Fifteenth National Conference on Artificial Intelligence (AAAI-) Tenth Conference on Innovative Applications of Artificial Intelligence (IAAI-) July -,  Monona Terrace, Madison, Wisconsin

Sponsored by the American Association for Artificial Intelligence Cosponsored by DARPA, NASA Ames Research Center, Microsoft Corporation, and the National Science Foundation In cooperation with the University of Wisconsin–Madison

Program & Exhibit Guide ■ Mobile Robot Exhibition Cochairs Contents Tucker Balch, Georgia Institute of Technology Karen Zita Haigh, Carnegie Mellon University Acknowledgments /  Conference at a Glance /  ■ Robot Building Laboratory, Chair DC- /  David Miller, KISS Institute for Practical Exhibition /  Robotics General Information /  ■ SIGART/AAAI- Doctoral Consortium Hall of Champions /  Chair IAAI- Program / - Janyce Wiebe, New Mexico State University Intelligent Systems Demonstrations /  ■ Student Abstract and Poster Chair Invited Talks / -, - Michael Littman, Duke University Registration /  ■ Tutorial Forum Cochairs Robot Building Laboratory /  Padhraic Smyth, University of California, Robot Competition and Exhibition /  Irvine Special Events and Programs / - Bart Selman, Cornell University Special Meetings /  Sponsoring Organizations /  ■ Workshop Chair and Cochair Technical Program Tuesday / - David Leake, Indiana University Technical Program Wednesday / - Raymond C. Mooney, University of Texas at Technical Program Thursday / - Austin Tutorial Forum / - A complete listing of the AAAI- and IAAI- Workshop Program / - Program Committee members appears in the conference proceedings. Acknowledgments Sponsoring The American Association for Artificial Intelli- gence wishes to acknowledge and thank the fol- Organizations lowing individuals for their generous contribu- tions of time and energy to the successful cre- AAAI gratefully acknowledges the generous ation and planning of the Fifteenth National contributions of the following organizations to  Conference on Artificial Intelligence and the AAAI- : Tenth Conference on Innovative Applications ■ ACM/SIGART of Artificial Intelligence. ■ Defense Advance Research Projects Agency ■ ■ AAAI- Program Cochairs Microsoft Corporation ■ Jack Mostow, Carnegie Mellon University NASA Ames Research Center ■ Chuck Rich, MERL – A Mitsubishi Electric National Science Foundation ■ Research Laboratory Office of Naval Research ■ University of Wisconsin – Madison ■ IAAI- Conference Chair: Bruce G. Buchanan, University of Pittsburgh ■ IAAI- Conference Cochair Ramasamy Uthurusamy, General Motors Research ■ Hall of Champions Chair and Cochair Jonathan Schaeffer, University of Alberta Dana Nau, University of Maryland ■ Intelligent Systems Demonstrations Cochairs

Contents / Acknowledgments Contents / Acknowledgments George Ferguson, University of Rochester Randolph M. Jones, University of Michigan ■ Mobile Robot Competition Cochairs Gregory Dudek, McGill University Robin Murphy, Colorado School of Mines David Kortenkamp, NASA Johnson Space Center

 Special Events & Programs . –   : PM   :   : from from  – from from registrants. A registrants.  per person fee  ,  :  for children) will for children)    .   .  from from  Rendezvous Rendezvous will be held in the will be Rendezvous Opening Reception is sponsored Reception Opening opening reception will be held in opening reception     registrants. A $ registrants. . This informal gathering will give at- gathering will give This informal . . This event will provide the traditional provide will This event . per person fee ($ PM for children) will be charged for spouses for children)  PM   .  : .     : be charged for spouses and other non-technical avail- tickets are Guest registrants. conference able in onsite registration. $ the Grand Terrace of the Monona Terrace Con- Terrace the Monona of Terrace the Grand July Tuesday, Center, vention AI Festival Exhibition will be held in the The AI Festival Cen- Convention Terrace of the Monona Hall July Wednesday, ter, with the opportuni- you will provide This event events— exciting among numerous ty to stroll Competition and Exhibition, Robot the Mobile Systems of Champions, the Intelligent the Hall by Posters—enlivened and the Student Demos, informal supper and conversation. Admittance to AAAI- is free to the reception tendees the opportunity to mingle in a relaxed Light snacks will be available. Su- atmosphere. Coordi- Social the Rendezvous santha Herath, attendees in- organizing interested will be nator, out to dinner after the to go to small groups Rendezvous. The AAAI- Opening Reception Opening  opportunity at the end for attendees to socialize of of the first day of technical sessions. A variety and a no-host bar will be avail- hors d’oeuvres to is free to the reception able. Admittance AAAI- ($ AAAI– The AAAI- – registrants. and other non-technical conference in onsite registration. available tickets are Guest The AAAI- Grand Terrace of the Monona Terrace Conven- Terrace of the Monona Terrace Grand July on Monday, tion Center Corporation. AAAI grate- in part Microsoft by generous contri- Microsoft’s fully acknowledges bution in support of this event. , : in in  and  PM PM  . (By in- . (By   : : . PM   Outstanding PM – –  :    :   : :    –  :  SRI International , from , from , from , from University of Southern University USC/Information USC/Information   Stanford University Stanford from from  Fellows Recognition Dinner will be Dinner Recognition Fellows AAAI Fellows Fellows AAAI in the Madison Ballroom, Monona Ballroom, in the Madison Program Committee Dinner will be Committee Dinner Program Program Committees. (By invitation Committees. (By Program followed by dinner at by followed   AM PM  California Sciences Institute Shoham, Yoav Wilkins, E. David George A. Bekey, A. Bekey, George Minton, Steven   : : The presentation of the three papers that have of the three The presentation as the AAAI- been recognized the Madison Ballroom, second floor, the Madi- second floor, Ballroom, the Madison to honor the contribu- son Concourse Hotel tions of all the members of the AAAI- IAAI- Session Paper into one special session will be combined Papers on in the invited talk track of the conference July Tuesday, only.) Outstanding Terrace Convention Center. Convention Terrace Presidential Address Presidential will Institute, NEC Research Waltz, L. David on “The Im- Address the AAAI Presidential give July Tuesday, on portance of Importance” the Capitol Ballroom A, second floor, the Madi- A, second floor, the Capitol Ballroom will begin at A reception son Concourse Hotel.   held Tuesday, July held Tuesday, Committee Dinner AAAI- Program Program The vitation only). held Monday, July July held Monday, ■ ■ ■ ■ Each year the American Association for Artifi- Association for the American Each year small number of a recognizes cial Intelligence made significant sustained members who have field of artificialcontributions to the intelli- attained unusual distinc- gence, and who have AAAI is pleased to an- tion in the profession. for newly elected Fellows nounce the three  Dinner Recognition Student Abstract Annual Poster Program Business Meeting Students whose abstracts were chosen for inclu- The Annual Business Meeting will be held sion in the conference proceedings will display Thursday, July , from : – : PM in the their work at the Student Abstract Poster Ses- Hall of Ideas E & F, Monona Terrace Conven- sion in the Exhibition Hall, Monona Terrace tion Center. Convention Center on Wednesday, July  from : – : PM in conjunction with the AI Festival. In addition, participants in the Conference AAAI/SIGART Doctoral Consortium will dis- play their poster presentations during this ses- Committee Meeting sion. All students will be available for questions. The AAAI Conference Committee Meeting will The AAAI- Student Abstract Poster Program be held Wednesday, July , from : – : is sponsored by Microsoft Corporation. AAAI AM in the University Room A, second floor, gratefully acknowledges Microsoft’s generous Madison Concourse Hotel. contribution in support of this program.

AAAI/SIGART Doctoral Executive Council Consortium (DC-) Meeting The AAAI Executive Council Meeting will be The Third AAAI/SIGART Doctoral Consor- held Sunday, July , from : AM – : PM in tium program will be held on Sunday and Mon- the University Room A, second floor, Madison day, July  – ,  from : – : PM in Concourse Hotel. Continental breakfast will be the Senate Room, Madison Concourse Hotel. available at : AM. The Doctoral Consortium provides an opportu- nity for a group of Ph.D students to discuss and explore their research interests and career objec- AAAI Press Editorial tives in an interdisciplinary workshop together with a panel of established researchers. The six- Board Meeting teen students accepted to participate in this pro- gram will also participate in the Student Poster The AAAI Press Editorial Board Meeting will be program on Wednesday, July , from : – held Wednesday, July , from : – : PM : PM during the AI Festival. All interested in the Wisconsin Room, second level, Monona AAAI- student registrants are invited to ob- Terrace Convention Center. serve the presentations and participate in dis- cussions at the workshop. AAAI and ACM/ SIGART gratefully acknowledge grants from AAAI Publications the Office of Naval Research and Microsoft Committee Meeting Corporation for student travel to this event. The AAAI Publications Committee lunch meet- ing will be held Tuesday, July , from : – : PM in the Wisconsin Room, second level, Monona Terrace Convention Center.

Student Programs & Meetings Programs Student AIJ Editorial Board Meeting The AIJ Editorial Board lunch meeting will be held Monday, July , from : – : PM in the University Room A, second floor, Madison Concourse Hotel.

 Conference at a Glance   998 Fellows Dinner 998 Fellows      RBL– RBL– IAAI– Workshop W Workshop     ULY      ULY  , J ULY , J ULY , J RBL– RBL– IAAI– ULY AAAI-98 AAAI-98 MORNING AFTERNOON EVENING , J Workshops Workshops Workshops Workshops Registration Registration Registration Registration Hall AI Festival/Exhibition Registration Registration Reception Opening Registration Registration Rendezvous Registration Registration Invited Talks Invited Invited TalksInvited Talks Invited , J Workshop W Workshop Tutorial ForumTutorial Forum Tutorial MP Tutorial Special Tutorial ForumTutorial Forum Tutorial Robot ProgramRobot Program Robot Robot ProgramRobot Program Robot Program Robot Robot ProgramRobot Program Robot AAAI-98/IAAI-98 AAAI-98/IAAI-98 Session Poster Student AAAI-98/IAAI-98 AAAI-98/IAAI-98 Committee Dinner Prog. AAAI/SIGART DC AAAI/SIGART DC AAAI/SIGART DCAAAI/SIGART DC AAAI/SIGART Hall of Champions of Hall of Champions Hall Champions of Hall of Champions Hall of Champions Hall Hall of ChampionsHall of Champions Hall Exhibition/IS DemosExhibition/IS Demos Exhibition/IS Exhibition/IS DemosExhibition/IS Demos Exhibition/IS Demos Exhibition/IS Exhibition/IS DemosExhibition/IS Demos Exhibition/IS DAY EDNESDAY HURSDAY UESDAY UNDAY Presidential Address / Talks / Address Presidential Talks Invited Fri T W T Monday, July July Monday, S Tutorial Forum SP SA wish toobtainsyllabifromothertutorialsmaypurchase themseparatelyfor$ Tutorial attendeesmayredeem theirtutorialsyllabiticketsatthe rooms. Attendees who tutorial syllabi.Amaximumoffourconsecutive tutorialsmaybetakenduetoparallelschedules. Tutorial registration forum includesadmissiontoupfourtutorialsandthecorresponding four Tutorial Forum  MA : Session III:Monday, July SP SP SP : ‒: Session II:Sunday, July SA SA SA : Session I:Sunday, July fee. onsite registration. The Special Tutorial (MP         Hall ofIdeas H&I,Monona Terrace Convention Center Carla P. Gomes, Ken McAloon andCarol Tretkoff Hall ofIdeas E&F, Monona Terrace Convention Center Rina Dechter andIrina Rish Hall ofIdeas H&I,Monona Terrace Convention Center Marti A.Hearst andMichael J.Pazzani Hall ofIdeas E&F, Monona Terrace Convention Center Craig KnoblockandQiangChung Yang Madison Ballroom C,Monona Terrace Convention Center Michael I.Jordan Madison Ballroom D,Monona Terrace Convention Center Pandu Nayak andBrian Williams Madison Ballroom C,Monona Terrace Convention Center Udo Hahn andInderjeet Mani Madison Ballroom D,Monona Terrace Convention Center Tuomas Sandholm Madison Ballroom C,Monona Terrace Convention Center Rick Lathrop  : : : : : : : : AM AM : Computational Molecular Biology Intelligence: andArtificial AnIntroduction Automatic TextSummarization Recent Advances inAIPlanning Principles andStrategies ofAutomated Inference: AUnifying View Economically Founded Multiagent Systems Model-Based Autonomous Systems Advanced Techniques forInformation Access Integration Intelligence ofArtificial andOperationsResearch Techniques Graphical Models and Variational Approximation ‒ : – : PM PM PM     ) isopentoallAAAI-  registrants fornoadditional  .  per syllabusin Tutorial Forum & RBL  Robot  , in the Madison and , in the Madison  –   PM PM Computational Aspects of Knowledge Representations Computational Aspects of Knowledge Statistical Methods in Natural Language Processing in Natural Methods Statistical Intelligent Simulation Intelligent Genetic Algorithms, Operations Research and AI Research Operations Algorithms, Genetic Support Vector Learning Vector Support From Action Theories to Agent-Planning Control Strategies for Reactive Agents for Reactive Strategies to Agent-PlanningTheories Control Action From Getting that First Grant: A Young Scientist’s Guide to (AI) Funding in America to (AI) Funding Guide Scientist’s Young A Grant: that First Getting Learning Bayesian Networks from Data Networks Bayesian Learning : : : : : : : :         Marco Cadoli and Thomas Eiter Cadoli and Marco Center Convention Terrace H & I, Monona of Ideas Hall Hendler Jim Center Convention Terrace Monona E & F, of Ideas Hall Bernhard Schoelkopf Bernhard Center Convention Terrace Monona E & F, of Ideas Hall Feng Zhao and Chris Bailey-Kellog Feng Center Convention Terrace D, Monona Ballroom Madison Nir Friedman and Moises Goldszmidt and Moises Friedman Nir Center Convention Terrace C, Monona Ballroom Madison John LaffertyLillian Lee and John Center Convention Terrace H & I, Monona of Ideas Hall Darrell Whitley Darrell Center Convention Terrace D, Monona Ballroom Madison Kabanza Froduald and Chitta Baral Center Convention Terrace Monona E & F, of Ideas Hall Robot Building Lab Building Robot July and Monday, Laboratory Building will be held Sunday The Robot Wisconsin Ballrooms, Madison Concourse Hotel. Preregistration is required. AAAI- is required. Preregistration Concourse Hotel. Madison Ballrooms, Wisconsin LaboratoryBuilding easy or difficult it is to implement participants will spend the day seeing how will be grouped into small teams, each AI techniques on an actual robot. Participants their favorite RBL will startThe ba- with a quick tutorial on robot mobile robot. of which will build their own will spend most techniques. Participants programming sensors, effectors and real-time sics covering the laboratoryThroughout their mobile robot. of their time designing, building and programming specific aspects of robot design and programming. will be individual team tutorials covering there li- and an extensive systems and technologies will also take place, of other robot Demonstrations portions will be provided brary of the mobility system Some functions will be available. of robot will be There get a good start on a fully functional robot. assuring that all groups thereby prebuilt, the end of the session At testing and redesign. ample opportunity individual design, creativity, for has will see which robot Then we all the robots will participate in a double elimination tournament. lab)! of the robot the right stuff to best accomplish the task (which will be specified at the beginning and taught The lab is being organized attendees. This tournament will be open to all the conference from and assistants are Instructors (KIPR) for AAAI. Robotics for Practical the KISS Institute by is the lead instructor. Miller David trained staff. KIPR’s : ‒ : MP MP MP : ‒ : MP MP Session IV: Monday, July July Monday, IV: Session MA MA MA Workshop Program Attendance at the workshops is limited, and participation is by invitation only. All workshop par- ticipants must register for the AAAI- technical program or, in the case of the four cosponsored workshops, must register for one of the cosponsoring conferences. (Exceptions to these rules will be required to pay a $. fee per workshop.) Registration onsite for a workshop is possible with the prior permission of the corresponding workshop organizer. All workshops will begin at : AM and conclude at : PM, unless otherwise noted below.

