent societal frame-works. We take a microeconomic view of coordination.In this talk we report on our work on adaptive agent ar-

chitecture, agent organizations, and presence of middle agents.

Performance Models for Dialogue AgentsMarilyn Walker, ATT Laboratories3:10 – 4:10 PM, Wednesday, July 30

Recent advances in dialogue modeling and spoken lan-guage processing have made it possible to build spokendialogue agents for many tasks. However, one obstacleto progress in spoken dialogue is the lack of a generalperformance model for comparing agent strategies. Thistalk discusses recent empirical studies whose goal is todevelop and test such a performance model.

Market-Oriented ProgrammingMike Wellman, University of Michigan10:30 – 11:30 AM, Thursday, July 31

Market-oriented programming is the construction ofcomputational economies, where agents interactthrough a price system. Markets can provide effectiveallocation of resources for a variety of distributed envi-ronments, and economic analysis a powerful design toolfor interaction mechanisms. The spread of electroniccommerce puts a premium on market-aware agents, and

presents a case for market awareness on the part of agent developers andAI researchers as well.

Recent Advances in KDD (Joint Conferences Talk)4:30 – 6:00 PM, Tuesday, July 29

AAAI-97/IAAI-97 Joint ExhibitionTuesday – Thursday, July 29-31

The exhibit program will offer exhibits and demonstrations by leadingsuppliers of AI software as well as AI consultants and publishers display-ing the latest in AI books and periodicals. Recent AAAI Exhibitorshave included:• AAAI Press • Academia Book Exhibits• Academic Press • Angoss Software International, Ltd.• Elsevier Science Publishers• Franz, Inc. • Harlequin Ltd• ILOG, Inc. • Intelligent Automation, Inc.• Kluwer Academic Publishers• Lawrence Erlbaum Associates• The MIT Press • Morgan Kaufmann Publishers, Inc.• OXCO, Inc. • PCAI• Prentice Hall • PWS Publishers/ITP• Real World Interface, Inc. • Springer Verlag New York, Inc.• Talarian Corporation • Wizsoft

AAAI-97 Hall of ChampionsTuesday – Thursday, July 29-31

AAAI-97 will include demonstrations of the best computer players of avariety of classic games of strategy. The games selected are expected toinclude the following:

GAME PROGRAM AUTHOR

backgammon TD-Gammon Gerry Tesaurobridge GIB Matt Ginsbergcheckers Chinook Jonathan Schaefferchess Rebel Chess Ed Schroedergo Handtalk Sidney YuanOthello Logistello Michael BuroScrabble Maven Brian Sheppard

(Rebel Chess was selected over Deep Blue because the programs are ofroughly comparable strength and Rebel Chess runs on a microprocessor.)Several classic solved games will also be included, such as go-moku (tic-tac-toe on a large board where the object is to get five marks in a row),connect-4, and nine man’s morris.

AAAI-97 attendees will be able to interact with these programs in avariety of ways. first, all of the programs themselves will be available dur-ing the conference and attendees will be able to compete against them.Second, many of the programs’ authors will be available to discuss boththe technical issues involved in creating the programs and the social is-sues involved in introducing world-class computer players into tourna-ment play. And finally, human experts will be on hand to play a series ofchallenge matches against the programs themselves. The Hall of Cham-pions will include a spectators’ area where AAAI attendees can viewthese matches as they progress. Admittance to the Hall of Champions isincluded in the technical program registration fee or the onsite exhibits-only registration fee. High-school students are welcome and will be ad-mitted without fee upon presentation of a valid high-school student ID.Children under 12 must be accompanied by an adult conference regis-trant. The Hall of Champions is being organized by Matthew L. Gins-berg, ginsberg@cirl.uoregon.edu.

Joint Exhibition & Hall of Champions 5

1997 AAAI-97 Mobile Robot Competition and Exhibition Tuesday – Thursday, July 29-31

This year, four events and an exhibition will be held at the Sixth An-nual Mobile Robot Competition.

Event 1: Find Life on MarsThis task is inspired by the upcoming Pathfinder Mission to Mars, as wellas the tantalizing (albeit limited) possibility of life on Mars as depictedby a recent meteorite analysis. The robot’s mission is to explore a largearea of Mars, looking for signs of past and current life, and return the life-forms, and only those life-forms, to the lander for further analysis. Fromsatellite imagery, we have some clues as to where life may exist, but it isup to the robot to make a thorough exploration before its batteries rundead.

Event 2: Find the Remote ControlThis event is inspired by the need for robot assistants to perform ‘fetch-it’ tasks in partially known environments. Imagine a robot assistant help-ing a handicapped person around the home. The person might ask therobot to fetch an orange, the TV remote, a cup of coffee, and so on.While the robot may not know where all of these items are initially, overtime it will learn roughly where they are kept. The event will take placein an arena that contains tables, chairs, and shelves at varying heights.Scattered throughout the area, on the floor, the shelves, and the tables,will be 12 different objects. The robot will start the event near a humansitting in a chair (i.e., judge) who will ask the robot to fetch three items.Once these items have been returned, the human will ask for three moreitems. The winner will be the robot that can find and return the mostitems in the allotted time.

Event 3: Home VacuumThe point of this contest is to explore the usefulness of intelligence in atask that appears to only require essentially unskilled labor—simplehousehold vacuuming. We believe that unlike vacuuming in the serviceindustry (factories, warehouses, etc.) home vacuuming will require sen-sate intelligence to deal with the humans in an everyday environment.For the AAAI contest, vacuuming robots ought to be short on vacuummechanisms and long on intelligence. That is to say, simple suction, stor-age, and disposal devices are sufficient for these tasks, but the robots willprobably have to make reasoned trade-offs among subtasks in real timeto achieve a high score.

Event 4: (Special event) Hors d’Oeuvres Anyone?This event will occur at the AAAI main reception where there will beheavy interaction with the attendees. Judging will be conducted by theattendees. The goal is to provide solid refreshments to people in closequarters. Safety and self-protection are paramount. A human escort (on-ly one allowed per team within the area) will always be nearby for safe-ty and control of the robot (i.e., if it moves out of the designated area),but will be limited in their interaction with the attendees. The robotsmust be fully autonomous. The escort will also replenish the hors d’oeu-vres on an as needed basis.

ExhibitionThe intent of the Robot Exhibition is to showcase current research inrobotics that does not fit into the competition tasks. We welcome robotdemos (scheduled or continuous), robot soccer entries, posters with orwithout robots present, and videos.

More information, points-of-contact, and the official rules can all befound at the following web sites: http://spbtrc.gtri.gatech.edu/AAAI97/(Competition) and http://www.ai.mit.edu/people/holly/AAAI97/ (Exhi-bition.)

AAAI-97 Robot Building Laboratory (RBL)Sunday – Monday, July 27-28, 1997

AI meets the road! Participants will spend the day learning about howAI can (and can’t) be integrated into the world of mobile robots. Mostof the day will be hands-on: building and programming small mobile ro-bots to do a variety of tasks.

Much of the current AI research deals with the actions of embeddedagents. In this course it will become apparent that simulations of anagent’s environment are often inadequate for effective evaluation of sys-tems. The RBL will give the attendees the necessary information to startembedding their systems in physical agents mobile robots that can in-teract with realistic environments.

Material to be covered:• Realistic versus idealized robots• Major components of robot systems• Sensor and effector integration• A crash course in behavior control programming• Everything an AI researcher needs to know about PID• Vendors and suppliers for getting robots into your lab or home.

Functional mechanical modules will be available from the start of theprogram. Participants will be able to spend their time designing and pro-gramming the robot, with only a bare minimum of LEGO®-hacking toget their robots to move reliably. The lab will begin with a brief tutorialon sensors, effectors and robot capabilities to get everyone up to speed,followed by the actual robot building. Throughout the day there will bea series of short tutorials, both for individual teams and for the group asa whole, on particular aspects of robot building and programming. OnMonday, July 28, all the robots will be displayed in the arena to show offtheir special capabilities and to compete head to head in a contest ofspeed and intelligence. This exhibition will be open to all of the confer-ence attendees. Please see the registration form on page 21 for the feeschedule.

The goals of this lab event are to give all participants exposure to theintricacies of melding AI and robotics; show the value of performing AIexperiments on physical devices; familiarize the participants with thecurrent robotic experimental technology; and give everyone a chance toplay with AI that they can get their hands around.

