=Paper=
{{Paper
|id=Vol-1136/paper4
|storemode=property
|title=Interactions between Levels in an Agent Oriented Model for Generalisation
|pdfUrl=https://ceur-ws.org/Vol-1136/paper4.pdf
|volume=Vol-1136
|dblpUrl=https://dblp.org/rec/conf/agile/Maudet13
}}
==Interactions between Levels in an Agent Oriented Model for Generalisation==
https://ceur-ws.org/Vol-1136/paper4.pdf
Interactions between Levels in an Agent Oriented Model
for Generalisation
Adrien Maudet
Université Paris-Est, IGN, COGIT, Saint-Mandé, France
adrien.maudet@ign.fr
Abstract. Generalisation is a complex operation of the mapping process seek-
ing to simplify geographic data. In order to carry out this process, algorithms
are used. Multi-agent systems are an approach to orchestrate the application of
these algorithms. Models were proposed in the literature, but some situation are
not automatically generalised in a satisfying way. Our hypothesis is that, if the
behaviour of the agents is described in a way that takes into account the organi-
sation of geographic objects in levels, we may solve these issues. Methods to
explore this hypothesis are introduced in this paper.
Keywords: Cartography, Automated generalisation, Multi-agent systems, Mul-
ti-levels.
1 Introduction
The PhD project exposed in this paper deals with cartographic generalisation. Carto-
graphic generalisation is the process that aims at decreasing the level of detail of a
vector database to fit a given scale and specific users’ needs in order to create a legi-
ble map. Indeed, when the scale decreases, the extent on the map to show information
about the portion of the real world in reality is smaller (Figure 1).
Fig. 1. Generalisation of a detailed map from vector data (copyright IGN).
This is a complex process as the generalisation of different objects (buildings, roads,
etc.) is constrained by other objects. Among the different approaches used to auto-
mate the process, there are agent-oriented models. These models allow solving some
specific issues. We assume that a more generic way to describe relations between
objects will help us to generalise automatically more cases. In this position paper we
first describe the motivation of the PhD project, then we give research questions and
finally we describe our methodology.
2 Motivation
The main idea of the agent oriented generalisation models is the fact than geographic
objects are describe as an agent (Weiss, 1999) that attempts to satisfy personal con-
straint (e.g. minimal size constraint for a building) and shared constraints (e.g. prox-
imity constraint between a building and a road symbol). These autonomous entities
will execute interactions with other ones to solve locally issues produced by scale
change. An interaction is an operation executed by one or more agent(s) taking into
account their situation (Figure 2).
Fig. 2. Simplified scenario of the execution of an interaction (copyright IGN).
Interactions may be hierarchical. In AGENT (Ruas 1999; Barrault et al. 2001), rela-
tionships are described between components (e.g. building) and a “meso” object (e.g.
urban block), and the meso agents are able to trigger its components and give orders
to them. Interactions may occur between objects from same levels too (e.g. two close
buildings interact when one moves away from the other) like in the CartACom model
(Duchêne 2004; Duchêne et al., 2012). The GAEL model (Gaffuri et al., 2008) intro-
duces new interaction types when decomposing objects into points: the interactions
between the decomposed object and its points, and the interaction between the points
themselves. The interactions in these three models are showed in Figure 3.
Fig. 3. Interactions between objects in existing models (copyright IGN).
Some problems remain unsolved by these models. Indeed, other types of relations
should be used to improve generalisation:
─ The inclusion relationship type occurs when an object is included in another object,
e.g. accidents or bus stations on a road (Jaara et al., 2012) (Figure 2b).
─ Diagonal interactions between objects of different levels, e.g. two buildings inter-
acting as a whole with other neighbouring objects (Figure 2a).
─ Objects involved in a hierarchy can both interact with their “parent” and other
objects in a same level, e.g. bus stations staying on a road when the shape of the
road is modified while preserving consistency with other bus stations (Figure 2b).
─ An object can be included in two hierarchical relationships and therefore needs to
make decisions taking into account these two relationships at the same time, which
is currently not well handled, e.g. a building belonging to two alignments (Figure
2c) or a bridge included both in a road and a river (Figure 2d).
We assume that a generic interaction model would help to solve these issues. A
model used in the simulation domain called PADAWAN (Picault and Mathieu, 2011)
has interesting properties for the issue (e.g. multi-level hierarchy, interaction-based
model).
Fig. 4. Instances of relation types not processed yet (copyright IGN).