Sunday, July 

W: AI and Information Integration (-/ day workshop) Organizers: Craig Knoblock and Alon Levy : – : PM, University Room, Inn on the Park W: Integrating Artificial Intelligence and Assistive Technology Organizer: Rich Simpson Capitol Room West, Inn on the Park W: Knowledge Sharing across Biological and Medical Knowledge Based Systems Organizer: Gary Merrill and Dhiraj Pathak Board Room, Inn on the Park W: Recommender Systems Organizer: Henry Kautz Capitol Room East, Inn on the Park W: Representations for Multi-Modal Human-Computer Interaction (-day workshop) Organizers: Syed Ali and Susan McRoy Lower Level #, Inn on the Park W: Software Tools for Developing Agents

Workshop Program Workshop Organizers: Brian Logan and Jeremy Baxter Madison Room, Inn on the Park W: Textual Case-Based Reasoning Organizers: Mario Lenz and Kevin Ashley Lower Level #, Inn on the Park W: Using AI for Knowledge Management and Business Process Reengineering Organizer: Rose Gamble Lower Level #, Inn on the Park

Monday, July 

W: AI and Information Integration (-/ day workshop) Organizers: Craig Knoblock and Alon Levy : AM – : PM, University Room, Inn on the Park W: Case-Based Reasoning Integrations Organizers: David Aha and Jody Daniels Madison Room, Inn on the Park W: Functional Modeling and Teleological Reasoning Organizer: Jon Sticklen Lower Level #, Inn on the Park

 W: Learning for Text Categorization Invited Talks Jointly Sponsored by the International Conference on Machine Learning Organizer: Mehran Sahami Capitol Room East, Inn on the Park W: The Methodology of Applying Machine Learning: Problem Definition, Task Decompo- sition and Technique Selection Jointly Sponsored by the International Conference on Machine Learning Organizer: Robert Engels Board Room, Inn on the Park W: Predicting the Future: AI Approaches to Time-Series Analysis Jointly Sponsored by the International Conference on Machine Learning Organizer: Andrea Danyluk Lower Level #, Inn on the Park W: Representations for Multi-Modal Human-Computer Interaction (-day workshop) Organizers: Syed Ali and Susan McRoy Lower Level #, Inn on the Park W: Verification & Validation of Knowledge-Based Systems Organizer: Daniel O’Leary and Alun Preece Capitol Room West, Inn on the Park

Friday, July 

W: The Grounding of Word Meaning: Data and Models Jointly Sponsored by the Cognitive Science Society Organizer: Michael Gasser Capitol Ballroom A, the Madison Concourse Hotel

AAAI-/IAAI- Invited Talks All AAAI-/IAAI- invited presentations will be held in the Madison Ballroom, fourth level, Monona Terrace Convention Center, unless otherwise noted.

Tuesday, July 

: – : AM AAAI/Presidential Address: The Importance of Importance David L.Waltz, NEC Research Institute Introduction by Randall Davis (Past President, AAAI), MIT AI Lab Human intelligence is shaped by what we care most about—the things that cause ecstasy, despair, pleasure, pain, terror, security, satisfaction, and other intense emotions. Any system we would consider truly intelligent will depend critically on the ability to separate the important from among the unimportant. This ability underlies such faculties as attention, focus- ing, situation and outcome assessment, priority setting, judgment, taste, goal selection, credit assignment, the selection of relevant memories and precedents, assessment of meaning and significance; all of these are impor- tant in learning from experience. AI has for the most part focused on logic and reasoning in artificial situations where only relevant variables and op- erators are specified, and has paid insufficient attention to processes of re- ducing the richness and disorganization of the real world to a form where logical reasoning can be applied. This talk will discuss the role of impor- tance in intelligence, provide some examples of research that makes use of  importance judgments, and offer suggestions for new mechanisms, archi- tectures, applications and research directions for AI.

: – : AM Invited Panel: Eight Cool Things from the Collocated Conferences Organizer: Charles Rich (AAAI- Program Cochair), MERL—A Mit- subishi Electric Research Laboratory The following eight organizations have chosen to hold their meetings in Madison contiguous with AAAI- this year: ILP ‘, GP-, SGA-, COLT ‘, ICML ‘, UAI-, ST&D, and CogSci. In honor of this special occasion, we have invited a chairperson from each of these confer- ences to join a panel to answer the following question: “What is the most important recent result/experiment/discovery in the area of your confer- ence that the general AI audience doesn’t know or understand or appreci- ate, but should (and why)?”

: AM – : PM Invited Talk: Learning Sparse Representations: Machine Learning, Machine Vision and the Brain Tomaso Poggio, Massachusetts Institute of Technology Learning is becoming the central problem in trying to understand intelli- gence and in trying to develop intelligent machines. Poggio will outline Invited Talks Invited some of the recent efforts in the domain of vision to develop machines that learn and to understand the brain mechanisms of learning.

: – : PM Invited Talk: Modeling Satisfaction and Satisfactory Modeling: Modeling Problems So Constraint Engines Can Solve Them Eugene C. Freuder, University of New Hampshire Introduction by David Waltz, NEC Research Institute A wide variety of problems can be modeled as constraint satisfaction (or optimization) problems. Once they are so modeled, powerful search and inference methods can be brought to bear. Modeling itself, however, pre- sents a series of challenges. The ultimate challenge is to automate the mod- eling process.

: – : PM Special AAAI- Outstanding Paper Session The presentation of the three papers that have been recognized as the AAAI- Outstanding Papers will be combined into one special session in the invited talk track of the conference. The papers, listed alphabetically by first author, are: Learning Evaluation Functions for Global Optimization and Boolean Satisfiability Justin A. Boyan and Andrew W. Moore, Carnegie Mellon University The Interactive Museum Tour-Guide Robot Wolfram Burgard, Armin B. Cremers, Dieter Fox and Dirk Haehnel, Uni- versity of Bonn; Gerhard Lakemeyer, University of Aachen; Dirk Schulz and Walter Steiner, University of Bonn; Sebastian Thrun, Carnegie Mellon University Acceleration Methods for Numeric CSPs Yahia Lebbah and Olivier Lhomme, Ecole des Mines de Nantes (France)

: – : PM IAAI- Invited Panel: Hall of Champions Lecture Hall, Level , Monona Terrace Convention Center Organizer: Dana Nau, University of Maryland Panelists: David Fotland, Hewlett Packard; Jonathan Schaeffer, University of Alberta, Gerald Tesauro, IBM Research; and David Wilkins, SRI Inter- national  Wednesday, July  Invited Talks

: – : AM Invited Talk: How People Treat Computers Like Real People: Experimental Evidence of a New Paradigm Clifford Nass, Stanford University Introduction by Howard E. Shrobe, Massachusetts Institute of Technology This talk will describe a series of experimental studies that demonstrate that people apply the same social rules and expectations to computers that they apply to people. Areas to be discussed include politeness, personality, reciprocity, adaptation, gender, voice input and output, humor, and com- puter-mediated communication versus human-computer interaction.

: – : AM Invited Talk: Real-World Scheduling Applications— A Valuable Mine Field Where Search Algorithm Is Less Important Than Representation and Usability Monte Zweben, Entrepreneur-in-Residence, Institutional Venture Partners & Matrix Partners Introduction by James Crawford, i Technologies After six years of commercially developing, marketing, selling, and deploy- ing manufacturing scheduling systems, we learned that scheduling was nearly impossible to do generically. Yet companies that attempted to model in excessive detail generally failed, and those that planned more abstractly succeeded. A project was only successful if the key decision criteria was captured in the representation-an obvious point that was extraordinarily hard to execute.

: AM – : PM Invited Talk: Structured Probabilistic Models: Bayesian Networks and Beyond Daphne Koller, Stanford University Introduction by Eric Horvitz, Microsoft Corporation In recent years, Bayesian networks have had significant impact on many areas in AI, including diagnosis, planning, and learning. Koller describes this technology, and analyzes the reasons behind its success, suggesting that the use of structured model-based representations is one crucial compo- nent. These insights lead to richer probabilistic representations that can model significantly more complex domains, involving many components that interact and evolve over time. Koller argues that these representations can help us build agents that reason and act in complex uncertain environ- ments.

: – : PM Invited Talk: AI in Medicine: The Spectrum of Challenges from Managed Care to Molecular Medicine Russ B. Altman, Stanford University Introduction by Bruce G. Buchanan, University of Pittsburgh Al has embraced medical applications from its inception, and some of the earliest work in successful application of AI technology occurred in medi- cal contexts. Medicine in the twenty first century will be very different than medicine in the late twentieth century. Fortunately, the technical challenges to AI that emerge are very similar, and the prospects for success are high.

: – : PM Invited Talk: “Every Time I Fire a Linguist, My Performance Goes Up,” and Other Myths of the Statistical Natural Language Processing Revolution Julia Hirschberg, AT&T Labs — Research Introduction by Martha Pollack, University of Pittsburgh  In the past two decades, natural language processing has experienced a rev- olution, from rule-based symbolic approaches to statistical, corpus-based techniques—with remarkable success in applications such as machine translation, automatic speech recognition, and text-to-speech. But there are signs that this revolution may be finding its limits, signs this talk will ex- plore.

: – : PM Invited Panel: Science Fiction Writers Read the Futures of AI Organizer: David Miller, KISS Institute for Practical Robotics Panelists: Greg Benford, University of California, Irvine; James Hogan; and Sarah Zettel Visionary science fiction authors are the prophets of AI. Unencumbered by the burden of having to implement anything, they construct vivid images of where our work might lead—the good, bad, and ugly. They inspire and warn, challenge and scold, excite and lampoon, tickle and scare. They ask questions we need to think about. In this panel, some science fiction authors will articulate their best hopes, worst fears, and most interesting predictions about AI and its role in (future?) society. The ensuing discussion will attempt to raise our con- sciousness by discussing future issues the field of AI will need to consider

Invited Talks Invited as AI advances in its capabilities and pervasiveness.

Thursday, July 

: – : AM Invited Talk: When and Where Will AI Meet Robotics? Issues in Representation Ruzena Bajcsy, University of Pennsylvania Introduction by Reid Simmons, Carnegie Mellon University In the early days of AI, robotics was an integral part of our research effort. In the early s, all major AI laboratories had research programs in robotics. However, by the late s, robotics took its own course separate from the core activities of AI. In this presentation, Bajcsy explores the common issue that is pertinent to both AI and robotics, the issue of representation.

: – : AM Invited Talk: Experiments in Musical Intelligence David Cope, University of California, Santa Cruz Musical works contain code about the processes and influences that creat- ed them. The computer program Experiments in Musical Intelligence at- tempts to decipher this code and create new but stylistically-faithful music. Examples of output will be performed, followed by a discussion of how these principles can transfer to other media.

: AM – : PM Invited Panel: Evaluating Representations of Common Sense Organizer: Douglas B. Lenat, CYCORP Panelists: Ken Forbus, Northwestern University; Leora Morgenstern, IBM T.J. Watson Research Center; and Erik Mueller, Signiform Everyone knows that horses have heads, babies want milk, unsupported objects fall, falling eggs break, and so forth. To use such knowledge, our programs manipulate representations of them. But by what criteria should we evaluate various representations of common sense knowledge? How should we evaluate the different contradictory criteria for evaluating repre- sentations.

 IAAI Conference Innovative Applications of Artificial Intelligence All IAAI- sessions will be held in the Lecture Hall on the fourth level of the Monona Terrace Con- vention Center. Monday’s schedule is listed below. The remainder of the papers will be presented in parallel with the AAAI- technical program on Tuesday, July  and Wednesday, July . Please re- fer to the daily schedule on the following pages for times.

(D): deployed application; (E): emerging application

: – : AM Opening Remarks Bruce Buchanan, IAAI- Conference Chair

: – : AM Automated Intelligent Pilots for Combat Flight Simulation (D) Randolph M. Jones, John E. Laird, and Paul E. Nielsen A New Technique Enables Dynamic Replanning and Rescheduling of Aeromedical Evacuation (D) Alexander Kott, Victor Saks and Albert Mercer

: – : AM Coffee Break

: – : AM Intelligent Control of Life Support Systems for Space Habitats (E) Debra Schreckenghost, Daniel Ryan, Carroll Thronesbery, Peter Bonasso, and Daniel Poirot

Knowledge-Based Avoidance of Drug-Resistant HIV Mutants (D) Richard H. Lathrop, Nicholas R. Steffen, Miriam P. Raphael, Sophia Deeds- Rubin, Michael J. Pazzani, Paul J. Cimoch, Darryl M. See, and Jeremiah G. Tilles

: AM – : PM The NASD Regulation Advanced Detection Systems (ADS) (D) J. Dale Kirkland, Ted E. Senator, James J. Hayden, Tom Dybala, Henry G. Goldberg, and Ping Shyr

Countrywide Automated Property Evaluation System—CAPES (D) Ingemar A. E. Hulthage and Iain Stobie

: – : PM Lunch Break

: – : PM Success in Spades: Using AI Planning Techniques to Win the World Championship of Computer Bridge (D) Stephen J. J. Smith, Dana S. Nau, and Thomas A. Throop

Control Strategies in HTN Planning: Theory Versus Practice (E) Dana S. Nau, Stephen J.J. Smith, and Kutluhan Erol

: – : PM Producing BT’s Yellow Pages with Formation (D) Gail Anderson, Andrew Casson-du Mont, Ann Macintosh, Robert Rae, and Barry Gleeson

ANSWER: Network Monitoring Using Object-Oriented Rules (D) Gary M. Weiss, Johannes P. Ros, and Anoop Singhal

: – : PM Coffee Break

: – : PM Turbine Engine Diagnostics (TED): An Expert Diagnostic System for the M Abrams Turbine Engine (D) Richard Helfman, Ed Baur, John Dumer, Tim Hanratty, and Holly Ingham Using Artificial Intelligence Planning to Automate SAR Image Processing for Scientific Data Analysis (D) Forest Fisher, Steve Chien, Edisanter Lo, and Ronald Greeley  ⁄ : ‒ : AM : ‒ : AM : AM – : PM

Welcome and Opening Remarks Invited Panel Invited Talk Jack Mostow and Charles Rich, AAAI-98 Program Cochairs Eight Cool Things from the Collocated Learning Sparse Representations: Machine Conferences Learning, Machine Vision and the Brain Presidential Address Organizer: Charles Rich (AAAI-98 Program Tomaso Poggio, Massachusetts Institute of The Importance of Importance Cochair), MERL—A Mitsubishi Electric Research Technology David L. Waltz, NEC Research Institute Laboratory Introduction by Randall Davis Panelists: David Page (ILP), John Koza (GP), David Goldberg (SGA), Peter Bartlett (COLT), Jude Shavlik (ICML), Greg Cooper (UAI), Arthur Graesser (ST&D), and Sharon Derry (COG SCI) Madison Ballroom Madison

Integrated AI Systems: Planning and Modeling the Web Problem Solving Session Chair: Shlomo Zilberstein Session Chair: Milind Tambe What Can Knowledge Representation Do for TRIPS: An Integrated Intelligent Problem- Semi-Structured Data? Solving Assistant Diego Calvanese, Giuseppe De Giacomo and Mau- George Ferguson and James F. Allen rizio Lenzerini Integrating AI Components for a Military Modeling Web Sources for Information Planning Application Integration Marie A. Bienkowski and Louis J. Hoebel Craig A. Knoblock, Steven Minton, Jose Luis Am- bite, Naveen Ashish, Pragnesh Jay Modi, Ion Muslea, Andrew G. Philpot, and Sheila Tejada Madison Ballroom C Ballroom Madison