This lab is aimed at AI researchers and practitioners who want tomove their systems out of simulations and into the physical world. A ba-sic understanding of common AI techniques and programming lan-guages will be assumed.

The lab is being organized and taught by the KISS Institute for Prac-tical Robotics (KIPR) for AAAI. Instructors and assistants will comefrom KIPR’s trained staff. David Miller will be the lead instructor. Forupdated information about this event, please refer to www.kipr.org/rbl97.

6 Robot Program

AAAI-97 / IAAI-97 Paper Sessions

Monday, July 28(IAAI-97 Award Winning Deployed Application Papers Only)

9:00 – 10:00 AM

Session 1: IAAI-97: Introduction/Scheduling I

10:30 – 11:30 AM

Session 2: IAAI-97: Scheduling II

11:40 AM – 12:40 PM

Session 3: IAAI-97: Planning/Layout

2:00 – 3:00 PM

Session 4: IAAI-97: Regulatory Compliance/Eligibility Determination I

3:10 – 4:10 PM

Session 5: IAAI-97: Regulatory Compliance/Eligibility Determination II

4:30 – 5:30 PM

Session 6: IAAI-97: Computer Diagnosis

Tuesday, July 2910:30 – 11:30 AM

Session 7: Planning Under UncertaintySession 8: Qualitative Reasoning 1Session 9: Description LogicsSession 10: Constraint Satisfaction Problems: SymmetrySession 11: IAAI-97: Invited Talk: Alexa Macray

11:40 AM – 12:40 PM

Session 12: ClassificationSession 13: Automated Reasoning/DiagnosisSession 14: Multi-Agent SystemsSession 15: IAAI-97: Information Extraction ISession 16: IAAI-97: Complex Systems Design I

2:00 – 3:00 PM

Session 17: Natural Language GenerationSession 18: Constraint Satisfaction Problems and Bayes NetworksSession 19: Text Retrieval and LearningSession 20: IAAI-97: Military ISession 21: IAAI-97: Transportation Diagnosis

3:10 – 4:10 PM

Session 22: Reactive BehaviorSession 23: Agent ArchitectureSession 24: Knowledge Representation: OntologiesSession 25: IAAI-97: Military IISession 26: IAAI-97: Design II

4:30 – 6:00 PM

Session 27: Spatial UncertaintySession 28: Plan GenerationSession 29: Model Selection and OverfittingSession 30: Structure of Constraint Satisfaction Problems

6:00 – 7:00 PM

AAAI-97 Student Abstract Poster Session and SIGART/AAAI Doctoral Consortium Poster Session (in conjunction with Opening Reception)

Wednesday, July 309:00 – 10:00 AM

Session 31: Problem Solving & Computational Resources

Session 32: Belief and DecisionSession 33: Information RetrievalSession 34: Computational Systems for Education

10:30 – 11:30 AM

Session 35: Parallelism in LearningSession 36: DiagnosisSession 37: IAAI-97: SpaceSession 38: Flexible Hierarchical PlanningSession 39: IAAI-97: Multimedia

11:40 AM – 12:40 PM

Session 40: Knowledge Discovery in DatabasesSession 41: Knowledge Representation: Nonmonotonic LogicSession 42: Agent CoordinationSession 43: Heuristics for SchedulingSession 44: IAAI-97: Space/Knowledge Management

2:00 – 3:00 PM

Session 45: Knowledge Representation: Reasoning about Action ISession 46: Local Search: Beyond SATSession 47: Automated Reasoning & the User InterfaceSession 48: Modeling for Decision ProcessesSession 49: Information Extraction II

3:10 – 4:10 PM

Session 50: NegotiationSession 51: Knowledge Representation: Expert SystemsSession 52: Probability and PlanningSession 53: Search (Cost)Session 54: IAAI-97: Invited Talk

4:30 – 6:00 PM

Session 55: Navigation & PerceptionSession 56: Constraint Satisfaction TechniquesSession 57: Language and LearningSession 58: Formal Analyses of LearningSession 59: IAAI-97: Design III

Thursday, July 319:00 – 10:00 AM

Session 60: Machine Learning (Probabilistic)Session 61: Optimal PlanningSession 62: Knowledge Representation: Theorem ProvingSession 63: Efficient Reasoning

10:30 – 11:30 AM

Session 64: SchedulingSession 65: Reasoning about Physical SystemsSession 66: Building and Modifying Knowledge BasesSession 67: Natural Language

1:00 – 2:00 PM

Session 68: Knowledge Representation for Automated ReasoningSession 69: Learning In Linguistic DomainsSession 70: Case-Based Reasoning and PlanningSession 71: Local Search Techniques

2:10 – 3:10 PM

Session 72: Reasoning about Action IISession 73: Experimental MethodologySession 74: Techniques for Temporal Reasoning

A complete AAAI-97/IAAI-97 program schedule can be found athttp://www.aaai.org/Conferences/National/1997/

Paper Session Topics & Schedule 7

1997 AAAI Tutorial ForumThe 1997 AAAI Tutorial Forum features sixteen four-hour tutorials thatexplore hot topics within and outside the AI field. Each tutorial istaught by experienced scientists and practitioners in AI. We encourageall attendees to take advantage of this opportunity to learn about ad-vances in areas outside their personal focus. One low fee will entitle tu-torial forum registrants to attend up to four consecutive tutorials, in-cluding four tutorial syllabi.

SESSION I: Sunday, July 27, 9:00 AM – 1:00 PM

SA1: Belief Networks and Decision-Theoretic Reasoning for Artificial IntelligenceDaphne Koller and Jack Breese

SA2: Evolutionary Computation and Artificial LifeMelanie Mitchell and John Batali

SA3: Agent Development in SoarJohn Laird, Clare Congdon, and Randolph Jones(Please note: This is a full-day tutorial.)

SA4: Data MiningUsama Fayyad and Evangelos Simoudis

SESSION II: Sunday, July 27, 2:00 – 6:00 PM

SP1: Reinforcement LearningLeslie Pack Kaelbling and Richard S. Sutton

SP2: Model-Based Autonomous SystemsBrian Williams and Pandurang Nayak

SP3: Modeling with Defaults: Causal and Temporal ReasoningHector Geffner

SP4: Principles of Ontological EngineeringMichael Gruninger and Mike Uschold

SESSION III: Monday, July 28, 9:00 AM – 1:00 PM

MA1: Topics in the Theory of the Practice of Machine LearningMichael Kearns

MA2: Genetic ProgrammingAstro Teller and David Andre

MA3: The Database Perspective on Knowledge Representation and Information IntegrationAlon Levy and Jeffrey D. Ullman

MA4: Physics-Based Modeling for Vision and Virtual Human AnimationDimitris Metaxas and Norman Badler

SESSION IV: Monday, July 28, 2:00 – 6:00 PM

MP1: Compute-Intensive Methods in Artificial IntelligenceHenry Kautz and Bart Selman

MP2: Computer VisionDaniel Huttenlocher and Todd Cass

MP3: Practical PlanningSteve Chien and Brian Drabble

MP4: Mobile Robot Control ArchitecturesR. James Firby and Reid G. Simmons

Agent Development in Soar (SA3)John Laird, Clare Congdon and Randolph Jones9:00 AM – 6:00 PM, Sunday, July 27

The major thrust of this tutorial will be to give the participants an un-derstanding of the details of Soar so that they can create simple Soarprograms. This will be a full-day tutorial, with an emphasis in the morn-ing on understanding the syntax and structure of the architecture (thememories and processes), and an emphasis in the afternoon on agent de-velopment. In the morning, participants will run, modify, and debugsmall demonstration programs that illustrate the various parts of Soar’sstructure. In the afternoon, we will work on simple agents that interactwith a dynamic simulated environment. The students will build theirown complete agents that navigate and compete in a simple maze world.The class will culminate with a competition among the agents designedby students.

We expect participants to have general programming experience anda basic understanding of symbolic processing, match, AI, problem solv-ing, etc. No prior knowledge of Soar or rule-based systems required.

John Laird received his Ph.D. in computer science fromCarnegie Mellon University in 1983. Since 1986 he hasbeen on the faculty of the University of Michigan inthe Electrical Engineering and Computer Science De-partment where he is currently an associate professorand director of the Artificial Intelligence Laboratory.His research interests center on the development of in-

tegrated intelligent agents and their underlying architecture. Laird isone of the original developers of Soar and is actively involved in its con-tinual evolution and application.

