Aug/Sep 96 Travel Report
<
http://cdr.stanford.edu/people/petrie/Aug-96-travel-plan.html
>
This was a trip to attend ECAI-96 and continue some German
collaborations, most notably with the University of
Kaiserslautern. See the last trip
report. This work was in support of the DSC project.
11 Aug 96
8:25 SFO DEL 964
15:54 Atlanta
17:20 DEL 146 AA84
12 Aug 96
10:55 Budapest
Attending
ECAI-96 Workshop on Non-Standard Constraint Processing
Stayed at
Hotel Platanus
1087 Budapest Koenyves Kalman krt 44
Tel: (36-1) 210-2599,
(36-1) 210-2592,
(36-1) 133-6505
Tx: 22-2100, Fax: 210-4386
At this conference, I
presented the paper Combining
Constraint Propagation and Backtracking for Distributed
Engineering. I also gave a
live demo using Java applets as agent clients. Our approach
was unique among the other papers presented in that we gave a
framework and services for problem solving as opposed to
presenting a novel constraint satisfaction algorithm that
attempted to solve the problem. The paper that was closest to
ours was "Dynamic Meta Constraints" by Janet Vander
<jvanderlinden@brookes.ac.uk>. We are studying the
relationship of this technique to our decision-dependent
constaints.
At this conference, I was on the panel "What is an Agent?" of
the
Agent Theories, Architechtures, and Languages (ATAL-96)
Workshop. My position paper will be by Springer-Verlag in
the LNAI series under the title "Intelligent Agents III".
Note: the trip after this point was funded by my
German hosts and myself.
16 Aug 96
Train to Berlin
Working at the DaimlerBenz
Contact:
Dr. Kurt
Sundermeyer <sun@DBresearch-berlin.de>.
At this center, I gave a talk
on distributed engineering and agents.
I also saw demonstrations and reviewed technologies as follows.
- SyDeR (System Design for Reuse): A tool for the design of
hierarchical systems. The approach focuses on the reuse
of exisitng designs based on requirements and interfaces.
Blocks of design are assigned to engineers who are warned
of global effects of internal inconsistencies. Some ideas
are the use of a specific design model to inherit
rationales and deduce rationales from design moves.
Contact:
Helmuth Ritzer <ritzer@dbresearch-berlin.de>.
- DASEDIS (Development And Simulation Environment for
Distributed Intelligent Systems): with this environment,
systems of agents can be implemented and their collective
behavior studied. I saw a demonstration of automobile
trafic in which each car was an agent following
various autonoma rules. Another application I saw
was a shop floor in which pallets were collected
for transport. The demo would allow delays, breakdowns,
and new orders. One strategy being explored was a
contract net; another was 1rst bid 2nd price auction.
This project is an outgrowth of COSY, part of the
old ESPRIT Prometheus project. The shop floor
control
contact is Frank Feldkamp
<feldkamp@DBresearch-berlin.de>
.
The main previous contact
is Afsaneh Haddadi
<afsaneh@DBresearch-berlin.de> who has
since come to the
Palo Alto DB research office. Dr. Haddadi's research
has been on "strong" agents but now says that iteraction
protocols, perhaps compiled from strong theories, are
sufficient and important. Our project is built around
the Redux interaction theory and we hope to collaborate
with Dr. Haddadi while she is in Palo Alto.
- A similar unnamed system was described. It uses
distributed agents and constraint satisfaction. It is a
simulation of the body, paint, and assembly shop floor
DB plant in Stuttgart. Each tool/resource is an
agent. Planning proceeds by backward chaining from the
production goal. The search is roughly depth-first. At
each level, constraint propagation
and consolidation is performed. The constraint solver
uses Simplex to determine if a solution exists.
The OZ system from the DFKI performs finite element
constraint satisfaction to arrive at a solution.
Contact: Stefan Bussman
<bussman@DBresearch-berlin.de>.
- I also reviewed an unnamed formal configuration
system built by Rudolf Müller
<mueller@DBresearch-berlin.de> who has done
extensive formal work with constraints and
scheduling. We discovered that we were in surprising
agreement about tractiability and approaches. In fact,
I recognized the Redux model of goals and constraints
in Dr. Müller's formal model. This turns out to be
an indirect transfer from Helmuth Ritzer, who was
familiar with Redux from his Kaiserslautern work.
21 Aug 96
Train to Kaiserslautern
Working at the University of
Kaiserslautern
Invited speaker at
Coordinating Work Processes
Worked here remotely for three weeks.
Contact:
Frank Maurer <maurer@informatik.uni-kl.de>
Lodging:
Universitaet Gaestehaus Room 215
Phone: 0631/311515
While working at with this group, we reviewed the dependency
structures of Redux,
CoMoKit, and
Procura. While Procura is more dynamic than
CoMoKit, it is clear that there are certain directions we should
all pursue. One extension to Procura is guidance by a
pre-existing plan, perhaps developed in DesignRoadmap. OTOH,
CoMoKit needs to deal with inputs/outputs different from those
planned.
Discussions with Frank Maurer and Prof. Richter lead me to think
that two extensions might be made to Redux. The most important
is to support metadecisions to get information in
order to decide which of several alternatives is best. This
sort of activity is important but not represented by Redux.
Another extension is an approval cycle. Frank Maurer's
aplication is extremely concerned with authority and
correctness. Someone should determine whether the decision
of a given agent to accomplish a task really does accomplish the
task in an acceptable manner. This is an important consideration
and we worked on the dependencies in detail. Much of
this work should show up in the next revision of CoMoKit.
Contact: Barbara
Dellen
10 Sep 96
Train to Bonn
ab 15.02 SE 3169 an 16.03 Mannheim Hbf
ab 16.07 EC 112 an 18.07 Bonn
Lecture on Wednesday at the
GMD AI Colloquium
Contact:
Hans Voss
<hvoss@mail.gmd.de>
also Gloria Mark at <gloria.mark@gmd.de>
02241 342 808
I gave my agents' definition talk and had the first spirited
debate with an Tom Gordon who insisted that legal
authority to act in someone's behalf was the best definition,
but he still had not yet formalized this notion to
exclude any particular set of computer programs.
The projects I reviewed were:
- ZENO:
this is a collaborative decision-making support system
based upon a system of argumentation and logic
considerably more advanced than the usual GiBis model.
Contact:
Tom Gordon <thomas.gordon@mail.gmd.de>
-
FABEL: This is an interesting attempt to combine
case-based reasoning (CBR) and expert systems.
I was shown a working
example of an architectural design.
Contact: Carl-Helmut Coulon <coulon@gmd.de>
That concluded the working portion of this trip.