Learning Parallel AI / Agents and Representation Coffee Break Session Chair: Lars Asker Session Chair: Matthew Evett Iterated Phantom Induction: A Little Natural Language Multiprocessing: A Case Knowledge Can Go a Long Way Study Mark Brodie and Gerald DeJong Enrico Pontelli, Gopal Gupta, Janyce Wiebe and David Farwell SUSTAIN: A Model of Human Category AAAI‒ Learning Metacognition in Software Agents Using Bradley C. Love and Douglas L. Medin Classifier Systems Zhaohua Zhang, Stan Franklin and Dipankar Dasgupta

Madison Ballroom D Ballroom Madison &

KR for Robotics Integrated AI Systems/ Evolvable Hardware IAAI‒ Session Chair: Wolfram Burgard A Formal Methodology for Session Chair: Justinian Rosca Verifying Situated Agents BIG: A Resource-Bounded Information Phan Minh Dung Gathering Agent An Algebra for Cyclic Ordering of 2D Victor Lesser, Bryan Horling, Frank Klassner, Ani- Technical ta Raja, Thomas Wagner and Shelley XQ. Zhang Orientations  : ‒ : Amar Isli and Anthony G. Cohn Evolvable Hardware Chip for High Precision Printer Image Compression Hidenori Sakanashi, Mehrdad Salami, Masaya Iwata, Shogo Nakaya, Tsukasa Yamauchi, Takeshi

Hall of Ideas E & F of Ideas Hall Sessions Inuo, Nobuki Kajihara, and Tetsuya Higuchi

Plan Recognition Graph Plan Session Chair: Diane Litman Session Chair: Jim Blythe Needles in a Haystack: Plan Recognition in Conformant Graphplan

Tuesday, July July Tuesday, Large Spatial Domains Involving Multiple David E. Smith and Daniel S. Weld Agents Extending Graphplan to Handle Uncertainty Mark Devaney and Ashwin Ram & Sensing Actions Acquisition of Abstract Plan Descriptions Daniel S. Weld, Corin R. Anderson, and David E. for Plan Recognition Smith Mathias Bauer Hall of Ideas H & I of Ideas Hall

IAAI-98 Hybrid Knowledge Based System for Auto- matic Classificaton of B-scan Images from Ultrasonic Rail Inspection (E) J. Jarmulak, E. J. H. Kerckhoffs, and P. P. van’t Veen Expert System Technology for Nondestructive Waste Assay (E) J. C. Determan and G. K. Becker Lecture Hall Lecture

 : ‒ : PM : ‒ : PM : ‒ : PM

Invited Talk Invited Talk AAAI-98 Outstanding Paper Session To be Announced Modeling Satisfaction and Satisfactory Session Chair: Jack Mostow Modeling: Modeling Problems So Constraint Learning Evaluation Functions for Global Op- Engines Can Solve Them timization and Boolean Satisfiability Eugene C. Freuder, University of New Hampshire Justin A. Boyan and Andrew W. Moore Introduction by David L. Waltz The Interactive Museum Tour-Guide Robot Wolfram Burgard, Armin B. Cremers, Dieter Fox, Dirk Hähnel, Gerhard Lakemeyer, Dirk Schulz, Walter Steiner, and Sebastian Thrun Acceleration Methods for Numeric CSPs Yahia Lebbah and Olivier Lhomme

Integrated AI Systems: Intelligent Tutoring Graphical Probabilistic Models Intelligent Environments Session Chair: Eric Horvitz Session Chair: Peter Haddawy Session Chair: Pandurang Nayak Generating Coordinated Natural Language Structured Representation of Complex Design Principles for Intelligent Environments and 3D Animations for Complex Spatial Stochastic Systems Michael H. Coen Explanations Nir Friedman, Daphne Koller and Avi Pfeffer Stuart G. Towns, Charles B. Callaway and James Solving Very Large Weakly Coupled Markov Cooperating with People: The Intelligent C. Lester Classroom Decision Processes David Franklin Procedural Help in Andes: Generating Hints N. Meuleau, M. Hauskrecht, K. Kim, L. Peshkin, Using a Bayesian Network Student Model L. Kaelbling, T. Dean, and C. Boutilier Abigail S. Gertner, Cristina Conati and Kurt Van- Speech Recognition with Dynamic Bayesian Lehn Networks Geoffrey Zweig and Stuart Russell

Planning Reinforcement Learning KR: KB Design Lunch Break Lunch

Session Chair: Craig Boutilier Session Chair: Sven Koenig Coffee Break Session Chair: Mukesh Dalal Applying Online Search Techniques to Usability Issues in KR Systems

Improving Big Plans  Neal Lesh, Nathaniel Martin and James Allen Continuous-State Reinforcement Learning Deborah L. McGuinness and Peter Patel-Schneider

Scott Davies, Andrew Y. Ng, and Andrew Moore :3 Controlling Communication in Distributed Representing Scientific Experiments: Planning Using Irrelevance Reasoning Bayesian Q-Learning  Implications for Ontology Design and Richard Dearden, Nir Friedman and Stuart Russell Knowledge Sharing

Michael Wolverton and Marie desJardins – Natalya Fridman Noy and Carole D. Hafner OKBC: A Programmatic Foundation for 

: Knowledge Base Interoperability Vinay K. Chaudhri, Adam Farquhar, Richard  Fikes, Peter D. Karp and James P. Rice

Information Extraction I Information Extraction II Model Construction and Analysis Session Chair: Marti Hearst Session Chair: Nicholas Kushmerick Session Chair: Gautam Biswas Answering Questions for an Organization Learning to Extract Symbolic Knowledge from Multimodal Reasoning for Automatic Model Online the World Wide Web Construction Vladimir A. Kulyukin, Kristian J. Hammond and Mark Craven, Dan DiPasquo, Dayne Freitag, An- Reinhard Stolle and Elizabeth Bradley Robin D. Burke drew McCallum, Tom Mitchell, Kamal Nigam and Discovering Admissible Simultaneous Towards Text Knowledge Engineering Seán Slattery Equations of Large Scale Systems Udo Hahn and Klemens : ‒ : Schnattinger Information Extraction from HTML: Applica- Takashi Washio and Hiroshi Motoda tion of a General Machine Learning Approach Decompositional, Model-Based Learning and Dayne Freitag its Analogy to Diagnosis Brian C. Williams and William Millar — AAAI Opening Reception, Grand Terrace, Monona Terrace Convention Center Convention Terrace Monona Terrace, Grand Reception, — AAAI Opening

Fuzzy Logic Nonmonotonic Reasoning Theorem Proving PM

Session Chair: John Yen Session Chair: Leora Morgenstern Session Chair: Neelakantan Kartha  :

Logical Representation and Computation of Fixpoint 3-Valued Semantics for An Algorithm to Evaluate  – Tuesday, July 28 July Tuesday, Tuesday, July July Tuesday, Optimal Decisions in a Qualitative Setting Autoepistemic Logic Quantified Boolean Formulae Didier Dubois, Daniel Le Berre, Henri Prade, and Marc Denecker, Victor Marek and Miroslaw Marco Cadoli, Andrea Giovanardi and Marco  Régis Sabbadin Truszczynski Schaerf :  A Fuzzy Description Logic Reducing Query Answering to Satisfiability in Two Forms of Dependence in Propositional Umberto Straccia Nonmonotonic Logics Logic: Controllability and Definability Riccardo Rosati Jérôme Lang and Pierre Marquis Anytime Approximate Modal Reasoning Fabio Massacci

IAAI-98 IAAI-98 A Prototype Application of Fuzzy Logic and Hall of Champions Panel: AI Game-Playing Expert Systems in Education Assessment (E) Techniques: Are They Useful for Anything James R. Nolan Other than Games? Bayesian Network Models for Generation of Organizer: Dana Nau, University of Maryland Crisis Management Training Scenarios (E) Panelists: David Fotland, Hewlett Packard; Eugene Grois, William H. Hsu, Mikhail Voloshin Jonathan Schaeffer, University of Alberta; Gerald and David C. Wilkins Tesauro, IBM Research; and David Wilkins, SRI International

 ⁄ : ‒ : AM : ‒ : AM : ⅞ ‒ : PM

Invited Talk Invited Talk Invited Talk How People Treat Computers Like Real-World Scheduling Applications—A Structured Probabilistic Models: Bayesian Real People: Experimental Evidence of a New Valuable Mine Field Where Search Algorithm Networks and Beyond Paradigm Is Less Important Than Representation and Daphne Koller, Stanford University Clifford Nass, Stanford University Usability Monte Zweben, Entrepreneur-in-Residence, Institu- Introduction by Eric Horvitz Introduction by Howard E. Shrobe tional Venture Partners & Matrix Partners Introduction by James Crawford Madison Madison Ballroom A & B Ballroom

Agent Interaction Natural Language Generation — Natural Language Generation Argumentation Session Chair: Tuomas Sandholm Session Chair: Daniel Marcu Optimal Auctions Revisited Session Chair: James Lester Generating Inference-Rich Discourse through Dov Monderer and Moshe Tennenholtz Hermes: Supporting Argumentative Discourse Revisions of RST-Trees Minimal Social Laws in Multi-Agent Decision Making Helmut Horacek David Fitoussi and Moshe Tennenholtz Nikos Karacapilidis and Dimitris Papadias Machine Learning of Generic and User- Bayesian Reasoning in an Abductive Mecha- Focused Summarization nism for Argument Generation and Analysis Inderjeet Mani and Eric Bloedorn Ingrid Zukerman, Richard McConachy and Kevin B. Korb Madison Ballroom C Ballroom Madison Coffee Break Reinforcement Learning Learning from Sequences Time and Representation Session Chair: Tom Dietterich Session Chair: Andrea Danyluk Session Chair: Takashi Washio The Dynamics of Reinforcement Learning in Feature Generation for Sequence The Temporal Analysis of Chisholm’s Paradox Cooperative Multiagent Systems Categorization Leendert W. N. van der Torre and Yao-Hua Tan Caroline Claus and Craig Boutilier Daniel Kudenko and Haym Hirsh Temporal Reasoning with Qualitative and Tree Based Discretization for Continuous Concepts from Time Series Quantitative Information about Points and State Space Reinforcement Learning Michael T. Rosenstein and Paul R. Cohen Durations William R. B. Uther and Manuela M. Veloso Rattana Wetprasit and Abdul Sattar Madison Ballroom D Ballroom Madison

Plan Efficiency Analysis of Search Random Approaches to Search Session Chair: David Wilkins Session Chair: Brian Drabble Session Chair: Robert Schrag Inferring State Constraints for The Branching Factor of Regular Search Boosting Combinatorial Search through Domain-Independent Planning Spaces Randomization

Alfonso Gerevini and Lenhart Schubert Stefan Edelkamp : ‒ : and Richard E. Korf Carla P. Gomes, Bart Selman and Henry Kautz Analyzing External Conditions to Complexity Analysis of Admissible Heuristic Which Search Problems Are Random? Improve the Efficiency of HTN Planning Search Tad Hogg Reiko Tsuneto, James Hendler and Dana Nau Richard E. Korf and Michael Reid Hall of Ideas E & F of Ideas Hall

Constraint Satisfaction Problems Frameworks for Plan Generation Planning as Satisfiability Session Chair: Eugene Freuder Session Chair: Reid Simmons Session Chair: Subbarao Kambhampati “Squeaky Wheel” Optimization Automatic OBDD-Based Generation of Uni- Act, and the Rest Will Follow: Exploiting De-

David E. Joslin and David P. Clements July Wednesday, versal Plans in Non-Deterministic Domains terminism in Planning as Satisfiability Reversible DAC and Other Improvements for Alessandro Cimatti, Marco Roveri and Paolo Enrico Giunchiglia, Alessandro Massarotto,and Solving Max-CSP Traverso Roberto Sebastiani Javier Larrosa, Pedro Meseguer, Thomas Schiex, Hybrid Planning for Using Caching to Solve Larger Probabilistic and Gérard Verfaillie Partially Hierarchical Domains Planning Problems Subbarao Kambhampati, Amol Mali and Stephen M. Majercik and Michael L. Littman Biplav Srivastava Hall of Ideas H & I of Ideas Hall

IAAI-98 Multi Machine Scheduling: An Agent-Based Approach (D) Rama Akkiraju, Pinar Keskinocak, Sesh Murthy and Frederick Wu Realtime Constraint-Based Cinematography for Complex Interactive 3D Worlds (E) William H. Bares, Joël P. Grégoire and James C. Lester Lecture Hall Lecture

 : ‒ : PM : ‒ : PM : ‒ : PM

Invited Talk Invited Talk Invited Panel AI in Medicine: The Spectrum of Challenges “Every Time I Fire a Linguist, My Performance Science Fiction Writers Read the Futures of AI from Managed Care to Molecular Medicine Goes Up,” and Other Myths of the Statistical Organizer: David Miller, Natural Language Processing Revolution Russ B. Altman, Stanford University KISS Institute for Practical Robotics Julia Hirschberg, AT&T Labs – Research Introduction by Bruce G. Buchanan Panelists: Greg Benford, University of California, Introduction by Martha Pollack Irvine; James Hogen; and Sarah Zettel

Grammar and Language Robot Navigation Temporal Reasoning Session Chair: Eric Brill Session Chair: Leslie Kaelbling Session Chair: James Crawford A Sampling-Based Heuristic for Tree Search Position Estimation for Mobile Robots in Dy- Backtracking Algorithms for Disjunctions of Applied to Grammar Induction namic Environments Temporal Constraints Hugues Juillé and Jordan B. Pollack Dieter Fox, Wolfram Burgard, Sebastian Thrun, Kostas Stergiou and Manolis Koubarakis Ambiguity and Constraint in Mathematical and Armin B. Cremers Fast Transformation of Temporal Plans for Expression Recognition Integrating Topological and Metric Maps for Efficient Execution Erik G. Miller and Paul A. Viola Mobile Robot Navigation: A Statistical Ap- Ioannis Tsamardinos, Nicola Muscettola and Paul proach Morris Sebastian Thrun, Jens-Steffen Gutmann, Dieter Fox, Wolfram Burgard and Benjamin J. Kuipers Lunch Break Lunch Qualitative Reasoning Techniques Qualitative Modeling Coffee Break Understanding Sound Session Chair: Richard Doyle Session Chair: Dan Clancy Session Chair: Ian Horswill Qualitative Analysis of Distributed Physical An Ontology for Transitions in Physical The Role of Data Reprocessing in Complex Systems with Applications to Dynamic Systems Acoustic Environments Control Synthesis Pieter J. Mosterman, Feng Zhao, and Gautam Frank Klassner, Victor Lesser, and Hamid Nawab Christopher Bailey-Kellogg and Feng Zhao Biswas Sound Ontology for Computational Auditory Qualitative Simulation as a Temporally- A New Architecture for Automated Modelling Scene Analysis Extended Constraint Satisfaction Problem Neil Smith Tomohiro Nakatani and Hiroshi G. Okuno Daniel J. Clancy and Benjamin J. Kuipers

Uncertainty Search and Optimization Search and Limited Resources GA Applications Session Chair: Nir Friedman Session Chair: Sven Koenig Session Chair: Richard Belew Fast Probabilistic Modeling for Combinatorial A* with Bounded Costs Optimal 2D Model Matching Using a Messy Optimization Brian Logan and Natasha Alechina Genetic Algorithm  : ‒ : Shumeet Baluja and : ‒ : Scott Davies Stochastic Node Caching for J. Ross Beveridge Highest Utility First Search Across Multiple Memory-Bounded Search Learning Cooperative Lane Levels of Stochastic Design Teruhisa Miura and Toru Ishida Selection Strategies for Highways Louis Steinberg, J. Storrs Hall and David E. Moriarty and Pat Langley Brian D. Davison — AI Festival, Exhibition Hall, Monona Terrace Convention Center Convention Terrace Monona Hall, Exhibition — AI Festival, PM  : 