Clare Congdon is a visiting assistant professor of com-puter science at Bowdoin College. She earned herPh.D. in computer science at the University of Michi-gan in 1995. Her research interests include machinelearning and robotics, and she is currently focused onteaching issues, especially the role of computer sciencein liberal arts schools.Randolph Jones holds a Ph.D. in information and com-puter science from the University of California, Irvine.He has served on the research faculty at Carnegie Mel-lon University, the University of Pittsburgh, and cur-rently at the University of Michigan. His research in-terests include intelligent agents, machine learning,and psychological modeling.

Belief Networks and Decision-Theoretic Reasoning for AI (SA1)

Daphne Koller and Jack Breese9:00 AM – 1:00 PM, Sunday, July 27

Uncertainty is an inherent feature in real life. Therefore, reasoning anddecision making under uncertainty are central to intelligent behavior.Over the past decade, a new and successful technology has emerged forautomating these capabilities. In this approach, an agent’s uncertaintyand objectives are explicitly modeled using probabilities and utilities;principled decision-theoretic techniques are then used to determine theagent’s beliefs given its limited observations, and to decide on a rationalcourse of action (one that maximizes expected utility).

In this tutorial, we describe some of the fundamental representationsand algorithms in this area. We present Bayesian belief networks, whichcombine qualitative information about causal structure with probabilis-

8 Tutorial Forum

tic information to provide a compact naturalencoding of complex probabilistic models. Weoutline several key algorithms for exact andapproximate probabilistic inference in beliefnetworks. We describe how this technologycan be scaled up to more complex domains byexploiting additional types of structure (e.g.,hierarchical decomposition), by automatedconstruction of models from knowledge bases,and by learning models directly from data. Fi-nally, we illustrate the use of belief networks aspart of a broader architecture for decision-the-oretic planning.

Participants should be familiar with basicprobability theory. No prior knowledge of be-lief networks is required.

Daphne Koller received herPh.D. from Stanford Universi-ty in 1993. After two years atthe University of California,Berkeley, she returned to Stan-ford as an assistant professor.She has done extensive work

on representation, inference, and learning inprobabilistic models. She is the recipient of theArthur Samuel Award for best thesis in theComputer Science Department and of theSloan Foundation Faculty Fellowship.

Jack Breese is a senior re-searcher in the Decision The-ory and Adaptive SystemsGroup in Microsoft Research.He has published over the pastten years in the areas of uncer-tainty in artificial intelligence

and computational decision analysis. He re-ceived his Ph.D. in 1987 from the Departmentof Engineering Economic Systems at StanfordUniversity and joined Microsoft Research in1993.

Compute-Intensive Methods in Artificial Intelligence (MP1)

Henry Kautz and Bart Selman2:00 – 6:00 PM, Monday, July 28

In the 1970s and 1980s the success of knowl-edge-intensive approaches to problem solvingeclipsed earlier work on compute-intensiveweak methods. However, in recent years com-pute-intensive methods have made a surprisingcomeback. One of the most prominent exam-ples is the success of IBM’s Deep Blue in defeat-ing Gary Kasparov in the first game of the 1996ACM Challenge match. The game led Kas-parov to exclaim, “I could feel — I could smell— a new kind of intelligence across the table.”Yet Deep Blue derives its strength mainly fromhighly optimized search.

Another dramatic development in the com-pute-intensive approach was the recent com-puter proof resolving the Robbins problem(McCune, Oct. 1996). The Robbins problem isa well-known problem in Boolean algebra, andwas open for over sixty years. The computerproof was found by applying powerful searchtechniques guided by general search tactics.Unlike previous results in the area of automat-ed theorem proving, this time several aspects ofthe computer proof could be called “creative”by mathematicians’ standards.

Deep Blue’s performance and the resolutionof the Robbins problem are good examples ofthe dramatic improvement in performance ofcompute-intensive approaches compared tojust a few years ago. Other examples of the re-cent success of such methods are in planning,qualitative reasoning about physical systems,and learning.

This tutorial will explore how and why quan-titative changes in compute-intensive methodscan lead to dramatic qualitative changes in over-all performance. These technique promise to al-leviate the knowledge acquisition bottleneck inAI by allowing expert-level performance to beachieved with only a a minimal amount of do-main knowledge. We will discuss how theselessons can be broadly applied to problems in ar-tificial intelligence, and we will speculate onwhere the next breakthroughs might occur.

Henry Kautz is head of theArtificial Intelligence Princi-ples Research Department ofAT&T Labs. His research in-terests include knowledge rep-resentation, intelligent soft-ware agents, and efficient

reasoning systems. He received his Ph.D. fromthe University of Rochester in 1987, andjoined AT&T Bell Labs that year. In 1989 hereceived the Computers and Thought Awardfrom the International Joint Conference onArtificial Intelligence. His work on stochasticsearch for planning won the best paper awardat the 1996 conference of the American Asso-ciation for Artificial Intelligence.

Bart Selman is a member ofthe Artificial IntelligencePrinciples Research Depart-ment at AT&T Laboratories.He holds a Ph.D. and M.Sc. incomputer science from theUniversity of Toronto, and a

M.Sc. in physics from Delft University of Tech-nology. His research has covered many areas inartificial intelligence, including tractable infer-ence, knowledge representation, search, plan-ning, default reasoning, constraint satisfaction,and natural language understanding. He has re-

ceived best paper awards at both the Americanand Canadian national artificial intelligenceconferences, and at the International Confer-ence on Knowledge Representation. His cur-rent research projects are on efficient reasoning,stochastic search methods, planning, knowl-edge compilation, and software agents.

Computer Vision (MP2)Daniel Huttenlocher and Todd Cass2:00 – 6:00 PM, Monday, July 28

The goal of computer vision is to extract con-tent from images by understanding the proper-ties of the physical world that gives rise tothose images. While image processing is con-cerned primarily with the manipulation andtransformation of image signals, computer vi-sion goes beyond this by studying properties ofthe world and of the cameras which produceimage signals.

This tutorial will provide a broad introduc-tion to the many techniques and applicationsin computer vision. We will consider how thegeometry of the world and the physics of theimaging process encode the structure of scenesin images, and how computer vision tech-niques uncover and utilize this structure in in-terpreting images. We will also cover applica-tions of computer vision to problems in areassuch as image databases, analyzing video, androbot navigation.

Examples of topics include low-level extrac-tion of image properties such as detecting in-tensity edges, computing visual motion, andanalyzing color information, middle-level pro-cessing which recovers three-dimensional in-formation using two or more viewpoints inspace or time, such as stereopsis and structurefrom motion, and higher-level processing suchas recognition methods which identify knownobjects in an image for use in visual inspection,image database searches, and human facerecognition.

This is an introductory tutorial, requiring noprior knowledge of computer vision, and rely-ing only on college-level material in computerscience and mathematics.

Todd Cass received his Ph.D.in electrical engineering in1992 from the MassachussettsInsitute of Technology wherehe was a member of the MITArtificial Intelligence Labora-tory. Since 1992 he has been a

member of the research staff at the Xerox PaloAlto Research Center (Xerox PARC). His cur-rent interests include model-based objectrecognition, digital watermarking and embed-ded data.

Tutorial Forum 9

Daniel Huttenlocher is an as-sociate professor of computerscience at Cornell University,and a principal scientist in theImage Understanding Area atXerox PARC. His research in-terests are in computer vision,

digital libraries and computational geometry.He received his bachelors degree from Univer-sity of Michigan in 1980 and Ph.D. from MITin 1988.

The Database Perspective onKnowledge Representation andInformation Integration (MA3)

Alon Levy and Jeffrey D. Ullman9:00 AM – 1:00 PM, Monday, July 28

The fields of deductive database systems andknowledge representation have both developedconcepts and techniques for modeling and ma-nipulating data. Even though the two fieldsconsidered very closely related problems, thetechniques developed emphasized the specificneeds of each field. While database systemsdeal with large amounts of relatively simple da-ta, knowledge representation systems tend tomanipulate smaller amounts of more complexdata. The tutorial will survey several bodies ofwork in the database community that are close-ly related to knowledge representation prob-lems. Special emphasis will be put on illustrat-ing the relevance of the results to the artificialintelligence community. We begin by coveringseveral fundamental topics such as the tradeoffsin designing query languages for database sys-tems and different treatments of negation. Wedescribe the problems of query containmentand rewriting queries using views, that areclosely related to the problem of performing in-ference in knowledge representation systemsbased on first-order logic. We illustrate the useof these techniques in the problem of informa-tion integration, and compare the different ap-proaches taken to this problem in the two fiel-ds. Finally, we describe several recent researchdirections in database systems, such as themanagement of semistructured data and activedatabase systems.

