Newsgroups: comp.robotics
Path: brunix!news.Brown.EDU!noc.near.net!howland.reston.ans.net!agate!linus!linus.mitre.org!starbase!bonasso
From: bonasso@starbase.mitre.org (R. Peter Bonasso)
Subject: AAAI Fall Symposium on Real World Robots
Message-ID: <bonasso.737155661@starbase>
Summary: ENclosed Is An Official Call for participation for the above
Keywords: autonomous agents, robot intelligence, real-world tasks
Sender: news@linus.mitre.org (News Service)
Nntp-Posting-Host: starbase.mitre.org
Organization: The MITRE Corporation
Distribution: usa
Date: Tue, 11 May 1993 21:27:41 GMT
Lines: 195


   Here is the official call for participation in the 1993 
   AAAI Fall Symposium Series. Included only is the discussion for the
   robot symposium "Instantiating Real World Agents".

   AAAI
   Fall Symposium Series
   22 - 24 October, 1993
   Raleigh, North Carolina


   AAAI presents the 1993 Fall Symposium Series to be held Friday through 
   Sunday, October 22-24, 1993 at the Sheraton Imperial Hotel & 
   Convention Center, Research Triangle Park, Raleigh, North Carolina.

   The topics of the five symposia in the 1993 Fall Symposium Series are:
     - Automated Deduction in Nonstandard Logics
     - Games: Planning and Learning
     - Human-Computer Collaboration: Reconciling Theory, Synthesizing
	Practice 
     - Instantiating Real-World Agents
     - Learning in Computer Vision: What, Why and How?

   Most symposia will be limited to approximately 60 participants. Each 
   participant will be expected to attend a single symposium. Working notes 
   will be prepared and distributed to participants in each symposium. 

   A general plenary session will be scheduled in which the highlights of 
   each symposium will be presented, and an informal reception will be held 
   on Friday evening, October 22.

   In addition to invited participants, a limited number of other interested 
   parties will be allowed to register in each symposium. Some student 
   scholarship money may be available. Registration information will be 
   available in late July, 1993. To obtain it, write to:

   AAAI
   Fall Symposium Series
   445 Burgess Drive
   Menlo Park, California 94025
   (415) 328-3123
   fss@aaai.org


   Submission Requirements

   Submission requirements vary with each symposium, and are listed in the 
   descriptions of the symposia. Please send your submissions directly to the 
   address given in the description.

   DO NOT SEND submissions to AAAI. 

     - All submissions must arrive by June 4, 1993.
     - Acceptances will be mailed by July 2, 1993.
     - Material for inclusion in the working notes of the symposia must be 
       received by August 23, 1993.


   Instantiating Real-World Agents

   Rather than being centered on a research area, or a general unsolved 
   problem area (e.g., intelligent agents), this symposium will concentrate on 
   one specific real-world problem in the hopes that progress can be made on 
   at least this problem, and that a solution to this specific problem will 
   provide clues to solutions to the more general problems.

   This symposium will concentrate on AI as applied to a physically
   instantiated robot for vacuuming household floors.  The target problem is to
   autonomously vacuum your living room, while doing the right thing with
   furniture, pets, trash, etc.  In particular, research on navigation,
   planning, spatial representation, multi-agent control, behavior control,
   obstacle avoidance, perception, exploration, NLP interfaces, etc., will be
   of interest as long as they are related to household vacuuming.  Theoretical
   work, simulations, and implemented systems will all be of interest.
   However, work presented at this symposium should be set in the target
   domain.

   Limiting the discussions to a specific task to be performed without 
   allowing the engineering of a specific solution still leaves a plethora of 
   issues to be explored.  We hope that significant progress can be made on 
   this problem, and that new research methods might grow out of this type 
   of symposia.  Perhaps most importantly, we are hopeful that a common 
   problem domain may obviate the vocabulary problems that have crept up 
   in recent years when researchers involved in different problems try to talk 
   to one another.  By having a common problem, we hope a common 
   language will emerge.

   Additionally, by concentrating on a real-world problem domain, we hope that
   some practical progress can be made in this domain, and in the related
   research areas.  Robotics and planning work are too often detached from
   their application areas. It is hoped that this symposium will bring to focus
   some research areas that are of more than just academic interest. It may
   even be possible that participants in this symposium will work on a
   commercial version of this robot-- allowing them to really clean up!