Constraint Satisfaction Problems — Constraint Satisfaction Problems Belief Revision and Inconsistency – Understanding Intractability

Session Chair: David Smith Session Chair: David Poole  :

Session Chair: Tad Hogg Generalizing Partial Order and Dynamic Reasoning under Inconsistency Based on  Wednesday, July July Wednesday, Wednesday, July July Wednesday, Hard Problems for CSP Algorithms Backtracking Implicitly-Specified Partial Qualitative David G. Mitchell Christian Bliek Probability Relations: A Unified Framework Supermodels and Robustness S. Benferhat, D. Dubois, J. Lang, H. Prade, A. The Constrainedness Knife-Edge Saffiotti and P. Smets Toby Walsh Matthew L. Ginsberg, Andrew J. Parkes and Amitabha Roy Belief Revision with Unreliable Observations Craig Boutilier, Nir Friedman, and Joseph Y. Halpern

IAAI-98 IAAI-98 An Expert System for Alarm System Planning Split Up: The Use of an Argument Based (E) Knowledge Representation to Meet Akira Tsurushima, Kenji Urushima, Daigo Sakata, Expectations of Different Users for Hiroyuki Date, Masatomo Nakata, Yoshinobu Discretionary Decision Making (E) Adachi and Kazuhisa Takahashi Andrew Stranieri and John Zeleznikow Warfighter’s Information Packager (E) Conversation Machines for Transaction Yigal Arens, Weixiong Zhang, Yongwon Lee, Jon Processing (E) Dukes-Schlossberg, and Marc Zev Wlodek Zadrozny, Catherine Wolf, Nanda Kamb- hatla and Yiming Ye

 ⁄ : ‒ : AM : ‒ : AM : AM – : PM

Invited Talk Invited Talk Invited Panel When and Where Will AI Meet Robotics? Experiments in Musical Intelligence Evaluating Representations of Common Sense Issues in Representation David Cope, University of California, Santa Cruz Organizer: Douglas B. Lenat, CYCORP Ruzena Bajcsy, University of Pennsylvania Panelists: Ken Forbus, Northwestern University; Introduction by Reid Simmons Leora Morgenstern, IBM T.J. Watson Research Center; and Erik Mueller, Signiform Madison Madison Ballroom A & B Ballroom

Formal Models of Agents’ Commitments Social Agents Heuristic Search Session Chair: Jonathan Gratch Session Chair: Tuomas Sandholm Session Chair: Foster Provost Leveled Commitment Contracts with Agents that Work in Harmony by Heuristic Search in Cyclic AND / OR Graphs Myopic and Strategic Agents Knowing and Fulfilling their Obligations Eric A. Hansen and Shlomo Zilberstein Martin R. Andersson and Tuomas W. Sandholm Mihai Barbuceanu Single-Agent Search in the Presence of Anytime Coalition Structure Generation with What Is Wrong With Us? Improving Deadlocks Worst Case Guarantees Robustness through Social Diagnosis Andreas Junghanns and Jonathan Schaeffer Tuomas Sandholm, Kate Larson, Martin Anders- Gal A. Kaminka and Milind Tambe Complete Anytime Beam Search son, Onn Shehory, and Fernando Tohmé Weixiong Zhang Madison Ballroom C Ballroom Madison Coffee Break Plan Execution Representation of Action Agents: Motivation and Emotion Session Chair: Jeff Rickel Session Chair: Neelakantan Kartha Session Chair: Pete Bonasso Managing Multiple Tasks in Complex, An Action Language Based on Causal A Motivational System for Regulating Dynamic Environments Explanation: Preliminary Report Human-Robot Interaction Michael Freed Enrico Giunchiglia and Vladimir Lifschitz Cynthia Breazeal (Ferrell) Maintaining Consistency in Hierarchical Abductive Planning with Sensing Emotion Model for Life-Like Agent and Its Reasoning Matthew Stone Evaluation Robert E. Wray, III and John Laird Hirohide Ushida, Yuji Hirayama and Hiroshi Nakajima When Robots Weep: Emotional Memories and Decision-Making Juan D. Velásquez Madison Ballroom D Ballroom Madison

Game Playing Constraint Satisfaction Problems Design and Diagnosis Session Chair: Susan Epstein Session Chair: David Etherington Session Chair: Ethan Scarl Opponent Modeling in Poker Using Arc Weights to Improve Iterative Repair Toward Design as Collaboration Darse Billings, Denis Papp, Jonathan Schaeffer John Thornton and Abdul Sattar Susan L. Epstein and Duane Szafron Extending GENET to Solve Fuzzy Constraint An Architecture for Exploring Large Design Finding Optimal Strategies for Imperfect Satisfaction Problems Spaces Information Games Jason H. Y. Wong and : ‒ : Ho-fung Leung John R. Josephson, B. Chandrasekaran, Mark Car- Ian Frank, David Basin, and Hitoshi Matsubara roll, Naresh Iyer, Bryon Wasacz, Giorgio Rizzoni, Qingyuam Li, and David A. Erb Constructing the Correct Diagnosis When Symptoms Disappear

Hall of Ideas E & F of Ideas Hall Nancy E. Reed

Constraint Satisfaction Problems — Search Control in Theorem Proving Constraint Satisfaction Problems Local Search Session Chair: Eric Horvitz Session Chair: Stephen Smith Session Chair: Russell Greiner A Feature-Based Learning Method for Branch and Bound Algorithm Selection by Local Search for Statistical Counting Theorem Proving Performance Prediction Olivier Bailleux July Thursday, Matthias Fuchs Lionel Lobjois and Michel Lemaître A Tractable Walsh Analysis of SAT and Learning Investment Functions for An Integer Local Search Method with Appli- Its Implications for Genetic Algorithms Controlling the Utility of Control Knowledge cation to Capacitated Production Planning Soraya Rana, Robert B. Heckendorn and Darrell Oleg Ledeniov and Shaul Markovitch Joachim P. Walser, Ramesh Iyer and Narayan Whitley Venkatasubramanyan Hall of Ideas H & I of Ideas Hall

Concepts and Context Nonmonotonic Reasoning Learning in Natural Language Session Chair: Robert McCartney Session Chair: W. Lewis Johnson Session Chair: Andy Kehler Knowledge Intensive Exception Spaces Experimenting with Power Default Reasoning Learning to Classify Text from Labeled and Sarabjot S. Anand, David W. Patterson and John Eric Klavins, William C. Rounds, and Guo-Qiang Unlabeled Documents G. Hughes Zhang Kamal Nigam, Andrew McCallum, Sebastian Probabilistic Frame-Based Systems Thrun and Tom Mitchell Daphne Koller and Avi Pfeffer Knowledge Lean Word—Sense Disambiguation Ted Pedersen and Rebecca Bruce

Lecture Hall Lecture Learning to Resolve Natural Language Ambiguities: A Unified Approach Dan Roth  : ‒ : PM : ‒ : PM

Closing Remarks Jack Mostow and Charles Rich, AAAI-98 Program Cochairs

Inductive Learning Session Chair: Mehran Sahami Boosting in the Limit: Maximizing the Margin of Learned Ensembles Adam J. Grove and Dale Schuurmans Boosting Classifiers Regionally Richard Maclin Robust Classification Systems for Imprecise Environments Foster Provost and Tom Fawcett

Lunch Break Lunch Tractable Inference Session Chair: Peter F. Patel-Schneider Algorithms for Propositional KB Approximation Yacine Boufkhad A Non-Deterministic Semantics for Tractable Inference James M. Crawford and David W. Etherington Computing Intersections of Horn Theories for Reasoning with Models Thomas Eiter, Toshihide Ibaraki, and Kazuhisa Makino

Human-Robot Interaction

AAAI Anual Business Meeting AAAI Anual Business Session Chair: Ken Forbus Alternative Essences of Intelligence R. Brooks, C. Breazeal, R. Irie, C. Kemp, M. Mar- janovic, B. Scassellati and M. Williamson Eye Finding Via Face Detection for a Foveated

 : ‒ : Active Vision System Brian Scassellati Template-Based Recognition of Pose and Motion Gestures on a Mobile Robot

: ‒ : S. Waldherr, S. Thrun, R. Romero and D. Margaritis

Constraint Satisfaction Problems Session Chair: Thomas Schiex On the Conversion between Non-Binary and Binary Constraint Satisfaction Problems

Thursday, July July Thursday, Fahiem Bacchus and Peter van Beek On the Computation of Local Interchangeability in Discrete Constraint Satisfaction Problems Berthe Y. Choueiry and Guevara Noubir A Fast Algorithm for the Bound Consistency of Alldiff Constraints Jean-Francois Puget

Learning about People Session Chair: Michael Wolverton Recommendation as Classification: Using Social and Content-Based Information in Recommendation Chumki Basu, Haym Hirsh, and William Cohen Learning to Predict User Operations for Adap- tive Scheduling Melinda T. Gervasio, Wayne Iba and Pat Langley Adaptive Web Sites: Automatically Synthesiz- ing Web Pages Mike Perkowitz and Oren Etzioni

 Booth # Exhibition ActivMedia Robotics The exhibition will be held in the Exhibit Hall  Hancock Road  on the first level of the Monona Terrace Conven- Peterborough, NH Tel: () - or () - tion Center, Tuesday, July  through Thursday, Email: [email protected] July . Admittance is restricted to badged con- Web: www.activmedia.com/robots ference attendees. Vendor-issued guest passes Pioneer Robots will be plentiful at AAAI-. must be redeemed at the Exhibitor Registration Visit ActivMedia Robotics to learn more about Counter, at Lakeside Commons in the foyer of the craftsmanship, accessories, software, docs the exhibit hall, on the first level of the Monona and support that make these platforms such a Terrace Convention Center. Further information popular base for AI research. Come see the Pio- regarding access to the Exhibition can be ob- neer AT climb mountains! See the classy new tained from the Exhibitor Registration Desk. PTZ Robotic Camera in action—plus a sneak Exhibit Hours preview of what’s in the pipeline for Pioneer this fall! Oh, yes, and there will be more of the ever-    AM   PM Tuesday, July : – : popular Pioneer T-shirts. Wednesday, July  : PM – : PM : PM – : PM Booth # Thursday, July  : AM – : PM Brightware, Inc.  Ignacio Blvd. Novato, CA  Exhibitors Tel: () - Fax: () - ■ AAAI Press Web: www.brightware.com ■ ACM Brightware is the market leader in automat- ■ ActivMedia Robotics ed customer interaction on the Net. Bright- ■ Brightware, Inc. ware’s application products intelligently interact Exhibit Program Exhibit ■ Elsevier Science with on-line customers, giving them an auto- ■ Franz, Inc. mated experience on the Net that rivals the ex- ■ Harlequin Inc. perience they would have on the phone or in ■ IEEE Computer Society person with company marketing, sales, and ser- ■ Kluwer Academic Publishers vice employees. Brightware delivers unprece- ■ The MIT Press dented ROI through fully automated customer ■ MacroVu, Inc. interaction, opens the Net as a new source of ■ Morgan Kaufmann Publishers revenues and market share for companies that ■ NASA Ames Research Center today sell through people, and enhances cus- ■ Numan Intelligence tomer loyalty through a helpful and rewarding ■ PC AI Magazine on-line experience. Brightware is targeting For- ■ Prentice Hall tune  companies who are using the Net as ■ Real World Interface, Inc. their newest customer channel, and the new ■ Register Machine Learning Technologies, Inc. generation of enterprises whose businesses are ■ Springer-Verlag based on the Net. For more information, please ■ Stottler Henke visit Brightware’s Web site.