Alon Levy is a principal mem-ber of technical staff at AT&TLabs. He received his Ph.D. incomputer science from Stan-ford University in 1993 andhis BSc in mathematics andcomputer science from He-

brew University, Jerusalem, in 1988. In hiswork he has made contributions to the fields ofartificial intelligence and database systems.

The focus of his work in the past few years hasbeen on the problem of information integra-tion, leading the Information Manifold projectat AT&T Labs. His other research interests in-clude description logics, knowledge baseverification, abstraction and reformulation ofcomputational theories, query optimization indatabases, materialized views, web-site man-agement systems, and management of semi-structured data.

Jeffrey D. Ullman (Ph.D.,Princeton University, 1966),is a professor at Stanford Uni-versity, and a leading re-searcher in database systems,knowledge-base systems, theo-retical computer science,

analysis of algorithms, and programming lan-guages and compilers. He served as the com-puter science department chair at StanfordUniversity, from 1990 to 1994 and has servedon the editorial boards of many journals, in-cluding the SIAM journal Computing,ACM,Computer and System Sciences, and Logic Pro-gramming. He is working on the efficient imple-mentation of data cubes for data warehousingand data mining.

Data Mining (SA4)Usama Fayyad and Evangelos Simoudis9:00 AM – 1:00 PM, Sunday, July 27

Knowledge discovery in databases (KDD) is arapidly growing AI field that combines tech-niques from machine learning, pattern recogni-tion, statistics, databases, and visualization toautomatically extract knowledge (or informa-tion) from lower level data (databases). Thisknowledge is subsequently used to support hu-man decision-making, e.g., prediction and clas-sification tasks, summarize the contents ofdatabases, or explain observed phenomena.The use of KDD systems alleviates the problemof manually analyzing the large amounts of col-lected data which decision-makers face cur-rently. Successful KDD systems have been im-plemented and are currently in use in finance,fraud detection, market data analysis, astrono-my, diagnosis, manufacturing, and biology. Thistutorial presents a comprehensive picture ofcurrent research paradigms in the field of KDD.The tutorial will provide an introduction toKDD, define the basic terms and the relationbetween data mining and the KDD process,present methods for data preparation and pre-processing, describe major data mining tech-niques from the fields of AI, pattern recogni-tion, databases, and visualization, discuss majorKDD systems from academia and industry andprovide a guide for developing a KDD system.

In the process, the tutorial addresses such issuesas role of various steps in the KDD process suchas sampling, selection, projection and dimen-sionality reduction, extraction of patterns andmodels, and the use of extracted knowledge.There are no prerequisites for this tutorial oth-er than familiarity with basic concepts in AI.

Evangelos Simoudis is vicepresident, Global Business In-telligence Solutions—IBMNorth America, where he isresponsible for the develop-ment and deployment of datamining and decision support

solutions to IBM’s customers worldwide. Priorto joining IBM, Evangelos worked at LockheedCorporation where he led the company’s datamining research, and was responsible for thedesign and commercial introduction of the Re-con data mining system, as well as its applica-tion to the financial and retail markets.Simoudis received a BA in physics from Grin-nell College, a BS in electrical engineeringfrom California Institute of Technology, anM.S. in computer science from the Universityof Oregon, and a Ph.D. in computer sciencefrom Brandeis University. Before Lockheed,Simoudis worked as a principal research staff atDigital Equipment Corporation’s Artificial In-telligence Center where he conducted researchon machine learning and pattern recognition,knowledge-based systems, and distributed ar-tificial intelligence. His research work at DEChas been incorporated in products for engi-neering design and diagnostic tasks. Simoudishas written extensively on data mining andmachine learning, and is editor in chief of theArtificial Intelligence Review.

Usama Fayyad is a senior re-searcher at Microsoft Re-search. Prior to joining Mi-crosoft, he headed theMachine Learning Group atJPL and was principal investi-gator on several tasks in tar-

geting automated data analysis in science andNASA data sets. He received a NASA medal(1994) and the 1993 Lew Allen Award for Ex-cellence at JPL. He cochaired KDD-94 andKDD-95 and was general chair of KDD-96. Heis a coeditor of Advances in Knowledge Discoveryand Data Mining (AAAI Press / The MIT PRess1996), and editor-in-chief of the new journalon this topic (Kluwer).

10 Tutorial Forum

Evolutionary Computation andArtificial Life (SA2)

Melanie Mitchell and John Batali9:00 AM – 1:00 PM, Sunday, July 27

Biologically inspired computation and com-putational models of complex adaptive sys-tems are new approaches of fundamental im-portance for the artificial intelligencecommunity. This tutorial will provide a se-lected survey of recent research in these ar-eas, especially those relevant to machinelearning and cognitive science.

Topics will include the theory of geneticalgorithms and their applications to prob-lems such as automatic programming, decen-tralized parallel computation, and evolvingneural networks; new results in the theory ofevolutionary computation; behavioral ecolo-gy and the mathematical theory of evolution;and artificial life models of the evolutionarydynamics of learning, altruism, communica-tion and social behavior. No previous knowl-edge of these areas is required.

Melanie Mitchell received aPh.D. in computer sciencefrom the University ofMichigan in 1990. Since1992 she has been on the re-search faculty of the Santa FeInstitute in Santa Fe, New

Mexico, and directs the Institute’s program inadaptive computation. She is also on the re-search faculty of the Computer Science De-partment of the Unversity of New Mexico inAlbuquerque. She is the author of An Intro-duction to Genetic Algorithms (MIT Press, 1996)and has published two other books and nu-merous research papers in the fields of artificialintelligence, cognititve science, and complexsystems, with particular focus on evolutionarycomputation and adaptive systems.

John Batali received aPh.D. in electrical engineer-ing and computer sciencefrom the MIT Artificial In-telligence Laboratory in1991. He is currently an as-sistant professor in the Cog-

nitive Science Department at the Universityof California at San Diego.

Genetic Programming (MA2)Astro Teller and David Andre9:00 AM – 1:00 PM, Monday, July 28

The tutorial will introduce the ideas and ap-plications of genetic programming (GP), acomputer search procedure inspired by thetheory of natural evolution. Starting with aprimordial ooze of randomly created pro-grams composed of functions, constants, andvariables appropriate to a problem, a popula-tion of computer programs is progressivelyevolved over many generations by applyingthe Darwinian principle of survival of thefittest, a recombination operation, and occa-sional mutation. Genetic programming hasfound applications in a wide variety of differ-ent areas of artificial intelligence includingpattern recognition, data mining, distributedAI, optimal control, molecular biology, sys-tem identification, and engineering design.

We will briefly review the properties andmechanics of genetic programming, andthen discuss the techniques that have beenemployed to successfully apply genetic pro-gramming to a variety of difficult real-worldproblems. Topics include multi-part pro-grams, automatically defined functions, iter-ation, recursion, memory structures, mentalmodels, architecture-altering operations, cel-lular encoding, neural programming, geneticdesign of electrical circuits, assembly codeevolution, evolvable hardware, promisingapplication areas for genetic programming, athorough presentation of the important cur-rent problems and issues for GP research, andthe implications of GP research to the fieldof artificial intelligence.

No advance knowledge about genetic al-gorithms, genetic programming, or biology isrequired.

Astro Teller is pursing re-search in artificial intelli-gence at Carnegie MellonUniversity. Teller is an ex-pert in the area of signalclassification in genetic pro-gramming. He is the author

of several book chapters, journal articles, andconference papers on the subject of noveltechniques in genetic programming.

David Andre is pursuingresearch on genetic pro-gramming and artificial in-telligence at UC Berkeley.He has published morethan 30 chapters and paperson genetic programming

and is working on an upcoming book. DavidAndre was copresenter of a tutorial on ge-netic programming at AAAI-96.

Mobile Robot Control Architectures (MP4)

R. James Firby and Reid G. Simmons2:00 – 6:00 PM, Monday, July 28

This tutorial will present the state-of-the artin architectural frameworks for controllingautonomous mobile robots. In recent years, aconsensus has developed around a three-lev-el architectural style — a bottom layer pro-viding behavioral (reactive) control, a toplayer providing planning (deliberation), anda middle (sequencing) layer bridging the gapbetween them.