   It is important to keep in mind that we are not looking for an engineered
   solution, but rather one oriented on general techniques which could be used
   in similar tasks. We also welcome both "explore, learn, map, and plan"
   as well as reactive approaches, with a view toward uncovering the value and
   trade-offs of these different approaches.

   Potential participants for the symposium on Instantiating Intelligent 
   Agents should submit a short position paper (2 - 6 pages) that describes 
   either their approach towards addressing the vacuuming problem, or their 
   current research and how it can be related to the floor vacuuming problem.


   Pete Bonasso
   The MITRE Corporation
   MS Z459
   7525 Colshire Drive
   McLean, VA 22102

   Organizing Committee: Pete Bonasso, MITRE, cochair and contact person 
   (bonasso@starbase.mitre.org); David Miller, JPL, cochair; Ramesh Jain, 
   UCSD; Ben Kuipers, University of Texas at Austin


   Machine Learning in Computer Vision: 
   What, Why, and How?

   This symposium will bring together researchers from different specialties 
   in machine learning and computer vision to address issues raised by 
   examining the use of machine learning in computer vision:
     - What elements of a computer vision system might be learned rather 
       than hand-crafted by the designer?
     - What machine learning paradigms are appropriate to the computer 
       vision domain (especially across the signal to symbol transition)?
     - Why or how would learning improve the performance or efficiency of 
       computer vision systems?
     - How do we go about implementing or exploiting the machine learning 
       paradigms which seem most appropriate to the computer vision 
       domain?

   One of the acknowledged problems with computer vision systems is that 
   they tend to be hand-crafted application-specific efforts that embody or 
   reflect rather little in the way of general principles which can adapt easily 
   from one application environment to another. While some in the computer 
   vision are currently reconsidering the goal of general purpose vision 
   systems as possibly too difficult or not relevant, there is still the clearly 
   motivated desire to learn something from the experience in creating a 
   vision system for one application domain that can be used to make it easier 
   to create the vision system.

   Since much of the effort in creating a vision system often lies in creating a 
   database of examples and facts, and in tuning the parameters and 
   operations of the system to the application domain, learning techniques 
   may be of use in addressing this problem. However, it is not yet clear what 
   learning capabilities computer vision systems should have, why these 
   capabilities should result in computer vision systems that display greater 
   competence and generality, or how to go about building vision systems 
   that incorporate learning capabilities.

   >From the standpoint of machine learning systems, visual domains present 
   some interesting problems. The images and outputs of low-level image 
   processing operations tend to be noisy, making it difficult to get true 
   segmentation of images. Thus it is unreasonable to assume that the 
   transition from image signal to symbol is made completely and correctly. 
   Also, large numbers of exactly labeled examples suitable for inductive 
   learning are generally not available. Some domain knowledge and clear 
   examples are  often available suggesting a multi-paradigm learning 
   approach.

   Format: The symposium will contain both invited and submitted papers. 
   There will be several longer talks by invited experts and a number of short 
   talks. We will emphasize an interactive discussion of issues. One panel 
   will be held on the obstacles to applying learning to vision and promising 
   approaches. Panel suggestions are encouraged and may be given to any 
   program committee member. There will be a poster session to encourage 
   broad participation and discussion. 

   To present a paper or poster, submit an extended abstract of three to five 
   pages by email (ascii, latex or postscript) to: hall@csee.usf.edu or 
   kwb@csee.usf.edu or by hard copy to:

   AAAI-Fall Workshop
   Dept. of Comp. Sci. & Engineering,
   ENG 118, 4202 E. Fowler Ave.
   University of South Florida
   Tampa, Fl. 33620

   To attend the workshop without presenting, send a supporting note.

   Program Committee: Kevin Bowyer, University of South Florida, cochair; 
   Chris Brown, University of Rochester; Bruce Draper, University of 
   Massachusetts; Lawrence Hall, University of South Florida, cochair; Tom 
   Mitchell, Carnegie-Mellon University; Dean Pomerleau, Carnegie-Mellon 
   University; Larry Rendell, University of Illinois