Booth # Booth # AAAI Press Elsevier Science  Burgess Drive  Avenue of the Americas Menlo Park, CA  New York, NY  Tel: () - Tel: () - Fax: () - Fax: () - Email: [email protected] Visit the Elsevier Science booth for an online demonstration of the journal Artificial Intelli- Booth # gence’s electronic features—including the new ACM mega index—and see how easily you can access  Broadway-th Floor the latest research results in the field. Of course, New York, NY  you can also browse through a wide range of our Tel: () - books and journals on display. Collect free sam- Fax: () - ple copies of the journals of your choice or take  advantage of the special % discount on all IEEE Computer Society, one of the most Exhibitors book titles! Our staff will be happy to assist you prestigious professional associations in the and we look forward to meeting you there. world, serves its members through numerous publications, conferences, and workshops. Booth # Membership information and complimentary Franz, Inc. copies of Computer and IEEE Intelligent Systems  University Avenue (formerly IEEE Expert) magazines will be avail- Berkeley, CA  able. Some of our latest book releases, including Tel: () - The Pattern Recognition Basis of Artificial Intelli- Fax: () - gence, by Donald Tveter; Mathematical Methods Email: [email protected] in Artificial Intelligence, by Edward Bender; and Web: www.franz.com/ Franz Inc. offers the most powerful dynamic Stiquito: Advanced Experiments with a Simple object-oriented programming system available for and Inexpensive Robot, by James M. Conrad and enterprise-wide, complex application develop- Jonathan W. Mills (book includes robot kit) are ment on Unix and Windows platforms. Allegro on sale. We will have live demonstrations of CL’s Dynamic Objects technology provides the Stiquito each day. power to develop complex, mission-critical appli- Booth # cations and the flexibility to handle user-modifi- cations easily, even after deployment. Allegro CL Kluwer Academic Publishers is powered by CLOS, which supports full multiple  Philip Drive inheritance; first-class exceptions; dynamic class, Norwell, MA  Tel: () - object and method redefinition; interpretive se- Fax: () - mantics; dynamic recompilation; high-perfor- Email: [email protected] mance garbage-collection; and the Meta-Object On-line Catalog: www.wkap.nl Protocol. Allegro CL also provides connectivity to Kluwer Academic Publishers invites you to the ORBIX family of products, simplifying the pro- visit our display of the premier journals in the cess of creating large-scale CORBA/ORBIX applica- Artificial Intelligence area. We are proud to an- tions in situations where run-time customizabili- nounce the inaugural issue of our new journal ty is desirable. Software developers in Fortune Autonomous Agents and Multi-agent Systems edit-  companies worldwide are using Allegro CL ed by Nicholas Jennings, Katia Sycara, and for mission-critical applications in every domain. Michael Georgeff. Free sample copies of all journals are available. Booth #  Harlequin Inc. Booth # One Cambridge Center MacroVu, Inc. Cambridge, MA   High School Road Box  Tel: () - Bainbridge Island, WA  Fax: () - Tel: () - Email: [email protected] Fax: () - Harlequin will be demonstrating our complete Can Computers Think? The Issue Map Series: line of advanced software development tools. Our Seven large, full-color argumentation maps visu- Common Lisp solutions are unsurpassed, includ- ally present the history of the -year machine in- ing LispWorks for the Windows operating sys- telligence debate. The maps summarize each con- tem, LispWorks for workstations, and Liquid tribution (more than  major moves from  Common Lisp (formerly Lucid Common Lisp). AI research scientists and philosophers world- New CORBA interfaces support Lisp develop- wide); rebuttals and counter rebuttals are linked ment of components for distributed environ- in threads of dispute. These maps can save stu- ments. Harlequin will also be showing Harlequin dents entering the field hundreds of hours trying Dylan, the new dynamic object-oriented language to comprehend the overall context and history of for the Windows Platform, as well as MLWorks, the debate’s  issue subtopics and the several the commercial implementation of Standard ML contending “camps” of protagonists. The maps for UNIX and Windows. provide the current frontiers of the arguments and a chronological intellectual history. Project Booth # Director Robert Horn, a visiting scholar at Stan- IEEE Computer Society ford University, will present at the booth.  Los Vaqueros Circle Los Alamitos, CA  Tel: () -  Booth # develop intelligent, self-monitoring, and adap- The MIT Press tive systems to enhance aerospace safety and ef- Five Cambridge Center ficiency. These information technologies will Cambridge, MA  provide a catalyst for a new generation of em- Tel: () - bedded systems that promise profound social Fax: () - and economic impact. NASA has been a long- Email: [email protected] Web: mitpress.mit.edu term supporter of the AAAI and is pleased to ex- Publisher of academic books and journals in hibit some of our latest AI developments and artificial intelligence and computer science. demonstrations at this year’s conference. Stop by our booth to see Behavior-Based Booth #  Robotics by Ronald C. Arkin, Reinforcement Learning by Richard S. Sutton and Andrew G. Numan Intelligence, Inc. Barto and other new publications from MIT  Cobblestone Drive Press & AAAI Press. Troy, MI  Tel: () - Fax: () - Booth # Email: [email protected] Exhibitors Morgan Kaufmann Publishers Web: www.numan.com  Pine Street, Sixth Floor Numan Intelligence, Inc. will unveil a revolu- San Francisco, CA  tionary breakthrough in artificial intelligence at Tel: () - or () - AAAI-: NuIntelligence and Human-Computer Email: [email protected] Intelligence. NuIntelligence is an embodiment of Web: www.mkp.com the elementary operational components of heuris- Morgan Kaufmann publishes the finest infor- tic search, genetic algorithms, neural networks, mation resources for artificial intelligence re- optimization, and other artificial intelligence searchers and students, including graduate and techniques. NuIntelligence can be used to imple- undergraduate texts, monographs, collected vol- ment any of these techniques as well as hybridiza- umes, and conference proceedings. We believe tion’s and combinations of them. NuIntelligence’s strongly in seeking out the most authoritative, ex- companion problem solving methodology, Hu- pert authors. Our family of authors and series ed- man-Computer Intelligence, enables the reduc- itors includes many of the world’s most respected tion of problem complexity from exponential to computer scientists and engineers and their books linear. NuIntelligence seamlessly integrates with a often represent the wisdom gained from years of broad base of applications across many problem research, development, and teaching. Recently domains through DDE, DDE-OLE and Internet published in this area are Banzhaf, et al: Genetic Socket interfaces. Programming—An Introduction, Nils J. Nilsson: Artificial Intelligence—A New Synthesis, and Booth # Michael N. Huhns and Munindar P. Singh: Read- PC AI Magazine ings in Agents. Please visit us at our booth. Post Office Box  Phoenix, AZ  Booth # Tel: () - NASA Ames Research Center Fax: () - Contact: Michael Goldman Email: [email protected] Caelum Research Corporation Web: www.pcai.com/pcai/ NASA/Ames Research Center PC AI Magazine provides the information Mail Stop - necessary to help managers, programmers, execu- Moffett Field, CA  tives, and other professionals understand the Tel: () - quickly unfolding realm of artificial intelligence NASA’s bold missions in space exploration (AI) and intelligent applications (IA). PC AI ad- and aeronautics depend on world-class research dresses the entire range of personal computers in- in computer science and artificial intelligence. cluding the Mac, IBM, PC, NeXT, Apollo, and Toward this end, Ames Research Center, locat- more. PC AI is an application-oriented magazine ed in the heart of Silicon Valley, has been desig- designed to give readers useful “hands-on” infor- nated the NASA Center of Excellence in Infor- mation. PC AI features developments in expert mation Technology. NASA is seeking ways to systems, neural networks, object oriented devel- put an unprecedented level of intelligence into opment, and all other areas of artificial intelli- the vehicles we send out to explore the universe gence. Feature articles, product reviews, real- for us, to expand human capabilities through re- world application stories, and a Buyer’s Guide search in human-centered computing and to present a wide range of topics in each issue.  Booth # Learning Technology™. It will also demon- Exhibitors Prentice Hall strate Discipulus™ for Windows //NT. One Lake Street Discipulus™ is the first commercial product Upper Saddle River, NJ  based on AIM Learning Technology™. Discip- Tel: () - ulus™ is very fast program induction software, Fax: () - ideally suited for tasks such as data mining, Prentice Hall is proud to present informa- forecasting, function fitting, and classification. tion about the forthcoming second edition of the leading introductory AI textbook, Artificial Booth # Intelligence: A Modern Approach by Stuart Rus- Springer-Verlag sell and Peter Norvig.  Fifth Avenue Please stop by our booth to obtain informa- New York, NY  tion on this and other quality books, such as; Tel: --SPRINGER The Widely Used Common Lisp by Paul Graham, Fax: () - and books on a variety of subjects ranging from Email: [email protected] Data Mining to Java. We are also accepting pro- Web: www.springer-ny.com posals for the Prentice Hall AI (and related sub- Springer-Verlag is an international publisher of jects) book series, Stuart Russell and Peter books, journals, software, and the renowned se- Norvig, Series Editors. ries “Lecture Notes in Artificial Intelligence.” Stop by the booth for a special % AAAI dis- Booth # count on all AI titles, plus related books on fuzzy Real World Interface, Inc. logic, neural networks, evolutionary computing, and a wide selection of general interest topics.  Fitzgerald Drive Post Office Box  Featured titles include: Jennings and Jaffrey, NH  Wooldridge’s Agent Technology, Adami’s Introduc- Tel: () - tion to Artificial Life, Munakata’s Fundamentals of Fax: () - the New Artificial Intelligence, Michalewicz’s Ge- Web: www.rwii.com netic Algorithms + Data Structures = Evolution Pro- Real World Interface, Inc. (RWII), an estab- grams, Schaeffer’s One Jump Ahead: Challenging lished leader in indoor mobile research robots, Human Supremacy in Checkers, and Grillmeyer’s is pleased to announce their second All Terrain Exploring Computer Science with Scheme. Robot Vehicle—The ATRV-. The ATRV- is designed to provide the robotics researchers and Booth # scientists who are pioneering tomorrow’s rovers Stottler Henke Associates, Inc. (SHAI) with a vehicle capable of supporting demanding  South Amphlett Blvd., Suite  missions including security, de-mining, recon- San Mateo, CA  naissance, surveillance and hazmat. RWII also Tel: () - provides application specific mobile robot de- Fax: () - velopment services for industrial, entertain- Email: [email protected] ment, and research customers. For further in- Web: www.shai.com formation, visit the RWII at booth # or view SHAI is looking for highly motivated indi- our family of indoor and all terrain mobile viduals who want interesting, challenging work robots at: http://www.rwii.com. on a variety of AI research and development projects. We currently have job openings for Booth # programmers/software engineers and artificial Register Machine Learning intelligence programmers/software engineers. SHAI has been a leader in AI research and in- Technologies, Inc. telligent solutions development since our incep-  Grand Avenue tion in . We are results-oriented problem Oakland, CA  Tel: () - solvers with practical experience gained in over AIM Learning Technologies™ is a suite of  AI projects for commercial and government tools to create extremely fast and memory effi- clients. We have developed and fielded hun- cient learning systems. AIM technology is ideal- dreds of operational systems in daily use in do- ly suited for large induction and optimization mains as varied as space station planning and tasks and for embedded applications. At AAAI- scheduling, intelligent tutoring systems, and re- , Register Machine Learning Technologies, tail sales prediction. Inc.™ will announce the availability of custom learning applications and tools based on AIM

 ■ : AM: CAPES: Countrywide Automated AAAI- Intelligent Property Evaluation System ■ : AM: Virtual Mattie Activity Monitor Systems Demonstrations ■ : PM: Sensible Agents Operating under The Intelligent Systems Demonstrations will be Dynamic Adaptive Autonomy held in the Exhibit Hall on the first level of the ■ : PM: FindUR: A Web-based Environ- Monona Terrace Convention Center, and will ment for Conceptual Retrieval be open to registered conference attendees dur- ■ : PM: Interactive Characters with Tactile ing exhibit hours. The Intelligent Systems Interface Demonstrations showcase state of the art AI im- plementations. System builders will be on hand to present their work, and audience interaction Answering Questions for an will be encouraged where possible. Organization Online Vladimir A. Kulyukin, Kristian J. Hammond, and Robin D. Burke, Intelligent Information Laboratory, Demonstrations Schedule University of Chicago Tuesday, July  The World Wide Web continues to challenge organizations to make online access to their ex- ■ : AM: Distributed Coaching for an Intel- pertise convenient for their clients. One means of ligent Learning Environment expertise access that many clients find convenient ■ : AM: Cyclepad: An Articulate Virtual in everyday life is asking natural language ques- Laboratory for Engineering Thermodynamics tions of the organization. To support it online, ■ : AM: PowerConstructor: A Belief Net- we developed an approach to building organiza- work Learning Tool tion-embedded question-answering intermedi- ■ : PM: TRIPS: The Rochester Interactive aries, called “information exchange systems.” Planning System These systems use their knowledge of the organi- ■ : PM: Interactive Pet Robot with Emotion zation’s structure to answer the clients’ questions Model and to acquire new expertise from the organiza- ■ : PM: ARIADNE: A System for Integrat- tion’s experts. Our approach uses techniques of ing Information from the Web hierarchical and predictive indexing, combined ■ : PM: KANSEI: Image Retrieval Simulat- term weighting, abstraction-based retrieval, and ing the Human Preference negative evidence acquisition. We will demon- ■ : PM: The Intelligent Classroom strate these techniques with the Chicago Infor- ■ : PM: Answering Questions for an Orga- mation Exchange system, an information ex- nization Online change application embedded in the University ■ : PM: CiteSeer: Autonomous Citation In- of Chicago’s computer science department. dexing ■ : PM: Realtime Gesture-Speech Human Interface on Notebook Size Personal Com- puter ARIADNE: A System for Integrating Information from the Web Wednesday, July  Craig A. Knoblock, Steven Minton, Jose Luis Ambite, ■ : PM: “Squeaky Wheel” Optimization Naveen Ashish, Greg Barish, Pragnesh Jay Modi, Ion Demonstration Muslea, Andrew G. Philpot, and Sheila Tejada, Infor- ■ : PM: TacAir-Soar: Generating Au- mation Sciences Institute, Integrated Media Systems tonomous Behaviour for a Distributed Mili- Center, and Department of Computer Science, Univer- sity of Southern California tary Training Environment The Web is based on a browsing paradigm ■ : PM : A Description Logic-based Config- that makes it difficult to retrieve and integrate urator for the Web data from multiple sites. Today, the only way to ■ : PM: STEVE: A Pedagogical Agent for do this is to build specialized applications, Virtual Reality which are time-consuming to develop and diffi- ■ : PM: Self-Explanatory Simulators for Ed- cult to maintain. We are addressing this prob- ucation lem by creating the technology and tools for ■ : PM – : PM: AI Festival: All demos rapidly constructing information agents that ex- available tract, query, and integrate data from web Thursday, July  sources. We will demonstrate a system called ■ : AM: Presenting Web Site Search Results Ariadne for rapidly building agents to integrate

Intelligetn Systems Demonstrations Systems Intelligetn in Context Web sources. Our system makes it fast and easy  to build new information agents that access ex- signed to support a range of uses, from an in- Intelligent Systems Demonstrations isting Web sources. In this demo, we will show teractive appraiser’s assistant in which its inter- how we can query and integrate data from mul- nal operations can be controlled by an expert tiple web sources in several different domains. user to a fully automatic mode suitable for less Also, we will show the kind of data models knowledgeable users or batch runs. It has sever- built, examples of information gathering plans, al models, including heuristics derived from how wrappers are generated for individual Web company-specific business rules, and uses both sources, and how selected information is stored commercial and proprietary property databases. locally to improve performance.

CyclePad, an Articulate Virtual CiteSeer: Autonomous Laboratory for Engineering Citation Indexing Thermodynamics Steve Lawrence, C. Lee Giles, Kurt D. Bollacker, Kenneth D. Forbus, Leo Ureel, Julie Baher, and Sven NEC Research Institute E. Kuehne, Institute for the Learning Sciences, North- This demo presents CiteSeer, an autonomous western University; John Everett, Xerox Palo Alto Re- citation indexing system. CiteSeer autonomous- search Center, and Mike Brokowski, Department of ly locates and processes research articles on the Mechanical Engineering, Northwestern University Web in PostScript form. CiteSeer automatically CyclePad is an articulate virtual laboratory extracts information from the articles including (AVL) for learning engineering thermodynamics the header, abstract, individual citations to other by design. Design tasks are highly motivating, and papers, and the context of the citations. CiteSeer tie classroom learning to real-world concerns. Stu- organizes the literature, and allows the location dents using CyclePad can design power plants, re- of papers using keyword search, citation links, frigerators, engines, cryogenic systems, and other and citation co-occurrence. Citations to any giv- types of thermodynamic cycles. Currently Cy- en paper can be made in many different formats, clePad is used by over  students per year, and is and CiteSeer uses AI methods in order to cluster available for download via the web. In this identical citations. This allows CiteSeer to rank demonstration you will see how students create the cited articles according to the number of ci- designs, including helping work with modeling tations. CiteSeer can also group together and assumptions and make appropriate choices of pa- display the context of multiple citations to a giv- rameter values. You will also get a look under the en paper. The context of citations can be very hood at how CyclePad’s analysis and explanation useful for literature search and evaluation, e.g. systems work, the underlying knowledge base, subsequent articles may review a given article, and some of the subtleties involved in reasoning highlight limitations, or present follow up work. about thermodynamics.