The tutorial will begin by presenting fac-tors that influence the development of sucharchitectures: actuator and sensor control,complex changing environments, the desireto perform a wide variety of tasks, and theneed for execution monitoring and excep-tion handling. The tutorial will focus on casestudies of implemented systems that empha-size reactive and sequencing layers; planningfor mobile robots will be examined onlybriefly. The strengths and weaknesses of dif-ferent architectural approaches will be ana-lyzed in terms of the tasks and environmentsthey readily support, the role of perception intask achievement and execution monitoring,and how exceptions are propagated and han-dled. Finally, the tutorial will discuss the prosand cons of the three-layer consensus andexamine alternative avenues of research.

This is an intermediate-level tutorial in-tended for people who need to develop orevaluate mobile robotic systems. No previousexperience with robotics is required, but a fa-miliarity with control theory is helpful.

R. James Firby is an assis-tant professor of computerscience at the University ofChicago. His main interestis the construction of sys-tems that interact with peo-ple in complex, changing

environments. Firby is currently working onan architecture that integrates task planning,synchronization with external processes, andlow-level robot control. That architecturehas been implemented on an indoor mobilerobot. Firby received his Ph.D. in artificialintelligence from Yale University in 1989.His dissertation introduced the RAP systemfor reactive plan execution using situationspecific task methods. That research ad-dressed problems in error recovery, task-di-rected perception, and opportunism. From1989-1991 Firby worked as a research scien-tist at the Jet Propulsion Lab designing andbuilding local navigation software for thePathfinder Planetary Rover project.

Tutorial Forum 11

Reid G. Simmons is a seniorresearch scientist in theSchool of Computer Sci-ence at Carnegie MellonUniversity. Since coming toCMU in 1988, Simmons’sresearch has focused on de-

veloping self-reliant robots that can au-tonomously operate over extended periods oftime in unknown, unstructured environ-ments. This work involves issues of robot ar-chitectures that combine deliberative and re-active control, probabilistic robot planningand navigation, robust error detection andrecovery, and selective perception. The ideashave been used in a half-dozen robots, in-cluding indoor mobile manipulators andlegged and wheeled planetary rovers. Sim-mons has published over 50 articles and con-ference papers on his work in robotics androbot architectures. Simmons received hisPh.D. in artificial intelligence from MIT in1988. His dissertation focused on the combi-nation of associational and causal reasoningfor planning and interpretation tasks. Theresearch involved the use of multiple repre-sentations of the physical world, and devel-oped a domain-independent theory of debug-ging plans.

Model-Based AutonomousSystems (SP2)

Brian Williams and Pandurang Nayak2:00 – 6:00 PM, Sunday, July 27

A new generation of autonomous systems arebeing developed that have the potential forprofound social, environmental, and eco-nomic change. These include autonomousspace probes, chemical plant control systems,power grids, life support systems, and re-configurable traffic systems, to highlight buta few. To be economically viable these au-tonomous systems will need to be program-mable purely through high level composi-tional models, supporting a “plug and play”approach to software and hardware develop-ment. This tutorial is a comprehensive intro-duction to the science and art of building theexecutive kernel that provides the sense/re-sponse loop for such model-based au-tonomous systems.

The focus of building a model-based exec-utive kernel provides a framework for unify-ing a diverse set of research results. We dis-cuss representation formalisms, starting withcomponent-based propositional representa-tions and building up to concurrent transi-tion systems and qualitative algebras. We dis-cuss the core algorithms, starting with the

basics of model-based diagnosis, planning,real-time propositional reasoning, and modelcompilation, ultimately working towards amodel-based executive with extensive diag-nosis and planning embedded within the re-active control loop. We present a modeling“style” guide, where we discuss specific meth-ods for modeling a variety of different au-tonomous systems. We conclude with a dis-cussion of research on developing hybridmodel-based executives that coordinate con-tinuous adaptive estimation and controlmethods. Throughout the tutorial we illus-trate the issues with examples drawn fromfielded applications, including NASA’s firstautonomous space probe Deep Space One.

Brian Williams leads the in-telligent systems group atNASA Ames Research Cen-ter, and is a flight colead forthe autonomous spacecraft,Deep Space One. He re-ceived a Ph.D. at MIT, served

as guest editor of Artificial Intelligence and is onthe editorial board of JAIR. Research interestsinclude immobile robots, model-based autono-my, diagnosis, model-based learning, qualita-tive reasoning, and design.

Pandurang Nayak is a re-search scientist at theNASA Ames ResearchCenter. He holds a Ph.D. incomputer science from Stan-ford University (1992), andhis dissertation was an ACM

Distinguished Thesis. He is currently an asso-ciate editor of JAIR, and his research interestsinclude abstractions, model-based au-tonomous systems, diagnosis and recovery,qualitative and causal reasoning.

Modeling with Defaults: Causaland Temporal Reasoning (SP3)

Hector Geffner2:00 – 6:00 PM, Sunday, July 27

A robot pushes a block and expects the blockto move. The block however does not move.He pushes again but harder. The blockmoves. Inferences of this type are easy forpeople but hard for robots. Part of the prob-lem is that modeling languages in AI do notdeal with uncertainty in a natural way. Logi-cal languages do not handle uncertainty,while probabilistic languages deal with uncer-tainty at a precision and cost that is seldomneeded. Default languages are a new type ofmodeling languages that aim to fill the gapbetween the two languages, providing model-

ers with the means to map soft inputs intosoft outputs in a meaningful and principledway. Default models combine the conve-nience of logical languages with the flexibili-ty of a probabilistic semantics and the trans-parency of argumentation algorithms.

The goal of this tutorial is to provide a co-herent and self-contained survey of suchwork. I focus in particular on the problemsthat arise when modeling causality and time,analyzing what works, what doesn’t work,and why. I also illustrate the use of defaultlanguages in areas such as qualitative reason-ing, decision making, planning and control.The tutorial is intended for people interestedin commonsense modeling. There are noprerequisites except a basic knowledge oflogic and probabilities.

Hector Geffner earned hisPh.D. at UCLA under thesupervision of Judea Pearlwith a dissertation on de-fault reasoning that was thecowinner of the 1990 ACMDissertation Award. Then

he worked as Staff Research Member at theIBM T. J. Watson Research Center, beforereturning to the Universidad Simon Bolivarin Caracas, where he currently teaches.Geffner has served on the program commit-tees of the major AI conferences and is amember of the editorial board of the Journalof Artificial Intelligence Research.

Physics-Based Modeling for Vision and Virtual Human Animation (MA4)

Dimitris Metaxas and Norman Badler9:00 AM – 1:00 PM, Monday, July 28

In only a few years, the technology of virtualhuman modeling and animation has pro-gressed from crude wireframe robotic modelsto realistic shaded smooth deformable figures.Driven by underlying dramatic increases inworkstation compute power and 3D graphics,the future holds promise of ever more accu-rate (and beautiful) biomechanically robustmodels. The challenge, however, remains toprovide effective and easily learned user in-terfaces to control, manipulate and animatethese models. We discuss two coordinated di-rections: expressively effective and real-timephysics-based simulations, and high-levellanguage-motivated interfaces. This frame-work provides access to a rich behavioral vo-cabulary of parametrized virtual human ac-tions. Benefits to be realized from thesecombined approaches include enhanced vir-

12 Tutorial Forum

tual human behavior sets, articulated and de-formable motion capture from video andtheir use in animations, facial modeling, mul-ti-user team coordination studies, job train-ing, and instruction manual interpretationand generation. This tutorial will assume theparticipant has some knowledge of introduc-tory physics, mathematics, and programming,but will focus on the formulation, integra-tion, and application of the models ratherthan their implementation details.

Dimitris Metaxas is an as-sistant professor of computerand information science atthe University of Pennsylva-nia. He specializes inphysics-based modelingtechniques. He is the author

of Physics-Based Deformable Models: Applica-tions to Computer Vision, Graphics and MedicalImaging, a recipient of an NSF career award,and on ONR YIP.

Norman Badler is a profes-sor of computer and infor-mation science at the Uni-versity of Pennsylvania. Heis coeditor of Graphical Mod-els and Image Processing andcoauthor of Simulating Hu-

mans. He directs the Center for HumanModeling and Simulation, which has pro-duced the Jack virtual human software.

