Bertrand BENOIT Pages - Reactive multi-agents systems

Parameters adaptation into reactive multi-agents systems: application in mobile robotic

Domain
Artificial Intelligence
Context
Researsh's works in school period

Period
February 2004 - June 2004
State
Finished
Language(s)
Java

Link(s)
Report - Support - Open source project

Description
I have performed this work during a training period of Postgraduate Degree in Computer Science (research) within the MAIA team, supervised by Vincent Chevrier.

Subject: (It's recommended to read the Scientific context before.)

The aim of the work initially consisted in studying various methods suggested in the literature to answer the problem of the parameter's tuning.
Then, it was awaited a proposal specific to the model developed into the team, allowing an easy adaptation of the parameters when the experimental conditions change, with a continuous adaptation accorded with the system's evolution, as ambition.

My research leads me to analyze various related research works in order to observe the potential existence of the same problem in more or less different context, and the solutions considered. Thus, the use of an optimisation method seemed relevant to achieve the laid down goals; it's why we defined a protocol of resolution implementing - amongst other things - such a method. In the instantiation that we have made, in order to use it with the works of Franck Gechter, we used the descent of gradient defined by Laurent Jeanpierre. The results obtained offer interesting brief replies.

I place at the disposal my report of Postgraduate Degree in Computer Science (research) (in french) which offers a good global vision of the whole of my work, as my support of presentation (in french) which comprises some additional results.

Remark(s)
During my work, I have realized a Parameters Tuning Unity (PTU) which initially adapted the parameters of the reactive multi-agents system conceived and implemented by Franck Gechter, thanks to messages' exchanges between the various modules of the software architecture.
In order to make evolve the Parameters Tuning Unity and widen the range of systems being able to profit from it, while allowing the interested people to take part in his evolution, I have created a dedicated project Open source. You can find, on this site, a certain number of information concerning the Parameters Tuning Unity, and to reach the site of the project on Source Forge.

If you wish to test the Parameters Tuning Unity with the system created by Franck Gechter, you must contact him to obtain its source code.

Installation
To use the Parameters Tuning Unity, it is enough for you to have version 1.2 (or later) of the JRE (work was carried out with version 1.4).

Scientific context
   A reactive multi-agent system is a system characterized on the level of the agents by forms of behaviors and by simple interactions which allow the observation of complex collective ownerships. Such systems have several assets: simplicity, robustness, adaptivity, reactivity and flexibility. They thus offer an attractive prospect for the collective resolution of problems.

   Nevertheless, these systems still actually suffer of an important disadvantage: their construction and their exploitation in a particular application require to tune the parameters of the subjacent model by experiments for it to show, with honourable performances, the necessary properties.

   In the MAIA team, it has been proposed a reactive model of behavior and interaction based on the physical principle of attraction and repulsion forces. The team has applied this original model successfully to a task of localization in mobile robotics.

   This model and its application to the localisation provides attractive results and shows that we can resolve collectively this complex problem with very simple individual behaviour.
Nevertheless, there are still some issues when it is necessary to adapt the model's parameters with (quasi-) automatic methods when the utilization's context changes. For example, speed of the moving targets, type of trajectory, etc.

References
Van Parunak, H. and Brueckner, S. : "Entropy and Self-Organization in Multi-Agent Systems", International Conference on Autonomous Agents, 2001.

Van Parunak, H. : "'Go to the Ant': Engineering Principles from Natural Agent Systems", Annals of Operations Research 75:69-101, 1997.

Bonabeau, E. and Dorigo, M. and Theraulaz, G. : "Swarm Intelligence", Oxford University Press, 1999.

Gechter, Franck : "Modèle multi-agents réactifs pour la fusion de percepts, application à la localisation et au suivi en robotique mobile", thesis of UHP NANCY I, 2003.

Jeanpierre, Laurent : "Apprentissage et adaptation pour la modélisation stochastique de systèmes dynamiques rééls", thesis of UHP NANCY I, 2002.