3 Aims and research questions
The objective of the PhD is to define an agent-oriented generalisation process taking
into account relations between objects at different levels and to answer to those re-
search questions:
─ How can generalisation problems implying many levels of interaction be solved?
The main purpose of the PhD is to provide solutions in order to improve an auto-
matic generalisation process.
─ How to orchestrate interactions between agents? The order the agents are activated
and the order the algorithms are applied impacts the result of the generalisation. In
an agent modelling perspective, we want to provide a generic way of modelling the
agent behaviour in constraint solving problems.
4 Method
PADAWAN is identified as a potential basis to solve remaining questions on interac-
tions between agents in generalisation. First, I analysed similarities and differences
between this model and the existing ones, in order to adapt PADAWAN to the speci-
ficities of generalisation.
Then case study was defined: a map combining base topographic data and touristic
thematic information. Figure 3 gives an instance of the case study: a series of bends
can be schematised using a schematisation algorithm (Lecordix et al., 1997) that re-
moves some bends (Figure 3a and 3b). But when the bend has a proximity relation-
ship with a touristic point object (Figure 3c), how can this information be preserved?
Solutions to such problems will be proposed, and will be evaluated by means of ex-
periments.
From these specific cases I will analyse in a more generic way the relations be-
tween objects and then I try to propose a decision process to orchestrate the activation
of the agents. This process will have to be evaluated by means of experiments too.
Fig. 5. (a) They are 13 bends between the two junctions of the circled road (1:100k). (b) The
same portion gets only 8 bends (1:250k). (c) If we have to apply the same operation, how to
preserve bends in relation with other map elements (like the encircled church)? (copyright
IGN).
5 Conclusion and perspectives
First, we introduced agent approaches to automate generalisation process and their
limitations. Then, we expose our hypothesis: enhance the expression of interactions
may help us solve unsolved generalisation case. We identify some of the unsolved
case and their multi-levels aspects. Currently, the first results cover two aspects: the
adaptation of existing processes to the PADAWAN model and the study of specific
cases (Maudet et al. 2013). The next steps will be to give solutions to other specific
cases in order to propose a generic way to manage multi-level interactions.
6 References
1. Barrault M., Regnauld N., Duchêne C., Haire K., Baeijs C., Demazeau Y., Hardy P.,
Mackaness W., Ruas A. and Weibel R.: Integrating Multi-agent, Object-oriented, And Al-
gorithmic Techniques For Improved Automated Map Generalization. In: 20th International
Cartographic Conference, vol.3, 2110-2116 (2001).
2. Duchêne C.: Généralisation cartographique par agents communicants : le modèle CartA-
Com. PhD Thesis, Université Paris 6 – Pierre et Marie Curie (2004).
3. Duchêne C., Ruas A. and Cambier C.: The CartACom model: transforming cartographic
features into communicating agents for cartographic generalisation. International Journal
of Geographic Information Science, 26(9):1533–1562 (2012)
4. Gaffuri J., Duchêne C. and Ruas A.: Object-field relationships modelling in an agent-based
generalisation model. In: Proceedings of 11th ICA workshop on Generalisation and Multi-
ple Representation, Montpellier, France (2008)
5. Jaara K., Duchêne C. and Ruas A.: A model for preserving the consistency between topo-
graphic and thematic layers throughout data migration. In: 15th International Symposium
on Spatial Data Handling (SDH’12), Bonn, Germany (2012)
6. Lecordix F., Plazanet C. and Lagrange J.P.: A Platform for Research in Generalization:
Application to Caricature. GeoInformatica, 1(2), pp. 161-182 (1997)
7. Maudet A., Touya G., Duchêne C. and Picault S.: Improving multi-level interactions mod-
elling in a multi-agent generalisation model: first thoughts. In: Proceedings of 16th ICA
workshop on Generalisation and Multiple Representation, Dresden, Germany (2013)
8. Picault S. and Mathieu P.: An Interaction-Oriented Model for Multi-Scale Simulation. In:
22nd International Joint Conference on Artificial Intelligence (IJCAI/AAAI), Barcelona,
Spain, 332–337 (2011)
9. Ruas A.: Modèle de généralisation de données géographiques à base de contraintes et d'au-
tonomie. PhD Thesis, Université Marne La Vallée (1999)
10. Weiss G.: Multiagent systems: a modern approach to distributed artificial intelligence.
Cambridge, MA, USA: MIT Press (1999)