Practical Planning (MP3)Steve Chien and Brian Drabble2:00 – 6:00 PM, Monday, July 28

Automated planning is the generation of alow-level sequence of actions to achievesome desired world state while obeying do-main constraints. Planning systems can auto-mate procedure generation problems such asscience data analysis, image processing, crisisresponse, space payload operations, and com-munications antenna operations. Automatedplanning technology can reduce operationscosts, decrease manual errors, and reduce de-pendency on key personnel.

This tutorial covers basic domain-indepen-dent AI planning: search, representing plan-ning knowledge, plan and state space plan-ning, operator-based planning and hierarchical task network planning. Advanced con-cepts such as integrated planning and schedul-ing, decision theoretic planning, and mixedinitiative planning will also be briefly dis-cussed. This tutorial will answer the questions:• Are planning techniques applicable to my

problem?• What are the most appropriate planning

representations and techniques for myproblem?

• How to acquire, verify, and maintain, myplanning knowledge base?

• How to embed a planning system into anoperational setting?

This tutorial targets AI practioners seeking athorough overview of the state-of the art inAI planning technology and key issues infielding applications.; and planning and re-lated AI researchers seeking an overview ofthe current state of the art in AI planningand insights into key bottlenecks in fieldingAI planning systems.

Prerequisite knowledge of basic AI (search,logiclike representations) required; familiaritywith some planning and scheduling systems,basic search strategies, reactive systems,and/or scripting languages would be helpful.

Steve Chien is the techni-cal group supervisor of theArtificial IntelligenceGroup, at the Jet PropulsionLaboratory, California Insti-tute of Technology where heleads efforts in automated

planning and scheduling for spacecraft andground operations. He holds a BS, MS, andPh.D. in computer science, all from the Uni-versity of Illinois. Chien is also an adjunctassistant professor in computer science at theUniversity of Southern California.

Brian Drabble received hisPh.D. in artificial intelli-gence planning systemsfrom the University of As-ton in 1988. He is a seniorresearch associate at theComputational Intelligence

Research Laboratory at the University ofOregon. Previous to this, he was a member ofthe Artificial Intelligence Applications Insti-tute at the University of Edinburgh wherehas was project leader and coprincipal inves-tigator on the O-Plan project. His researchinterests are in intelligent planning andscheduling, temporal reasoning and qualita-tive reasoning. Drabble has presented tutori-als and courses on the practical applicationsof planning and scheduling to audiencesfrom both academia and industry.

Principles of OntologicalEngineering (SP4)

Michael Gruninger and Mike Uschold2:00 – 6:00 PM, Sunday, July 27

Disparate backgrounds, languages, tools, andtechniques are a major barrier to effectivecommunication among people, organisa-tions, and/or software systems. In this tutori-al, we show how the development and im-plementation of an ontology (an explicitaccount of a shared understanding in a givensubject area) can improve such communica-tion, which in turn can give rise to greaterreuse, sharing, interoperability, and more re-liable software.

After motivating their need, we clarify thedefinition of ontologies and the purposesthat they serve. The tutorial will be drivenby the examination of ontologies used inpractice in such diverse applications asprocesses, enterprise modelling, products,and engineering mathematics. After explor-ing concrete examples of ontologies, we out-line two approaches towards a methodologyfor developing and using ontologies, rangingfrom a suite of informal techniques to therole of formal languages and techniques inthe specification, implementation and evalu-ation of ontologies. Finally, we review thestate of the art and current practice in thisemerging field, considering various case stud-ies, software tools for ontology development,and future prospects.

The goal of the tutorial will be to provideparticipants with a working knowledge of thefield that will enable them to design and im-plement ontologies for various domains.

Mike Uschold has been ac-tively involved in the devel-opment and use of ontologiesfor over a decade. Uscholdreceived his bachelors degreein mathematics and physics,and a masters degree in com-

puter science from Rutgers University, wherehe worked as a research assistant on expert sys-tems projects until 1982. From 1983-1987 hewas a research associate and a lecturer in Edin-burgh. He received his Ph.D. in 1991, and hasbeen at AIAI since 1987. Uschold has beeninvolved in numerous ontology workshopsover the past few years.

Michael Gruninger hasbeen a research scientist inthe Enterprise IntegrationLaboratory at the Universi-ty of Toronto since 1993and is project manager ofthe Enterprise Engineering

Tutorial Forum 13

Project. He received a B.Sc. in computer sci-ence from the University of Alberta in 1987and his M.Sc. in computer science from theUniversity of Toronto in 1989. His doctoralwork at the University of Toronto has beenin the area of logic and object recognition incomputer vision, constructing ontologies tosupport 2D object recognition in scenes withocclusion.

Reinforcement Learning (SP1)Richard S. Sutton and Leslie Pack Kaelbling2:00 – 6:00 PM, Sunday, July 27

Reinforcement learning is learning about,from, and while interacting with an environ-ment in order to attain some objective. Inother words, it is a relatively direct model ofthe learning that people and animals do intheir normal lives. In the last two decades,this age-old problem has come to be muchbetter understood by integrating ideas frompsychology, optimal control, artificial neuralnetworks, and artificial intelligence. Newmethods and combinations of methods haveenabled much better solutions to large-scaleapplications than had been possible by all

other means. This tutorial will provide a top-down introduction to the field, coveringMarkov decision processes and approximatevalue functions as the formulation of theproblem, and dynamic programming, tempo-ral-difference learning, and Monte Carlomethods as the principal solution methods.The role of neural networks, evolutionarymethods, and planning will also be covered.The emphasis will be on understanding thecapabilities and appropriate role of each ofclass of methods within in an integrated sys-tem for learning and decision making.

Richard S. Sutton is one ofthe founders of the field ofreinforcement learning andthe author of the originalpaper on temporal-differ-ence learning. He is a seniorresearch scientist at the

University of Massachusetts in Amherst, andbefore that he worked for nine years at GTELaboratories. He is the author with AndrewBarto of a forthcoming text on reinforce-ment learning, to be published by The MITPress.Leslie Pack Kaelbling is associate professorof computer science at Brown University.

She previously held posi-tions at the Artificial Intel-ligence Center of SRI Inter-national and at TeleosResearch. She received anAB in philosophy in 1983and a Ph.D. in computer sci-

ence in 1990, both from Stanford University.Kaelbling has done substantial research onprogramming paradigms and languages forembedded systems, on mobile robot designand implementation, and on reinforcementlearning algorithms. Her current research di-rections include integrating learning mod-ules into systems programmed by humans, al-gorithms for learning and navigating usinghierarchical domain representations, andmethods for learning perceptual strategies.She is an NSF Presidential Faculty Fellow, amember of the AAAI Executive Council, amember of the IJCAI Advisory Committee,and the 1997 recipient of the IJCAI Com-puters and Thought Award.

Topics in the Theory of thePractice of Machine Learning(MA1)

Michael Kearns9:00 AM – 1:00 PM, Monday, July 28

This tutorial will concentrate on tools andtechniques from computational learning the-ory, statistics and related fields that directlybear on the practice of machine learning.The emphasis will be on understanding theunderlying ideas rather than on rigor and for-malization, and on how the various mathe-matical intuitions derived can inform practi-tioners. Possible topics include complexityregularization and model selection, boosting,and the VC dimension and its generaliza-tions. Wherever possible, examples from theexperimental literature will be used to high-light and examine the theory.

Michael Kearns received aPh.D. in computer sciencefrom Harvard University in1989 and was a postdoctoralfellow at MIT and the Inter-national Computer ScienceInstitute in Berkeley. He

joined AT&T Bell Labs in 1991, and is nowa principal member of technical staff in theMachine Learning and Information Re-trieval department at AT&T Labs Research.With Umesh Vazirani, he is coauthor of AnIntroduction to Computational Learning Theory(MIT Press 1994).