Countrywide Automated Property A Description Logic-Based Evaluation System—CAPES Configurator for the Web Ingemar A.E. Hulthage and Iain Stobie, Artificial In- Deborah L. McGuinness and Lori Alperin Resnick, telligence Division, Countrywide Home Loans AT&T Labs – Research; Charles Isbell, MIT; Matt Parker and Chris Welty, Vassar College; and Peter Pa- The purpose of CAPES is to estimate the tel-Schneider, Bell Labs Research. Corresponding Au- market value of residential properties in order to thor: Deborah McGuinness, AT&T Labs – Research assess the collateral on Countrywide loans. Description logics have a history of success CAPES estimates market value by comparison in configuration applications in major compa- of the subject property to other similar nearby nies including AT&T, Lucent, and the Ford properties, for which recent sales information is Motor Company. While our platform has pro- available. Characteristics of the subject and duced over  deployed commercial configura- comparable properties, such as the living area tors, we find a demonstration application to be and number of bedrooms, are used to the extent the best expository tool for describing how de- available. In some cases price indices describing scription logics can be best leveraged in tasks the change in property values over time are also such as configuration. We have developed a used. In addition to the estimated market value, demonstration system that was designed to con- CAPES produces a measure of the uncertainty tain reasoning processes analogous to those in in the result. Its accuracy has been validated ex- our deployed systems, yet works in the everyday tensively on batches of properties by comparing domain of configuring home stereo systems. its results to known sales prices. CAPES was de- The system is built using the CLASSIC knowledge  representation system and has recently been sites. Usage logs from the last  months and us- ported to a multi-user web platform. Our er studies show improved recall with little nega- demonstration contains, among other things, tive impact on precision. The knowledge base of examples of configurations from partial specifi- background information is also used to support cations, contradiction detection, explanation of query formation that is semantically richer than reasoning, automatic completion, alternative what is typically available from web search en- exploration, parts list examination, and config- gines. The demonstration systems show concep- uration from an example library. tual search in sites covering electronic yellow pages, community calendars, and competitive intelligence applications. Our system also in- Distributed Coaching for an cludes a distributed description logic-supported Intelligent Learning Environment environment for generating and maintaining background knowledge ontologies. Kenneth D. Forbus, Leo Ureel, Julie Baher and, Sven E. Kuehne, Institute for the Learning Sciences, North- western University; John Everett, Xerox Palo Alto Re- search Center, and Mike Brokowski, Department of The Intelligent Classroom Mechanical Engineering, Northwestern University David Franklin and Joshua Flachsbart, University of We are demonstrating a distributed coaching Chicago system for CyclePad, a deployed intelligent People frequently complain that it is too dif- learning environment in engineering thermody- ficult to figure out how to get computers to do namics. Part of the coach resides on the stu- what they want. However, with a computer sys- dent’s computer, with the rest on a server ac- tem that actually tries to understand what its cessed via email. The on-board coach uses users are doing, people can interact in ways that Bayesian reasoning about teleology to make are more natural to them. We have been devel- suggestions about parameter values, and helps oping a system, the Intelligent Classroom, that students debug contradictions. The email coach does exactly this. The Intelligent Classroom us- provides additional analysis help and uses anal- es cameras and microphones to sense a speaker’s ogy for design coaching, providing step-by-step actions and then infers his intentions from advice on how principles in a web-based library those actions. Finally, it uses these intentions to can be applied to a student’s particular design. decide what to do to best cooperate with the We demonstrate how the on-board coaching speaker. (The Classroom can play videotapes, works and how the RoboTA agent colony han- show slides, and also produce a video of the pre- dles email interactions. We show how sentation.) In the Intelligent Classroom, the MAC/FAC retrieves cases from a design library speaker need not worry about how to operate and how SME is used to generate concrete ad- the Classroom; he may simply go about his lec- vice about how to apply the principles of a case ture and trust the Classroom to assist him at the to the student’s design problem. Our case com- appropriate moments. piler, which takes expert-prepared materials and produces cases, will also be shown. Interactive Characters with FindUR: A Web-Based Environment Tactile Interface Hirohide Ushida, Yuji Hirayama, and Hiroshi Naka- for Conceptual Retrieval jima Deborah L. McGuinness, and Lori Alperin Resnick, This demonstration shows interactive charac- AT&T Labs – Research; Thomas W. Beattie and Steve ters which communicate with their users. The Solomon, AT&T Labs; and Harley Manning, Dow Jones Markets. Corresponding Author: Deborah characters are able to express emotions and per- McGuinness, AT&T Labs – Research sonalities. A behavior generation model is used to When documents contain few conceptually generate life-like behaviors of the characters. The related words, recall for naturally occurring model consists of reactive and deliberative mech- queries can drop below acceptable levels. We an- anisms. In the reactive mechanism, tactile sensors alyzed a number of web sites and found that up are used to realize physical interactions with the to  percent of the relevant retrievals were users. The characters can feel the user’s touch and missed. We devised and implemented a system react instantaneously. On the other hand, the de- architecture that improves search performance liberative mechanism generates goal-oriented be- using query expansion from description logic- havior as the result of interactions between cog- maintained ontologies. Our work has been de- nitive and emotional processes. This mechanism ployed on ten corporate and community web is based on a psychological theory, called the cog- Intelligent Systems Demonstrations Systems Intelligent  nitive appraisal theory. The model also has a provided. If a user picks up one of those as a key Intelligent Systems Demonstrations learning mechanism to improve behavioral pat- image, similar images in the database can be re- terns. The behaviors are represented by comput- trieved simulating the human preferences. er animation with sound. The results obtained from experiments showed that the model is ef- fective to give the users illusion of life. PowerConstructor: A Belief Network Learning Tool Jie Cheng, University of Ulster at Jordanstown Interactive Pet Robot PowerConstructor is an efficient and handy with Emotion Model belief network learning tool, which includes a Toshihro Tashima, Toshimi Kudo, Sachihiro Saito wizard interface and a construction engine. Our and Masaharu Osumi, Fuzzy Technology and Business system is currently available on -bit Windows Promotion Division, OMRON Corporation platforms (Windows , , NT). It takes a We propose a Pet Robot that interacts with database table as input and constructs the belief users and exhibits lifelike behavior based on its network structure as output. Our system has a emotion model. The Pet Robot can percept the number of main features. User-friendly interface: It stimuli from users or environment by using gathers input information through  simple steps some tactile and auditory sensors. The Emotion and there is online help for each step. Accessibili- Model generates emotions and desires by the ty: It supports most of the popular desktop stimuli. It consists of a reactive mechanism, database and spreadsheet formats. It also supports which is based on subsumption architecture, remote database servers through ODBC. and a deliberative mechanism, which is based Reusability: The construction engine is an inde- on a psychological theory. The Pet Robot be- pendent ActiveX code component so that it can haves reactively and emotionally and its behav- be easily integrated into other belief network, da- iors always changes as the emotion and desires ta mining, knowledge base, and decision support change. The Pet Robot wears a cat costume and systems for Windows. Efficiency: The system is behaves like a cat. For example, she wags her tail based on our information theoretic dependency delightedly if she is stroked, and she turns or analysis algorithm, which requires conditional in- looks around if she hears a big sound. She gets dependence test O(N) times for the general case sleepy when she feels satisfied, and she meows and O(N^) times when the node ordering is when she feels lonely. This robot is designed to known. Other kinds of domain knowledge can al- supply users with peace and enrichment of so be used in the learning process. PowerCon- mind, and we will evaluate it. structor . is now available for download from http://infosys. susqu.edu/bnpc/. KANSEI Image Retrieval Simulating the Human Preference Presenting Web Site Hideyuki Kobayashi, Yoriyuki Okouchi, and Shunji Search Results in Context Ota, Information Technology Research Center, OM- RON Corporation Michael Chen and Marti Hearst, Computer Science Department and School of Information Management & We have developed an image retrieval system Systems, University of California, Berkeley using KANSEI (feeling, impression or sensibility) We address search over large, heterogeneous features. This system can search the same sensu- web sites such as those found at universities and ous image from a large image storage not using within corporate intranets. The goal is to make text or word but an image. Therefore it doesn’t use of structure implicit within the site to pro- need indexing on each image for preparing im- vide context for the retrieved documents, even age retrieval. Our system extracts the KANSEI for those sites for which there is no centralized features from each image, and sets adequate organization. Most web search engines simply weights for combining those features. In com- list titles, urls, and abstracts, and thus do not bining procedure, we introduce a new value place the results in context. We will demonstrate called “adaptability.” It judges how much the our alternative: a simple but novel approach to features are extracted from the image. As a re- organizing and presenting the results of search sult, adaptability makes it possible to make a over the pages of a large, heterogeneous web site. KANSEI model depending on each image and to The main idea is to show, for each page match- calculate similarity between images. As color ing the query, the path of web links that a user image contents of our current system, Japanese would follow from a root page to the search hit. stamp, national flag and butterfly images are The result is a hierarchical characterization of  the search results that both shows the context in and customizing self-explanatory simulators. which the hits appear and educates the user Going under the hood, we will show how the about the structure of the web site. runtime architecture works, focusing on the structured explanation system, and the structure and organization of domain theories that fuel Realtime Gesture-Speech Human the simulation compiler. Interface on Notebook Size Personal Computer Sensible Agents Operating Under Ryuichi Oka, Hironobu Takahashi, Toshiro Mukai, Takuichi Nishimura, Takashi Endou, Masayuki Dynamic Adaptive Autonomy Nakazawa, Shigeki Nagaya, and Hiroshi Matumura, K. Suzanne Barber, The University of Texas-Austin Real World Computing Partnership The practical deployment of distributed agent- Four application programs are demonstrated based systems mandates that each agent behave using so-called multi-modal personal computer sensibly, incorporating an understanding of both of notebook size (MMPC) with a microphone global system goals and their own local goals. The and a CCD camera. The programs are as fol- Sensible Agent research seeks to prove: The oper- lows: () House design based on realtime spot- ational level of agent autonomy (i.e. types of roles ting recognition of spontaneous speech and ges- an agent plays in its interactions with other ture. A user can talk with an agent and share the agents) is key to an agent’s ability to respond to present status of the task displayed by CG. () dynamic situational context, (i.e. the states, Speech summary based on automatic segmenta- events, and goals that exist in a multi-agent sys- tion of topic from spontaneous speech. The tem), conflicting goals, and constraints on behav- output of the summary is a sequence of speech ior. Levels of autonomy are defined along a spec- segments. The program is language free. () Mu- trum ranging from command-driven (agent exe- tual retrieval between speech and video image of cutes commands from another agent), to consen- TV news data based on self-organized databases sus (agents work together to meet goals), to local- and spotting retrieval. The query is an endless ly autonomous (agent can initiate its own thread stream of speech or video image. () Flexible re- of execution), to master (agent controls other altime gesture recognition based on a new spot- agents). The Sensible Agent architecture and ca- ting matching method. About  categories of pabilities for each SA constituent module (action gesture are recognized allowing variations such planner, self agent modeler, external agent model- as stopping or reverse motions. er, conflict resolution advisor, and autonomy rea- soner) will be demonstrated for the domain prob- lem of radar frequency management among dis- Self-Explanatory Simulators for tributed naval ships. Education: A Demonstration Kenneth D. Forbus, Mike Oltmans, and George Lee, The Institute for the Learning Sciences, Northwestern “Squeaky Wheel” University Optimization Demonstration Creating new kinds of educational software has been one motivation for qualitative physics. David E. Joslin, i Technologies, and David P. Clements, University of Oregon Self-explanatory simulators combine the preci- This demonstration shows how squeaky sion of numerical models with qualitative repre- wheel optimization (SWO) has been applied to sentations to provide both numerical data and a scheduling domain. For other possible do- conceptual, causal explanations. This demon- mains and a detailed description of SWO see stration shows how we are using self-explanato- the proceedings. The core of SWO is the con- ry simulators in a new architecture for educa- struct/analyze/prioritize cycle. On each iteration tional software, active illustrations, that pro- a solution is constructed by a greedy algorithm, vides stand-alone simulation laboratories and a making decisions in an order determined by pri- new type of media for hypermedia systems. We orities assigned to the elements of the problem. will demonstrate several examples of self-ex- That solution is then analyzed to find the ele- planatory simulators for education, including ments of the problem that are handled poorly in the Evaporation Laboratory, an atmosphere that solution. The priorities of those elements simulation, and space-related simulations. We are then increased, causing the greedy construc- will explain how the SIMGEN self-explanatory tor to deal with them sooner on the next itera- simulation compiler works, and the authoring tion. (“The squeaky wheel gets the grease.”) environment we have developed for creating

Intelligent Systems Demonstrations Systems Intelligent This cycle continues until some termination  condition occurs. The construction, analysis demonstration will highlight the system’s ability Intelligent Systems Demonstrations and prioritization are all in terms of the ele- to perform air-to-air combat, and to interact ments that define a problem domain. In the with human controllers. scheduling problems shown in the demo, those elements are the orders to schedule. TRIPS: The Rochester Interactive Planning System STEVE: A Pedagogical George Ferguson and James Allen, University of Agent for Virtual Reality Rochester Jeff Rickel and W. Lewis Johnson, Information Sci- This demonstration showcases TRIPS, The ences Institute & Computer Science Department, Uni- Rochester Interactive Planning System, an intel- versity of Southern California ligent, collaborative, conversational planning To master complex tasks, such as operating assistant. TRIPS collaborates with its user using machinery, people need hands-on experience fac- both spoken dialogue and graphical displays to ing a wide range of situations. Since it is often im- solve problems in a transportation logistics do- practical to provide such training on real equip- main. The system understands the interaction ment, we are exploring the use of virtual reality in- as a dialogue between it and the human. The di- stead; training takes place in a D, interactive, alogue provides the context for interpreting hu- simulated mock-up of the student’s work envi- man utterances and actions, and provides the ronment. Since mentors and teammates are often structure for deciding what to do in response. A unavailable, we are developing an autonomous, variety of AI technologies, including planning, animated agent, Steve, that cohabits the virtual scheduling, and simulation, are integrated by world with students to play these roles. Steve can TRIPS to produce solutions in response to hu- demonstrate tasks as well as monitor students man guidance. With the human in the loop, while they practice tasks, providing assistance they and the system together can solve harder when needed. Steve integrates many AI tech- problems faster than either could solve alone. In niques: it can generate and recognize speech; our demonstrations, users are encouraged to sit demonstrate actions; use gaze and gestures; an- down and try the system, with only rudimenta- swer questions; construct, execute, and revise ry guidance from us. Further information is plans; and discuss past actions based on an episod- available in our AAAI- paper “TRIPS: An Inte- ic memory. Steve has been tested on a variety of grated Intelligent Problem-Solving Assistant.’’ naval operating procedures, and can provide in- struction in a new domain given only the appro- priate declarative knowledge. Virtual Mattie Activity Monitor Scott Dodson, University of Memphis TacAir-Soar: Generating Virtual Mattie (VMattie) is a clerical software Autonomous Behavior for a agent that is capable of actively gathering infor- Distributed Military Training mation from humans, composing announce- Environment ments of seminars, and distributing the an- nouncements without human intervention. Randolph M. Jones, John E. Laird, Paul E. Nielsen, Karen Coulter, Frank Koss, and Patrick Kenny, Arti- VMattie sends and receives information in the ficial Intelligence Laboratory, University of Michigan form of natural language, freeform email mes- TacAir-Soar is a software system that gener- sages. It must maintain the necessary distribu- ates complex, intelligent behavior in real time, tion lists, send announcements in a timely man- to support military training by simulation. It is ner, and will remind organizers to send informa- a large (, rules) rule-based system, which tion if needed. VMattie is a multi-agent system controls synthetic models of US military fixed- which embodies and extends several AI architec- wing aircraft. It autonomously performs all of tures including Maes’ behavior net, Hofstadter the missions typically performed aboard fixed- and Mitchell’s Copycat architecture, a black- wing aircraft, including defensive and offensive board, and a neural network. It is written com- counter-air, close-air support, suppression of pletely in Java. The VMattie Activity Monitor is enemy air defense, strategic attack, escort, air- a system used to test, monitor, and demonstrate borne early warning, reconnaissance, mid-air the capabilities of VMattie. It includes a client tanking, and forward air control. TacAir-Soar is application which is capable of executing as an implemented within the Soar architecture for applet (strictly in a web browser), and uses pub- cognition, which contains state-of-the-art tech- lish/subscribe, or “push,” technology to receive nology for real-time pattern matching. This real-time updates from VMattie.  ducing world-class computer players into tour- Hall of Champions nament play. The Hall of Champions includes two spec- Man versus machine — who is better? In artifi- tators’ areas where AAAI attendees can view cial intelligence, this battle is usually carried out matches as they progress. The Hall of Champi- by playing a game. In the short lifespan of com- ons will be open during exhibit hours (see puting science and artificial intelligence, con- schedule below). Admittance to the Hall of siderable effort has been devoted to creating Champions is included in the technical pro- game-playing programs capable of meeting and gram registration fee or the onsite exhibits-only exceeding human abilities. A scorecard of com- registration fee. High School students are wel- puter accomplishments in this area might read come and will be admitted without fee upon as follows: presentation of a valid high school student ID ■ Solved—Computers can play some games card. Children under  will also be admitted perfectly (Connect- and Go Moku, for ex- without fee, but must be accompanied by an ample). adult conference registrant. ■ Computer Champions—Computers are in- disputably better than all humans in games such as Checkers and Othello. Disclaimer ■ Undecided—It is not clear whether man or machine is better in games such as Backgam- This is an educational exhibition, not a compe- mon, Chess, and Scrabble. tition. The programs and humans participating ■ Emerging—Great strides have been made re- in the Hall of Champions are all outstanding; cently in Bridge and Poker,with the each participant may or may not be the human prospects of a computer program being a or computer champion of the game. The per- worthy challenger to the human world sons or programs currently holding champi- champion only a few years away. onships are determined by the governing orga- ■ Human dominance—Some games have been nizations of the various games. Participation in resistant to progress. For example, research the AAAI Hall of Champions has been deter- into achieving high-performance Go pro- mined primarily by excellence of play, but also grams is still in its infancy. by suitability for our educational mission and The Hall of Champions presents several by the scheduling constraints of the event.

