Text of Ottawa Presentations
- Sylvie Lamy, Anne Ruas (Institut Géographique
National, F)
- Yves Demazeau (Institut National Polytechnique de Grenoble,
F)
- Mike Jackson (Laser-Scan, UK)
- William Mackaness (University of Edinburgh, UK)
- Robert Weibel (University of Zurich, CH)
- This presentation provides an overview of the AGENT project.
Specific research results can be found in publications by
individual members of the five partner institutions of the AGENT
consortium.
- The R&D background of the partner institutions has also
been absorbed in the AGENT project and prototype.
- The contributions of the other members of the AGENT
consortium to the progress of the project are gratefully
acknowledged.
- Focus on topographic cartography (IGN) and Internet
- Large background:
- Many years R&D in generalization (IGN, UNI-ZH,
UNI-ED)
- Lamps2/Generalizer: Interactive generalization module
- Selection, displacement, typification, aggregation,
exaggeration, ...
- Traditionally is labor-intensive, tedious task involving
re-compilation
- Production agencies have style manuals
- Different cartographers give different results
- Computers have helped, but still manual
- There is still a need paper for products at different
scales.
- Generalization is increasingly important for "mapping on
demand", particularly for internet mapping
- AGENT =
- Automated GEneralization - New
Technology.
- 21 man years over a 3 year period
- Objectives of the project:
- Better algorithms
- Better measures and assessment tools
- Better constraints
- Self-evaluating methods
- Strategies and orchestration
- Better decisions via multi-agent systems
- Expected results: prototype implementation and two
demonstrators (and of course publications...)
|
Generalisation |
OODB |
MAS |
| Uni Zurich |
+++ |
|
|
| Uni Edinburgh |
+++ |
|
|
| IGN Paris |
+ |
+ |
+ |
| Laser-Scan |
+ |
++ |
|
| INP Grenoble |
+ |
|
++ |
- Real or virtual entity
- Evolves in an environment
- Is able to perceive this environment
- Is able to act in this environment
- Can communicate with other agents
- Exhibits, as a consequence of the above, an autonomous
behaviour
- Reactive (simple) or cognitive (more complex)
- A set of agents which interact in a common environment.
- Focus on the collaborative resolution of global problems by a
set of distributed entities.
- Agents attempt to satisfy their own local goals as well as
the collaborative global goals.
- Status of generalization in consortium:
- already object-oriented approach (Gothic/Lamps2)
- Why MAS for generalization?
- capturing the holistic aspects
- better catering to the multi-level problem
- allowing more dynamic choices
- enabling autonomy vs. full user control
- Why generalization for MAS?
- exploring a real-size technology transfer
- enhancing MAS methodology
INPUT: perception stimuli
- PC perception capabilities (first hand)
- CC communication capabilities (second hand)
- K knowledge
- RC reasoning capabilities
- DC decision capabilities
- EC execution capabilites
- G goals
- P possible plans
- C choices
- A actions
OUTPUT: actions on agents or environment
- Recursive subdivision and creation of agents
- Includes spatial processing: Partitioning,clustering,...
Generalization of urban situations needs knowledge about
neighbors
- Size constraint: Building should have a
minimum size to be legible. This size depends on scale and symbol
thresholds.
- Width constraint: The width within a
building should be wide enough to avoid conflict.
- Accuracy constraint: The absolute position
of a building should be preserved as much as possible.
- Orientation constraint: The orientation of a
building should be preserved.
- And others, such as shape constraints, etc.
- I should be big enough
- I should be wide enough
- I should try to preserve my shape
- I should not be too far to my initial position
- I should try to preserve my original orientation
| Size: |
> 300m2 |
| Internal_width: |
> 20m |
| Hausdorff_dist: |
< 20m (from initial position) |
| Main_orientation: |
in [Initial_main_orient - 0.1, Initial_main_orient +
0.1] |
Constraints / Goals
- Ensure min. distance
- Maintain alignment
- Maintain shared primitives
- ...
Actions
- Displace
- Amalgamate
- Eliminate
- About half-way, and on schedule
- Independent generalisation (micro-agents) is implemented and
currently being refined
- Contextual generalisation (meso-agents) is being developed
now, nearing completion (version 2)
- Strategic generalisation (macro-agents) follows next
year
- Partnership is working well, and first results look
promising
Home Page:
http://agent.ign.fr