14 Tutorial Forum

Cou

rtes

y, P

rovi

denc

e / W

arw

ick

Con

vent

ion

& V

isito

rs B

urea

u

1997 AAAI Workshops (by invitation only)

Abstraction, Decisions, and Uncertainty

Contact: Christopher Geib; geib@cs.ubc.caMonday, July 28

AI and Knowledge Management Contact: Bradley Whitehall; brad.l.whitehall@JCI.comSunday, July 27

AI Approaches to Fraud Detection and Risk Management Contact: Tom Fawcett; fawcett@nynexst.com Sunday, July 27

Building Resource-Bounded Reasoning Systems

Contact: Shlomo Zilberstein; shlomo@cs.umass.edu Sunday, July 27

Constraints and Agents Contact: Eugene C. Freuder; ecf@cs.unh.edu Sunday, July 27

Deep Blue Versus Kasparov: The Significance for ArtificialIntelligence

Contact: Robert Morris; morris@cs.fit.edu Monday, July 28

Language and Space(two-day workshop)Contact: Patrick Olivier; plo@aber.ac.uk Sunday and Monday, July 27-28

Multiagent Learning Contact: Sandip Sen; sandip@kolkata.mcs.utulsa.edu Monday, July 28

On-Line Search Contact: Sven Koenig; skoenig=search@cs.cmu.edu Monday, July 28

Robots, Softbots, Immobots: Theories of Action, Planningand Control

Contact: Chitta Baral; chitta@cs.utep.edu Monday, July 28

Spatial and Temporal Reasoning Contact: Frank D. Anger; fanger@nsf.gov Monday, July 28

Using AI in Electronic Commerce, Virtual Organizationsand Enterprise Knowledge Management to Reengineer theCorporation

Contact: Daniel E. O’Leary; oleary@rcf.usc.edu Monday, July 28

Verification & Validation of Knowledge-Based Systems Contact: Robert Plant; rplant@umiami.miami.edu Sunday, July 27

Registration Information

AAAI-97/IAAI-97 Registration FeesYour AAAI-97 / IAAI-97 program registration includes admission toall sessions, invited talks, exhibitions, the Student Abstract Poster Ses-

sion, the opening reception, and the AAAI-97 / IAAI-97 ConferenceProceedings. Nonmember registration includes a one-year AAAI mem-bership. Onsite registration will be located outside Hall D, second lev-el, Rhode Island Convention Center, One Sabin Street, Providence,Rhode Island, 02903-1859.

Early Registration (Postmarked by May 28)AAAI MembersRegular $395 Students $120

NonmembersRegular $470 Students $185

Late Registration (Postmarked by June 25)AAAI MembersRegular $445 Students $145

NonmembersRegular $520 Students $210

On-Site Registration (Postmarked after June 25 or onsite.)AAAI MembersRegular $495 Students $170

NonmembersRegular $570 Students $235

Workshop RegistrationWorkshop registration is limited to those active participants deter-mined by the organizer prior to the conference. Individuals attendingworkshops but not the AAAI-97 conference must pay a $150.00 perworkshop registration fee. Workshop registration materials will be sentdirectly to invited participants. (The workshop schedule is preliminary,and subject to change without notice.)

Robot Building LabYour robot building lab registration includes admission to the robotbuilding lab and the exhibition program. Fees are $150.00 for membersor nonmembers, and $75.00 for students. Attendance is limited andearly registration is strongly encouraged.

AAAI-97 Tutorial Forum FeesYour tutorial program registration includes admission to no more thanfour consecutive tutorials and the corresponding four tutorial syllabi. Amaximum of four consecutive tutorials may be attended due to parallelschedules. Extra syllabi from other tutorials will be available for pur-chase onsite for $15.00 each. Your tutorial program registration also in-cludes admission to the exhibition program.

Early Registration (Postmarked by May 28)AAAI MembersRegular $170 Students $75

NonmembersRegular $225 Students $105

Late Registration (Postmarked by June 25)AAAI MembersRegular $200 Students $100

NonmembersRegular $255 Students $130

On-Site Registration (Postmarked after June 25 or onsite.)AAAI MembersRegular $230 Students $125

NonmembersRegular $300 Students $155

Workshops & Registration Information 15

Payment InformationPrepayment of registration fees is required.Checks, international money orders, banktransfers, and traveler’s checks must be in USdollars. American Express, MasterCard,VISA, and government purchase orders arealso accepted. Registration applications post-marked after the early registration deadlinewill be subject to the late registration fees.Registration applications postmarked afterthe late registration deadline will be subjectto on-site registration fees. Student registra-tions must be accompanied by proof of full-time student status.

Refund RequestsThe deadline for refund requests is July 7,1997. All refund requests must be made inwriting. A $75.00 processing fee will be as-sessed for all refunds.

Registration HoursRegistration hours will be Sunday and Mon-day, July 27-28, 7:30 AM – 6:00 PM; Tuesdayand Wednesday, July 29-30, 8:00 AM – 6:00PM; and Thursday, July 31, 8:00 AM – 3:00PM. All attendees must pick up their registra-tion packets for admittance to programs.

HousingAAAI has reserved a block of rooms in Prov-idence properties at reduced conferencerates. Conference attendees must contact thehotels directly and identify themselves asAAAI-97 registrants to qualify for the re-duced rates. Important! Attendees mustsubmit their name, address, fax and phonenumbers when making reservations. Pleasenote the cut-off date for reservations and thereservation method/information under eachhotel. Hotel rooms are priced as singles (1person, 1 bed), doubles (2 persons, 2 beds),triples (3 persons, 2 beds), or quads (4 per-sons, 2 beds). Rooms will be assigned on afirst-come, first-served basis. All rooms aresubject to a 12% state and city tax.

The Westin Hotel Providence(Headquarters Hotel)

One West Exchange StreetProvidence, RI 02903Reservations: 1-800-WESTIN1Telephone: 401-598-8000Fax: 401-598-8290Single/Double/Triple/Quad: $115.00Additional Person: $20.00Distance to center: connectedCut-off date for reservations: 5:00 pmEDT June 25, 1997.

All reservation requests for arrival after 6:00PM must be accompanied by a first night roomdeposit, or guaranteed with a major creditcard. The Westin Hotel Providence will nothold any reservations after 6:00 PM unlessguaranteed by one of the above methods.Reservations received after the cut-off timewill be accepted on a space available basis.

Reservations accepted without a creditcard guarantee or advance deposit are subjectto cancellation at 6:00 PM on the day of ar-rival. Departure date changes resulting in ashortened stay at the hotel made after check-in to the hotel are subject to a $25.00 changefee plus applicable taxes.

Providence BiltmoreKennedy PlazaProvidence, RI 02903Telephone: 401-421-0700Fax: 401-455-3050Single/double: Superior room $109.00Single/double: Deluxe room $115.00Distance to center: Two blocksCut-off date for reservations: 5:00 pmEDT July 3, 1997.

All reservations must be accompanied byone night’s deposit or credit card guarantee.Room type requests are processed accordingto the availability at the time of reservation.Deposits will be retained if the individualtraveler does not arrive or cancels within 48hours of arrival.

Holiday Inn Providence Downtown

I-95 at AtwellsProvidence, RI 02903Telephone: 401-831-3900Fax: 401-751-0007Single: $79.00Double: $89.00Triple: $99.00Quad: $109.00Distance to center: One blockCut-off date for reservations: 5:00 pmEDT July 2, 1997.

Rooms to be guaranteed by credit card or ad-vance deposit of first nights room and tax.Individuals may cancel up until 6pm night ofarrival without penalty. Complimentaryshuttle service to and from T. F. Green Air-port based upon availability.

Student Housing, Brown UniversityAAAI-97 has reserved a block of dormitoryrooms at Brown University for student hous-ing during the conference. Accommodationsare double or single rooms with shared maleor female washrooms. Linen and towels areprovided upon arrival, but are not changedduring the conference. Telephone service isavailable in each room, but there are no tele-phones in rooms. Guests must bring theirown telephone instruments. The lines aretoll-direct lines, and local telephone calls arefree of charge. Long distance telephone callsmust be made by using a credit card, callingcard or call collect. Breakfast is available inthe Sharpe Refectory from 7:30-9:00 AM, butis not included in the housing package. Pub-lic transportation is available to the RhodeIsland Convention Center. More detailedgeneral information will be included withreservation confirmations.

Double room rate per person and night is$32.00 in non-airconditioned rooms and sin-gle room rate per person and night is $36.00in non-airconditioned rooms or $45.00 inairconditioned rooms (proof of full time stu-dent status must be included with the hous-ing reservation form). Reservations must bemade by June 21, 1997. Reservations re-ceived by June 21, 1997 will be confirmed. Areservation form is enclosed in this brochure.Room reservations must be guaranteed bycompleting the credit card authorization partof the housing reservation form. VISA, Mas-tercard and American Express only. Cancel-lations or changes must be submitted in writ-ing. One night’s payment is not refundable ifcancellation occurs after July 21, 1997.Reservations should be sent to:

AAAI-97 Student HousingBrown UniversityConference ServicesBox 1864Providence, RI 02912or fax to: 401-863-7300Parking is available on campus and the

cost is $1.00/day. Please use the reservationform to sign up for parking passes.