Hall of Champions Hall game-playing exhibitions. Competitions be- tween evenly matched opponents offer the most interest, as evidence by last year’s chess match Expert Players Schedule between Garry Kasparov and IBM’s Deep Blue. This year, AAAI is highlighting two undecided Tuesday, July  games: Backgammon and Scrabble. Who is bet- ■ : AM – : PM: Bridge: GIB vs Zia ter at Backgammon? Gerry Tesauro’s TD-Gam- Mahmoud & Michael Rosenberg mon or world champion Malcolm Davis? Who ■ : AM – : PM: Backgammon: TD is better at Scrabble? Brian Sheppard’s Maven or Gammon vs Malcolm Davis Grandmaster Adam Logan? Both matches will ■ : PM – : PM: Scrabble: Maven vs be played over several days, allowing for enough Adam Logan games to be played to get more insight into ■ : PM – : PM: Panel Discussion, “AI whether man or machine is the better player. Game-Playing Techniques: Are They Useful The Hall of Champions also features exhibi- for Anything Other than Games?” tions in the emerging games of Bridge and Pok- Wednesday, July  er, as well as in Go. ■ : PM – : PM: Backgammon: TD Gam- AAAI- attendees will be able to interact mon vs Malcolm Davis with these game-playing programs in a variety Go: Many Faces of Go vs James Kerwin of ways. First, attendees can watch the competi- ■ : PM – : PM: Scrabble: Maven vs tions. All games will have commentary provid- Adam Logan ed by both the game programmer and the hu- Bridge: Bridge Baron man opponent. Second, most of the programs will be available during the conference for at- Thursday, July  tendees to play against them. Finally, the pro- ■ : AM – : PM: Scrabble: Maven vs grams’ authors will be available to discuss both Adam Logan the technical issues involved in creating the pro- ■ : AM – : PM: Poker: Loki grams and the social issues involved in intro-  that we can highlight innovative research along Robot Competition Seventh Annual Mobile with the more robust systems. To this end, there will be rounds of differing challenge. Also, hu- Robot Competition & man participants will be required to attend a Exhibition post-competition workshop where they will de- scribe and discuss their techniques. The Robot Competition and Exhibition will be held in the Exhibit Hall of the Monona Terrace Convention Center, and will be open to regis- Event : Hors d’Oeuvers Any- tered conference attendees during exhibit hours. one? Robot Interaction Event Following in a long tradition of popular mo- bile robot competitions, this year’s event will This event will take place during the AI Festival provide conference attendees with a first hand on Wednesday evening in the exhibit hall. look at the progress in the fields of artificial in- Robots may either be in a penned area, or free telligence and robotics. The competition will to mingle with all attendees. consist of two events which will focus on de- Objective: The objective of this competition tecting signs of past and current life on Mars is to act as service robots, serving hors d’oeuvers and testing the robots ability to safely serve re- to attendees at the reception, and handing out freshments and interact with guests. The exhibi- flyers and making announcements between reg- tion will showcase current research in robotics ular conference sessions. This year, each contes- that does not fit into the competition tasks. tant is required to explicitly and unambiguous- AAAI gratefully acknowledge grants from ly demonstrate interaction with the spectators. DARPA and the National Science Foundation In keeping with the IJCAI panel “The Next Big for student travel to this event. Thing,” more natural modes of communication are necessary for society’s acceptance of robots. Event I: Find Life on Mars Furthermore, this helps distinguish the AAAI competition from other competitions. Also different from last year, robots will be Mission Objective: The goal of the Find Life on allowed to touch attendees! Specifically, in their Mars event is to seek out new life forms, collect attempt to serve food, a robot may “nudge” a them, categorize them, and return them back person in order to get through a crowd and safely to the Mars Lander. serve food to other groups of people. In addi- Scenario: The robot has just landed on Mars. tion to emphasizing interaction with attendees, It is an inhospitable place: polished cement manipulation is encouraged, either by refilling ground, large black rocks, danger zones, and serving trays autonomously, or in physically other obstacles jutting from the Martian land- handing out the food or flyers to the attendees. scape. Behind the robot sits the Mars Lander. It Awards: Of greatest importance this year will is the capsule that the robot rode for many be a series of Technical Innovation Awards that weeks to get here. It has two access doors on its will be given for specific accomplishments. narrow ends. This is where the robot will de- These will highlight entries that have some posit life forms. noteworthy innovation regardless of how well Time is of the essence: The robot only has the entry performed in the competition, and five to ten minutes to carry out its mission. As will be awarded in such areas as: distinguishing the robot boldly goes where no robot has gone humans from inanimate things (they don’t offer before, it sees nothing but the desolate Martian cookies to tables!), gesture recognition, nudging rock and obstacled landscape. But wait, there it using a manipulator, personality, enabling two- was again. A small colorful object about the size way conversations with a human being, use of of a tennis ball. The robot races to the Martian, vision-based sensing, recognizing VIP’s by rib- picks it up, carries it back to the Lander and bons on badges and addressing them different- places it carefully into one of the two access ly, and best integration effort. doors, and off it goes again. In addition to the Technical Innovation Spirit of the Games: The purpose of the Find Awards, the reception event will have a first, sec- Life on Mars event is threefold: Promote new ond, and third place award for technical merit, research and innovative ideas in robotics. En- based on the judges’ scores from the Qualifica- courage robust, real-world solutions. Enhance tion/Safety Round and from the performance in information exchange between researchers. the reception event. In order to determine these This year awards will be given for technical prizes, robots will actually be scored based on innovation, as well as performance. We hope  reaching various levels of competency. Some of ment, the silent presence of the plant fills an these competencies are binary, and others in- emotional niche. Unfortunately, this plant is of- volve some scoring function. The reception ten dying; it is not adapted to the fluorescent event will also have a popular vote for the at- lighting, lack of water, and climate controlled tendees favorite robot. air of the office. Office Plant # (OP#) is an ex- ploration of a technological object, adapted to the office ecology, which fills the same social Robot Event Judges & Chairs and emotional niche as a plant. OP# monitors the ambient sound and light level, and, employ- Robot Competition Cochairs ing text classification techniques, also monitors its owner’s email activity. Its robotic, sculptural Gregory Dudek, McGill University; Robin body, reminiscent of a plant form, responds in Murphy, Colorado School of Mines; and David slow, rhythmic movements to express a mood Kortenkamp, NASA/Ames Research generated by the monitored activity. In addi- Robot Exhibition Cochairs tion, low, quiet, ambient sound is generated to Tucker Balch, Georgia Institue of Technology express this mood; the combination of slow and Karen Zita Haigh, Carnegie Mellon Uni- movement and ambient sound thus produces a versity sense of presence, responsive to the changing ac- tivity of the office. OP# is a new instantiation Robot Competition Judges of our notion of *intimate technology*, that is, ■ Find Life on Mars: Maria Gini, University of technologies which address human needs and Robot Teams Robot Minnesota; Lisa Meeden; Douglas S. Blank, desires as opposed to technologies which meet University of Arkansas; Nicola Ferrier exclusively functional task specifications. ■ Hors d’Oeuvers Anyone?: Alan Schultz, Naval Research Lab; Illah Nourbaksh; Holly Yanko, Exhibitor Massachussets Institute of Technology Georgia Institute of Technology ■ Technical Merit Awards: Vladimir Lumelsky Robot: JavaBots and S.W. Zucker, Yale University Team: Tucker Balch JavaBots is a freely-distributable software sys- tem for developing and running multi-robot Mobile Robot Competition control systems on mobile robots and in simu- Workshop lation. The system was used by Georgia Tech to control their winning multi-robot entry in the ■ Organizers: Gregory Dudek, Robin Murphy AAAI- Mobile Robot Competition. JavaBots and David Kortenkamp is also used in a number of other laboratories in Thursday, July  ongoing research. At AAAI- we will demon- : AM – : PM strate the simulation capabilities of JavaBots as Hall of Ideas J, Monona Terrace well as a videotape of robots using the system.

Exhibitor and Competitor Robot Competition and Georgia Institute of Technology Exhibition Teams Robot: Pepe Team: Alexander Stoytchev and Rawesak Tanawong- Exhibitor and Competitor suwan Carnegie Mellon University Competitor Robot: Nomad Team Members: Mark Maimone, Reid Simmons, and McGill University Dimi Apostolopoulos Robot Name: Invader Advisor: Greg Dudek Exhibitor Team Leader: Francois Belair Team Members: Francois Belair, Scott Burlington, Carnegie Mellon University Robert Sim, Eric Bourque, Andrew Ladd, and Robot: Office Plant # Gillaume Marceau Team: Michael Mateas and Marc Boehlen Invader is a Nomad  built by Nomadic Walk into a typical, high tech office environ- Technologies equipped with  sonar sensors and ment, and, among the snaking network wires, a monocular color camera for external sensing. glowing monitors, and clicking keyboards, you Last years “Mars Mission” was a big success for In- are likely to see a plant. In this cyborg environ- vader, it is looking forward to this years challenge.  When Invader wakes on the foreign planet, our current  MIP processor). Rather than us- Robot Teams it will use all the information that it can attain ing a single, centralized symbolic world model, to help it on its quest. Combining all this infor- working memory is distributed amongst a num- mation by way of an extended Kalman filter, In- ber of different sensory-motor and memory vader will begin searching for aliens, every step subsystems, each of which supports representa- of the way incorporating aquired information tions that are tailored to a particular common with the information that it already has com- task. Variable binding is also performed by and piled, in a global map. This map will help In- distributed through these peripheral systems. vader track down all the aliens that it comes ac- cross for identification and the aquisition of in- Exhibitor teresting samples from the environment. SRI International/ Invader will identify some of its targets by Rochester University their color, others by their shape. Color seg- Robot: Realtime Stereo and People-Tracking mentation is fed to a principle components Team Members: Kurt Konolige and Chris Eveland analysis mechanism to help differentiate among SRI’s Small Vision System performs realtime all of the potential targets. Shape recognition is stereo analysis using standard PC hardware. We done by comparing edge, corner and curve in- will demonstrate this system in a people-track- formation against all the shapes that Invader ing application. knows about. Invader will use natural language to convey its findings, and of course can supple- Exhibitor and Competitor ment that by downloading a global map for lat- University of British Columbia er reference. Invader is looking forward to this Robot: Jose and/or Spinoza years journey. Team Members: Don Murray, Jim Little, Rod Barma, Cullen Jennings, and Stewart Kingdon Competitor MIT Artificial Intelligence Lab Exhibitor Robot: Cog University of Minnesota Team Leader: Brian Scassellati Robot: TBMin and new (yet to be named) robots Team: Maria Gini and Paul Rybski Competitor Naval Research Lab Competitor Robot: Coyote and Roadrunner University of New Mexico Team Leader: Alan Schultz Robot: Nomadic Scout Team Advisor: Dr. Greg Heileman Exhibitor and Competitor Team Members: Traci Vanek, Maureen Ballas, Northwestern University Melody Romero, Jane Canulette, Liz Kurens, and Rhonda Arkana Robot: Kludge One of the University of New Mexico’s en- Team Advisor: Ian Horswill Team Leader: Dac Le tries is the Nomadic Scout. The Nomadic Scout Team Members: Lars Bergstron, Robert Zubek, Mark was purchased by the UNM student branch of DePristo, Matt Brandyberry, Shashi Buluswar, Dac the IEEE Computer Society to enhance the ex- Le, and Ian Horswill isting robotics program. It is the first commer- Kludge is a low-cost robot that incorporates cial robot purchased by the school. In prepara- real-time vision with a novel cognitive architec- tion for the Mars Explorer phase of the compe- ture. Kludge can track up to three objects si- tition, the Scout was equipped with a vision sys- multaneously, avoid obstacles using vision, fol- tem for object identification. This vision system low simple instructions, play ball and chase is comprised of a Newton Labs Cognachrome games, etc. It’s simplified architecture consists of board utilizing the ARC development system a set of sensory-motor systems, a logic-based along with a CCD color camera. An onboard problem solver, a Society-of-Mind-like frame Gateway portable computer running a Pentium system, and a simple finite-state parser. The II processor on a Linux platform will handle all problem solver can perform forward-chaining high level control of the entire system. inference on a subset of modal logic involving single-place predicates and single-level quantifi- Competitor cation. Axioms are compiled into a TMS-like University of New Mexico feed-forward network, allowing the system to Robot: Lobotomous recompute all inferences from scratch on every Advisor: Dr. Greg Heileman cycle of the sense-decide-act loop (-Hz on Team Members: Traci Vanek, Maureen Ballas,  Melody Romero, Jane Canulette, Liz Kurens, Rhonda Exhibitor Arkana, and Dan Stormont University of Southern California One of the University of New Mexico’s en- Robot: Ullanta Theater Troupe tries in this year’s AAAI Mobile Robot competi- Team: Barry Brian Werger tion is Lobotomous. Lobotomous will be en- tered in the human interaction portion of the Competitor event, and this will be its third consecutive ap- University of Texas at Arlington pearance at this contest. Robot: Pioneer  Lobotomous was designed and constructed Team Advisor: Dr. Bill Carroll by UNM students in a senior level design class Team Leader: James Poole in preparation for the  AAAI competition. Team Members: Kiyoko Fujita, Brandon Hennegan, Sandia National Labs is a major contributor to Cary Pillers, Priyath Sandanayake, and Michael Tran the development of Lobotomous and has pro- Team pioneer is a senior design project at the vided much of the hardware used for the pro- University of Texas at Arlington in Arlington, ject. Since , students in the EECE depart- Texas. Our robot is a Pioneer mobile robot with ment for competitions and departmental pro- a gripper package and Fast Track vision system jects have continued development. In  installed. The goal of our project is to gain Lobotomous won first place in the AAAI vacu- knowledge and experience in the field of uming competition and competed in the hors robotics and artificial intelligence. d’oeuvres phase. Exhibitor and Competitor

Robot Teams Robot Exhibitor and Competitor VUB AI Lab University of North Dakota Robot: Babu and Pi Robot: Rusty the B.E.A.R Team: Paul Vogt Team Advisors: Sven Anderson, Henry Hexmoor and Bruce Maxwell Exhibitor and Competitor Team Leader: Bret Reese Independent Team Members: Elizabeth Gordon, Daniel Ibanez- Gomez, Brett Reese, Tim Thompson, Aron Tomson, Robot: Beast and Snake Matt Lafaray, and Mike Trosen Team Leader: Laurent Chabin We have worked on sensory interpretation and fusion of sonar, infrared, and vision. For the competition, we have developed intuitive navi- gation algorithms that detect mobile objects with human skin-color. The sensory detection is done by a low-quality/low-cost vision system. Our vision system detects skin-color and mo- tion at about Hz rate. One of our software modules captures sonar data and detects arti- facts in sonar occupancy grids. This software is also used as an offline analysis tool. It uses stored sonar data from previous runs to allow users to generate sonar occupancy grids over dif- ferent spans of time. Detection algorithms are run on this software for debugging before they are loaded onto the robot. We are extending this software to be a general testbed for other senso- ry data such as vision and infrared.

 Registration Registration Conference registration will take place outside the Exhibition Hall, Lakeside Commons, on the first level of the Monona Terrace Convention Center, beginning Sunday, July . Registration hours are: Sunday, July  : AM – : PM Wednesday, July  : AM – : PM Monday, July  : AM – : PM Thursday, July  : AM – : PM Tuesday, July  : AM – : PM Only checks drawn on US banks, VISA, MasterCard, American Express, government purchase or- ders, traveler’s checks, and US currency will be accepted. We cannot accept foreign currency or checks drawn on foreign banks.

