Call for Interest in Participation in IJCAI-99 Workshop

[Gerhard K. Kraetzschmar]

Dear reader of this news group or list: 
(Our apologies, if you receive this multiple times)

We plan to organize a workshop at IJCAI-99 in Stockholm.
The topic is
        Adaptive Spatial Representations for Dynamic Environments 
and we believe it may be of interest to you. Please read the 
draft for the workshop proposal in the attachment for more information
on the workshop.

You can contribute to the workshop by submitting a paper, giving one of 
the survey talks, or one of the commenting statements in a session.

Note: IJCAI permits only active participants for workshops.

If you come to the conclusion that you are indeed interested in the
workshop and want to actively participate, please take the time to respond
with a short email that indicates your interest and kind of contribution.
Please send the email to 
	gkk at acm.org
and include "IJCAI-WORKSHOP" in the subject line. Thanks a lot.

Please do respond soon, because the due date for the proposal is in a 
few days, and we need to collect a list of tentative participants for it.

-- 
Sincerely Yours, Gerhard
---------------------------------------------------------------------------
Dr. Gerhard K. Kraetzschmar                   University of Ulm
Fon: intl.+49-731-502-4155	              Neural Information Processing
Fax: intl.+49-731-502-4156	              Oberer Eselsberg
Net: gkk at neuro.informatik.uni-ulm.de          89069 Ulm   
     gkk at acm.org                              Germany
WWW: http://www.informatik.uni-ulm.de/ni/staff/gkk.html
---------------------------------------------------------------------------

Proposal for IJCAI-99 Workshop
========================================================
Adaptive Spatial Representations of Dynamic Environments
========================================================
Workshop Description:
====================
Spatial representations of some sort are a necessary requirement for
any mobile robot in order to achieve tasks like efficient,
goal-oriented navigation, object manipulation, or interaction with
humans. A large number of approaches, ranging from CAD-like and
topological representations to metric, probabilistic methods and
biologically-oriented representations, has been developed in the past.
Most approaches were developed for solving robot navigation tasks 
and successfully applied in a wide variety of applications. In many
approaches, the spatial representation is a strong simplification of 
the environment (e.g. occupied and free space) and does not permit an 
easy representation of the spatial structure of task-relevant objects.
Furthermore, these approaches can model only static elements of the
environment in an adequate manner. Dynamic elements, such as doors,
changed locations of particular objects, moving obstacles, and humans,
are usually dealt with in one of two ways:
- A purely reactive level temporarily has a transient representation 
	for an anonymous object. The representation is present only as 
	long as the object can be actively sensed; it vanishes
	thereafter and does not settle into some kind of permanent
	representation. (Doors, moving obstacles, humans.)
- Repeated exposure of the robot to both the old and new location of
	an object that changed its position leads to a slow adaptation
	of (long-term) spatial memory. (Moved, relocated objects)

The current representations are sufficient for many tasks that have 
been researched in the past and led to many successful applications.
However, in order to achieve truely useful robots, e.g. for the office
domain, the robot will have to acquire AND MAINTAIN more complete 
models of its environment. It will have to know the precise locations 
(or have a good estimate of it) of objects and subjects it has seen,
if they are relevant for completing a task. Examples: Location of
tools (screwdriver), books, special equipment (video beamer), persons, 
doors, etc.

Thus, for any existing spatial representation, the following 
questions arise?
- Which structural aspects of the environment can be modeled? How?
- Can the representation model dynamic aspects at all? Could it be extended?
- Which kinds of dynamic aspects can be modeled?
- How are the spatial dynamics of an object modeled?
- How is uncertainty dealt with?
- How are the dynamics used to predict various spatial aspects of
	objects?
- Which methods can be used to update/maintain the representation
	based on sensory information?
- What computational effort is required for updating the representation?

For many of the current approaches, we have little knowledge about how 
these questions must be answered. The goal of the workshop is to bring 
together researchers who develop and use various kinds of spatial 
representations in mobile robots and make them think about how to answer 
the above questions for their approaches. A secondary goal is to provide 
a forum on which various kinds of representations can be compared.

Workshop Actuality and Target Audience:
======================================= 
Currently, various successful, though specialized applications of
mobile robots (RHINO, etc.) are known. However, improved adaptive
spatial representations will be needed to build service robots for
more complex tasks in more complex environments. We consider such
representations an essential step for making progress towards this
goal. Thus, the target audience includes 
- researchers working in mobile robots, especially map building, 
  spatial modelling, navigation, object manipulation, and human-robot
  interaction, 
- AI researchers working on topological and other symbolic methods, 
  metric and probabilistic representations, and uncertainty. 
The workshop also appeals to researchers that have 
- studied spatial data structures in CAD, GIS, and image processing or
- studied spatial representations in biological systems 
and are now applying their models in robotic systems.

Preliminary Workshop Agenda:
============================
Depending on submissions, available time, and IJCAI constraints on the
schedule, we plan four to five sessions, each one will most likely be 
centered around one of the following themes:
- CAD/GIS-Inspired Representations (Frank/?Samet)
- Topological Representations (Kuipers/Cohn/Nebel)
- Metric and Probabilistic Representations (Burgard/?Kaelbling)
- Biologically-Inspired Representations (Recce/Tani/?Mallot)

In each session (90 minutes) covering one of the above four
approaches, we plan to implement the following session program:
- Invited survey talk by an experienced research scientist (25+5 min)
- Two short talks selected from the workshop paper submissions (10+5 min)
- Two to three rebutting/commenting statements by representatives 
	from the other approaches (5 min each)
- Session discussion (15 min), moderated by session chair
A general discussion session (45 to 60 minutes) will try to 
summarize results, draw conclusions, and define future activities,
like workshops, definition of benchmark problems, and others.

Tentative Attendees:
====================
(will be collected from response after announcements of various 
 news groups and lists:
	comp.robotics
	comp.ai
	comp.ai.neural-nets
	comp.ai.nlang-know-rep
	connectionists list
	hybrid list
 Please add any relevant news group or list)
	
Workshop Organizing Committee:
=============================
* To be confirmed.

Andrew Frank (GIS/CAD representations)
Bernhard Nebel (AI, relational/topological representations)
Anthony Cohn (AI, relational/topological representations)
Gerhard Kraetzschmar (chair) (AI, robotics, hybrid representations)
Benjamin Kuipers (AI, robotics, hybrid representations)
Michael Beetz (AI, robotics, metric/probabilistic representations)
Wolfram Burgard (AI, robotics, metric/probabilistic representations)
Gunther Palm (neuroscience, neural networks)
Michael Recce (neuroscience, biologically-inspired robotics)
Jun Tani (biologically-inspired robotics, dynamics) 

*Leslie Kaelbling
*Hanspeter Mallot
*Hanan Samet

Primary Contact:
===============
Gerhard K. Kraetzschmar
University of Ulm, Neural Information Processing
James-Franck-Ring, 89081 Ulm, Germany
Fon: +49-731-50-24155
Fax: +49-731-50-24155
Net: gkk at acm.org  or gkk at neuro.informatik.uni-ulm.de