Air Transportation and Car RentalProvidence, Rhode Island—Get there for less!Discounted fares have been negotiated forthis event. Call Conventions in America at800-929-4242 and ask for Group #428. Youwill receive 5% – 10% off the lowest applic-able fares on American Airlines, or the guar-

16 Housing

anteed lowest available fare on any carrier.Travel between July 24 and August 3, 1997.All attendees booking through CIA will re-ceive free flight insurance and be entered intheir bi-monthly drawing for worldwide trav-el for two on American Airlines! Hertz RentA Car is also offering special low conferencerates, with unlimited free mileage. Call Con-ventions in America at 800-929-4242, askfor Group #428. Reservation hours: M-F 6:30am–5:00 pm Pacific Time. Outside US andCanada, call 619-453-3686 / Fax 619-453-7679. Internet: scitravel@aol.com/24-houremergency service 1-800-748-5520. If youcall direct: American 800-433-1790, ask forindex #S 9485. Hertz 800-654-2240, ask forCV#24250.

Ground TransportationThe following information is the best avail-able at press time. Please confirm fares whenmaking reservations.

Airport ConnectionsAirport Van Shuttle

Telephone: 401-736-1900Located at T.F. Green Airport$9.00 one way, $14.00 round tripContact: Richard Sprague

Cline TransportationTelephone: 401-751-2546$9.00 one way, $16.00 round trip, Individual car or limo is $24.00 T. F.

Green Airport to ProvidenceAdvance reservations neededContact: Linda Cline

Hotel ShuttlesComplimentary, Providence-Holiday InnHotel

TaxiTaxis are available at T. F. Green Airport.Approximate fare from the airport to down-town Providence is $24.00.

BusBonanza Bus Lines—New York, Boston. Thedepot is located at Kennedy Plaza. For infor-mation on fares and scheduling, call 401-751-8800.

RailThe Amtrak station is located at CapitolHill. For general information and ticketing,call 800-872-7245.

City Transit SystemRhode Island Public Transit Authority (RIP-

TA) is a statewide bus transit. Schedules areavailable at the main depot located atKennedy Center (across from the BiltmoreHotel). Basic local fare is $1.00. For generalinformation call 800-244-0444.

ParkingA parking garage is available at the RhodeIsland Convention Center. The maximumdaily rate is $8.50.

DisclaimerIn offering American Airlines, the BiltmoreHotel, Brown University, Conventions inAmerica, Hertz Rent A Car, the HolidayInn, the Westin Hotel and all other serviceproviders, (hereinafter referred to as “Suppli-er(s)” for the National Conference on Ar-tificial Intelligence and the Innovative Ap-plications Conference, AAAI acts only inthe capacity of agent for the Suppliers thatare the providers of the service. BecauseAAAI has no control over the personnel,equipment or operations of providers of ac-commodations or other services included aspart of the AAAI-97 or IAAI-97 program,AAAI assumes no responsibility for and will

not be liable for any personal delay, inconve-niences or other damage suffered by confer-ence participants which may arise by reasonof (1) any wrongful or negligent acts or omis-sions on the part of any Supplier or its em-ployees, (2) any defect in or failure of any ve-hicle, equipment or instrumentality owned,operated or otherwise used by any Supplier,or (3) any wrongful or negligent acts or omis-sions on the part of any other party not un-der the control, direct or otherwise, ofAAAI.

Providence Visitor InformationThe Providence Warwick Convention &Visitors Bureau welcomes you to Providence!They can assist with dining reservations, di-rections, tour bookings, entertainment sug-gestions, transportation, and hotel informa-tion. Maps and brochures are available.

One West Exchange StreetProvidence, RI 02903Telephone: 401-274-1636URL: http://www.providenceri.com/home.html

Transportation 17

Cou

rtes

y, P

rovi

denc

e / W

arw

ick

Con

vent

ion

& V

isito

rs B

urea

u

18 Conferences at a Glance

MORNING AFTERNOON EVENING

Sunday, July 27Registration RegistrationTutorial Forum Tutorial ForumWorkshops WorkshopsRBL–97 RBL–97

Monday, July 28Registration Registration 1997 Fellows DinnerTutorial Forum Tutorial ForumWorkshops WorkshopsIAAI–97 IAAI–97AAAI/SIGART DC AAAI/SIGART DCRBL–97 RBL–97

Tuesday, July 29Registration Registration AAAI–97 Opening ReceptionAAAI–97 AAAI–97 Student Poster SessionKeynote & Invited Talks Invited TalksIAAI–97 IAAI–97Exhibition ExhibitionRobot Competition Robot CompetitionHall of Champions Hall of Champions

Wednesday, July 30Registration Registration Program Committee DinnerAAAI–97 AAAI–97Invited Talks Invited TalksIAAI–97 IAAI–97Exhibition ExhibitionRobot Competition Robot CompetitionHall of Champions Hall of Champions

Thursday, July 31Registration RegistrationAAAI–97 AAAI–97Invited Talks Invited TalksExhibition ExhibitionRobot Competition Robot CompetitionHall of Champions Hall of Champions

Brown University Housing Registration FormAmerican Association for Artificial Intelligence

July 25 - August 1, 1997Brown University

Providence, RI 02912

PLEASE PRINT:NAME: __________________________________________________________________ SEX: ■■ M ■■ F

Last First Middle Initial

MAILING ADDRESS: ______________________________________________________________________

______________________________________________________________________

______________________________________________________________________

TELEPHONE NUMBER: ________________________________ ________________________________Home Business

E-MAIL ADDRESS: __________________________________________________________________________

ARRIVAL DATE: ____________________________________ TIME: ______________________________

DEPARTURE DATE: ________________________________ TIME: ______________________________

HOUSING PREFERRED:

______ Nights in an air-conditioned single room @ $45.00/person/night $ __________________

______ Nights in a single dormitory room @ $36.00/person/night $ __________________

______ Nights in a double dormitory room @ $32.00/person/night $ __________________

NAME OF PERSON SHARING WITH YOU:

____________________________________________________

■■ Please assign a roommate for me.

(If we are unable to assign you a roommate, you will be billed for a single room.)

TOTAL HOUSING: $ __________________

PARKING: (Necessary only if staying in on-campus housing)

______ nights @ $1.00/night x ______ cars TOTAL PARKING: $ __________________

TOTAL DUE: $ __________________

continued on reverse

PAYMENT AUTHORIZATION: Your room reservation must be guaranteed by completing this credit card authorization form. Master Card, VISA and American Express only. One night’s payment is not refundable if cancellation occurs after July 21, 1997.

____________________________ __________________________________________ ____________Credit Card Type Credit Card Number Expiration Date(Master Card, VISA, Am EX only)

____________________________ __________________________________________Name on Credit Card Signature of Authorization(Please print)

■■ If you intend to pay by Purchase Order, please check here. You must list credit card

information to guarantee your reservation until the purchase order is received.

Please return this housing registration form by June 21, 1997 to: Brown University, Conference Services,Box 1864, Providence, RI 02912. Or FAX to (401) 863-7300. Reservations cannot be made by tele-phone. Reservations received by June 21 will be confirmed. Cancellations or changes must be submittedin writing directly with Brown University Conference Services.

PLEASE RETURN THIS FORM BY JUNE 21, 1997.

Please indicate below if you have any physical impairments or dietary requirements which we shouldknow about to help make your stay on the Brown campus as comfortable as possible:

Providence, Rhode Island 23

Non-Profit Org.U.S. Postage

PAIDPermit #132

Menlo Park, CA

American Association forArtificial Intelligence445 Burgess DriveMenlo Park, California 94025-3496(415) 328-3123

Postmaster: Time Dependent Material

Inside:

AAAI-97 / IAAI-97 Highlights / 2Keynote Address / 3AAAI-97/IAAI-97 Invited Talks / 3–5Exhibits / 5Hall of Champions (NEW!) / 5Mobile Robot Competition / 6Robot Building Laboratory / 6Paper Session Schedule / 7Tutorial Forum / 8–14Workshops / 15Registration/Housing / 15–16Conference at a Glance / 18Transportation / 17