Registration Fees AAAI-/IAAI- Technical Program Your AAAI-/IAAI- technical program registration fee includes admission to all technical paper sessions, invited talks and panels, exhibitions, the Student Abstract Poster Session, the opening recep- tion, the AI Festival, AAAI-/IAAI- Conference Proceedings and the Special Tutorial MP. Note: Students must present proof of full-time student status to qualify for student rate. Onsite technical program fees are: Regular Member $ Regular Nonmember $ Student Member $ Student Nonmember $ Tutorial Forum The tutorial forum registration includes admission to no more than four consecutive tutorials and the corresponding four tutorial syllabi. Extra syllabi from the other tutorials may be purchased sep- arately for $. each. The tutorial forum registration also includes admission to all exhibit hall programs. Please note that you need not register for the Tutorial Forum to attend the Special Tu- torial MP on Monday, July . Onsite Tutorial Forum fees are: Regular Member $ Regular Nonmember $ Student Member $ Student Nonmember $ Special Second-Day (Monday, July  only) Tutorial Forum Registration fee for COLT / ICML / UAI attendees only: Regular attendees may deduct $. and students $. from the fees listed above. Workshop Program Workshop registration is limited to those active participants determined by the organizer prior to the conference. All workshop participants must register for the AAAI- technical program or, in the case of the four cosponsored workshops, must register for one of the cosponsoring conferences. (Excep- tions to these rules will be required to pay a $. fee per workshop.) Registration onsite for a workshop is possible with the prior permission of the corresponding workshop organizer. Robot Building Lab The robot building lab registration fee includes admission to the robot building lab and the exhi- bition program. Fees are $. for members or nonmembers, and $. for students. Atten- dance is limited and preregistration is recommended. Check for availability onsite. Exhibition Admission to the exhibition hall programs is included in all other types of registration. For individ- uals interested in admittance to the exhibit hall only, an exhibits only registration is available in on- site registration. The fee is $. for a one-day pass, and $. for a three-day pass. Exhibit hall programs include vendor exhibits, the Hall of Champions, the Intelligent Systems Demonstrations, and the Robot Competition and Exhibition. High-school students are welcome and will be admit- ted without fee upon presentation of a valid high-school student ID. Children under  will also be admitted without fee, but must be accompanied by an adult conference registrant. Please note: The AI Festival, which will be held in the exhibit hall, is included in the technical registration fee only. All other attendees must pay an additional fee.

 AAAI Logo Shirts ■ Seventh floor, Sheraton Madison Hotel Hours: : AM – : PM daily Polo shirts with the AAAI logo will be for sale Services include fax, copies, laser printing, and during registration hours outside the Exhibition other general office services. The Madison Con- Hall, Lakeside Commons, on the first level of course, and the Sheraton Madison Hotel offer the Monona Terrace Convention Center. Sup- shipping by Federal Express and UPS. plies are limited. Price $. each onsite. Career Information Admission A bulletin board for job opportunities in the ar- Each conference attendee will receive a name tificial intelligence industry will be made avail- badge upon registration. This badge is required able in the registration area, outside the Exhibi- for admittance to the technical, tutorial, exhib- tion Hall, on the first level of the Monona Ter- it, IAAI and workshop programs. Workshop at- race Convention Center. Attendees are welcome tendees will also be checked off a master regis- to post job descriptions of openings at their tration list at individual rooms. Smoking, company or institution. drinking and eating are not allowed in any of the technical, tutorial, workshop, IAAI, or ex- hibit sessions. Child Care Services For information about child care services, you Baggage Holding may contact Be My Nanny at --. The agency requests forty-eight hours notice. (This in- There is no baggage holding area at the Monona formation is provided for your convenience and Terrace Convention Center. Please check your does not represent an endorsement of this agency luggage with the bellman at your hotel after you by AAAI. Responsibility for all child care arrange- have checked out. Neither the AAAI, the ments must be assumed by the parents.) Monona Terrace Convention Center, the Madi- son Concourse Hotel, the Best Western Inn on the Park, nor the Sheraton Madison Hotel ac- Coffee Breaks cept liability for the loss or theft of any suitcase, briefcase, or other personal belongings brought Coffee will be served in the Grand Terrace, lev- el four, Monona Terrace Convention Center; in General Information General to the site of AAAI-/IAAI-. the foyer space, second floor, Madison Con- course Hotel; and in the mezzanine and pool Banking terrace, second level, Inn on the Park. Coffee breaks in the Monona Terrace Convention Cen- The closest bank and automated teller machine ter and the Inn on the Park are scheduled for (ATM) are located at the M & I Bank at  West : – : AM and : – : PM each day. Main Street. The ATM networks available are Coffee breaks in the Madison Concourse Hotel American Express, MasterCard, Visa, Cirrus, are scheduled for : – : AM and : – : Plus and Money Network. The M & I Bank can PM during events in the hotel. also exchange all major foreign currencies. ■ The M & I Bank  West Main Street Copy Services Madison, WI  Telephone: () - Copy services are available at: ■ Econo Print Monday – Friday: : AM- : PM Closed Saturdays and Sundays Contact: Mark Kamplin  South Pinckney Street Madison, Wisconsin  Business Centers Telephone: -- Hours: : AM – : PM, Monday – Friday Business Centers are available at the following Copy service is also available at the Business locations: Centers in the conference hotels. ■ Lobby of the Madison Concourse Hotel Hours:  hours per day  Dining List of Attendees General Information

Madison dining information is available at the A list of preregistered attendees of the confer- Visitor Information Booth, near the main en- ence will be available for review at the AAAI trance on the fourth level of the Monona Ter- Desk in the registration area on the first level of race Convention Center. Concessions will be the Monona Terrace Convention Center. At- open on the Rooftop Terrace and on the fourth tendee lists will not be distributed. level of the Monona Terrace Convention center from, July  – . Message Center

Handicapped Facilities See Information Desk

The Monona Terrace Convention Center, the Madison Concourse Hotel, the Best Western Parking Inn on the Park and the Sheraton Madison Ho- tel are all equipped with handicapped facilities. Parking is available at the Monona Terrace Con- vention Center. The maximum daily rate is $.. Housing

For information regarding hotel reservations, Press please contact the hotels directly. For student housing reservations assistance, please contact All members of the media are requested to reg- the University of Wisconsin – Madison Confer- ister in the Press Room on the fourth level of the ence Groups Office, University Housing at - Monona Terrace Convention Center, Meeting -, : AM – : PM, Monday – Friday. Room N. Press badges will only be issued to in- Students requiring assistance after hours should dividuals with approved credentials. The Press refer to the contact information provided in the Room will be open during the following hours: student housing confirmation letter. Monday, July  : AM – : PM Tuesday, July  : AM – : PM Wednesday, July  : AM – : PM Information Desk Thursday, July  : AM – : PM An AAAI- volunteer will be on duty during An information desk/message desk will be press room hours to assist the members of the staffed during registration hours, Sunday press and media. through Thursday, July  – . It is located near the main entrance, on the fourth level of the Monona Terrace Convention Center. Mes- Printed Materials sages will be posted on the message boards adja- cent to the desk. The telephone number for Display tables for the distribution of promo- leaving messages only is () -. Paging tional and informational materials of interest to attendees is not possible. conference attendees will be located outside the Exhibition Hall on the first level of the Monona Terrace Convention Center. Internet

Internet access in Hall of Ideas G on the fourth Proceedings level of the Monona Terrace Convention Cen- ter. The internet room will be open during reg- Each registrant for the AAAI- technical pro- istration hours. As a courtesy, please limit your gram and IAAI- will receive a ticket with the access time to - minutes if others are waiting registration materials for one copy of the confer- to use the service. The AAAI- Internet Room ence Proceedings. During registration hours on is sponsored by Microsoft Corporation. AAAI Sunday, July , Monday, July  and until : gratefully acknowledges Microsoft’s generous AM on Tuesday, July . Proceedings tickets can be contribution that makes this service available. redeemed at the AAAI Press Proceedings desk, lo- cated near the main entrance on the fourth level of the Monona Terrace Convention Center.  After : AM on Tuesday, the AAAI- speakers visit this room to organize their materi- /IAAI- Proceedings ticket may be redeemed als. The room will be open from : AM – : at the MIT Press booth # , located in the Ex- PM Sunday, July  through Wednesday, July  hibition Hall, during exhibit hours. and from : AM – : PM, Thursday, July . Extra Proceedings may be purchased at the Invited speakers are asked to come to Meet- conference site at the above locations. Thursday, ing Room K one day prior to their speech. Rep- July , will be the last day to purchase extra resentatives from AV Headquarters will be avail- copies of the Proceedings onsite. able from : AM – : PM Sunday, July  The AAAI-/IAAI- Proceedings can also be through Wednesday, July  and from : AM redeemed by mailing the ticket with your name, – : PM, Thursday, July  to confirm your shipping address and e-mail to: audiovisual needs, and assist with the prepara- ■ Exhibits tion of your materials, if necessary. The MIT Press  Cambridge Center Cambridge, MA  Transportation Postage must be prepaid with a check or Mas- terCard/Visa and expiration date. USA: $.; The following information provided is the best Outside USA: $. surface or $. airmail. available at press time. Please confirm fares when making reservations. Proceedings Shipping Airlines and Rental Cars The American Association for Artificial Intelli- A Mail Boxes Etc. booth will be located outside gence has selected American Airlines and Unit- the Exhibition Hall, Lakeside Commons, on ed Airlines as the official co-carriers and Avis the first level of the Monona Terrace Conven- Rent A Car as the official car rental agency for tion Center. The booth will be open on Tuesday, AAAI-/IAAI-. If you need to change your July  and Wednesday, July  from : AM – airline or car rental reservations, please call : PM and on Thursday, July  from : – Conventions in America, our official travel    : PM. agency at - - and ask for Group #. E-mail: fl[email protected] Airport Shuttles Recording Complimentary Hotel Airport Shuttles: The Madison Concourse Hotel, the Best Western Inn

General Information General No audio or video recording is allowed in the on the Park and the Sheraton Madison Hotel. Tutorial Forum. Audiotapes of the plenary ses- sions, invited talks and panels, and the IAAI ses- Taxi sions will be for sale near the main entrance on Taxis are available at Dane County Regional the fourth level of the Monona Terrace Con- Airport. Approximate fare from the airport to vention Center. A representative from Audio downtown Madison is $.. Archives will be available to take your order during registration hours, beginning on Tues- Bus day, July . Order forms are included with reg- Van Galder Bus Lines—Downtown Chicago, istration materials. Tapes may also be ordered by O’Hare Airport. The depot is located at the mail from: University of Wisconsin-Madison Memorial ■ Audio Archives International, Inc. Union. For information on fares and schedul-  Foothill Blvd., Suite  ing, call --. La Crescenta, CA  Badger Bus Lines—Mitchell Field, Milwaukee Telephone: -- Airport provides service to the University of Fax: -- Wisconsin – Madison campus and the Madison Badger Bus Depot at  South Bedford Street, Madison, WI . For information, call - Speaker Ready Room -.

The Speaker Ready Room will be located in City Transit System Meeting Room K on the fourth level of the Madison Metro Transit System is a citywide bus Monona Terrace Convention Center. This room transit system. Schedules are available in the has audio-visual equipment to assist speakers Monona Terrace Convention Center. Basic local with their preparations. It is important that fare is $.. You may buy a booklet of ten rides  for $.. There is a Free Fare Zone from : acts only in the capacity of agent for the Suppli- General Information AM – : PM around the center of Madison. For ers which are the providers of the service. Be- general information, call --. cause AAAI has no control over the personnel, equipment or operations or providers of accom- modations or other services included as part of Tutorial Syllabi the AAAI-/IAAI- program, AAAI assumes no responsibility for and will not be liable for Extra copies of AAAI- tutorial syllabi will be any personal delay, inconveniences or other available for purchase in the registration area, damage suffered by conference participants outside the Exhibition Hall, Lakeside Com- which may arise by reason if (I) any wrongful or mons, on the first level of the Monona Terrace negligent acts or omissions on the part of any Convention Center. Supplies are limited. Cost Supplier or its employees, () any defect in or is $. per syllabus. Preregistration tutorial failure of any vehicle, equipment or instrumen- syllabi tickets may be redeemed in the tutorial tality owned, operated or otherwise used by any rooms. Supplier, or () any wrongful or negligent acts or omissions on the part of any other party not under the control, direct or otherwise, of AAAI. Visitor Information

The Monona Terrace Convention Center will have a Visitor Information desk near the main entrance on the fourth level of the Monona Ter- race Convention Center. Maps and brochures with information on shopping, restaurants, out- door activities, parks, and tours will be avail- able. Hours are : – : PM, Sunday, July  – Monday, July ; : – : PM, Tuesday, Ju- ly  – Wednesday, July ; : – : PM, Thursday, July .

Volunteer Room

The volunteer room is located in Meeting Room R of the fourth level of the Monona Ter- race Convention Center. Hours are : AM – : PM, Sunday, July  – Wednesday, July  and : AM – : PM, Thursday, July . Ex- tra volunteer instructions and schedules will be available. All volunteers should check in with Josette Mausisa, AAAI Registrar, in the registra- tion area prior to their shifts. The volunteer meeting will be held Saturday, July  at : PM in Hall of Ideas E&F of the Monona Terrace Convention Center.

Disclaimer

In offering American Airlines, Avis Rent A Car, Be My Nanny, Conventions in America, the Best Western Inn on the Park, the Madison Concourse Hotel, the Sheraton Madison Hotel, United Airlines, University of Wisconsin – Madison, and all other service providers (here- inafter referred to as “Supplier(s)” for the Na- tional Conference on Artificial Intelligence and the Innovative Applications Conference), AAAI 

.

d

R

)

n

1

N o t o 5 r h t y h g p o u w o r H t Dane County t D ( M r. S H Regional ➢ . y wy Airport N w 1 I 13 n

H t . e e v r

Ave. s ty . t

A a e

v n t to e A g ) 9

Coun 1 n 0 an i h 5 1 - Ln. s 9

rm a International y 4 W w he Anderson St. . (H

S E Packers

N.

A b erg A ve Airport Rd. Cen . tu ommercial Ave. ry C ) A 2 v Hwy 30 Interstate 94 1 e . ks Ave. y LAKE MENDOTA a w O St. H Milwaukee ( . . d t ir v S

l a F

(Hwy 14) B o g

n 1st a e A

l b t l w e oo A n d . n A y i v w e W . rove Rd. H Cottage G t. e . S n t University t. S n S so Bay Dr. H n m ig State m o a . s h a n li l l d a rh h i n R Street o Jo W. Belti W d ampus D G C r. l Blair St. M keye Rd. Old Sauk University Buc

Rd. Ave. o

n o A a

u n v q

n e e t S nn a .

o gen t. i Ter.

e Breese I R Regent St. D n m Proudfit W

r t

e m .

N r a d. LAKE d. ich Rd. Pflaum R s

y ols t G a a

Blv t. Erin t e Mineral Point Rd. W S MONONA e 9

e o l r 0

y n a

e . o d v Speedway Rd. John Nolen Dr. n M t R d i e i LAKE h o eg M

W S WINGRA Odana Rd. ay . Broadw d

R .

a t

m S o k k a r 8) N a (Hwy 12-1

P UPPER

. .

d . S MUD

R d . R

Beltline Hwy. LAKE d

k )

R

c 1

) o n

r 5

2-14-18-151) 4 (Hwy 1 o 1 t

m y

i

h

w R w

g

. y y

u

y r H

H ( e ( o . w h t d H tc S R . le a S a o H n n o i r h LAKE s e m i e F V S WAUBESA (Hwy 18-151)

