- Multi-Agent Simulation (present and past)
- Social Robotics (present and past)
- Adaptive Interfaces (past)
- Multi-Agent Problem Solving (past)
- Agent-Oriented Development (present and past)
My main goals are to combine and develop interdisciplinary studies and multi-agent tools for the implementation of complex intelligent artificial systems and the simulation of complex (real) systems.
Each one of the projects presented below takes place in a perspective of research where they will serve as a basis for the design of communities of software or hardware autonomous agents, immersed in the real world and interacting in a natural way with human beings. This is what I call Pervasive Intelligence. It will impose us to radically revise our current design processes and implement original ones, among which we are certain to find Participatory Design processes, involving designers, users and artificial agents in shared environments. These two topics are a the heart of my current researches, which are (and, for some of them, have been) roughly organized along the following categories:
[Warning : some of the projects presented below are linked to their own presentation page which, in some cases, is really incomplete. I apologize for this]
- Participatory Design of Agent-Based Simulation (2002-2005) - More to come... with the PhD Thesis of Minh Nguyen, master thesis of Paul Guyot, Research Network of the CNRS, etc. Most of these subjects will be expanded in the near future. Meanwhile, see this entry in the glossary, and related internships.
- The SimPop2 project (2002-2005) - I supervise, in collaboration with Denise Pumain and Lena Sanders, from the PARIS CNRS Laboratory, the PhD Thesis of Benoît Glisse on the simulation of urban growth in Europe these last 2000 years...
- The ArchiSim project (1997-2003) - I supervise, in collaboration with Stéphane Espié, at the INRETS, the PhD thesis of Sameh El Hadouaj on the simulation of realistic traffic used in conjunction with a car simulator.
- Intra-urban Migrations (1998-2003) - I supervise, in collaboration with Marie Piron, of the IRD (ex-ORSTOM, Geodes Laboratory), the PhD thesis of Diane Vanbergue, on the multi-scale simulation of intra-urban migrations in the town of Bogota.
- The CUBES project (1999-2003) - I supervise, in collaboration with Thierry Bouron, from France Telecom R&D, the phD thesis of Lamjed Ben Said, on the simulation of consumers' behaviors.
- The Manta project (1991-1999) -The application of EMF to the simulation of an ant colony. MANTA is certainly the most complete multi-agent simulation of social insects so far, since it covers all the individual activities and collective phenomenas that have been observed in two species of ants. The program and the sources (in Delphi 3 and now Java) are available and freely distributed.
- The Rivages project (1997-2001) - I have supervised the work of David Servat, at the IRD (ex-ORSTOM, Geodes Laboratory), in collaboration with Jean-Pierre Treuil and Edith Perrier, on the simulation of hydrological phenomenas using multi-agent techniques. One of the most exciting aims of this project is to implement multi-level simulations using agents at all levels of abstractions (water bowls, lakes, rivers, etc.).
- The Emergence of Language project (1997-2000) - I have supervised, in collaboration with Luc Steels, at Sony CSL Paris, the PhD thesis of Frédéric Kaplan on the emergence of language among populations of simulated agents (using selectionist techniques). When Artificial Life meets computational linguistics...
- The Microbes project (1999-2003) - It is a long-term study of the survival of a group of robots (10) in a standard environment. The robots will live for a minimum of three years, starting in Sept. 99, within the LIP6 laboratory, 24 hours a day, recharging themselves when necessary, using people as a resource and trying to cooperate for staying alive. This is probably the biggest experiment ever made with fully autonomous robots operating in a non-controlled environment. A dedicated website (still to be build) will allow people to follow our progresses in this exciting project.
- Interactive Situated Learning (2000-2003) - I have supervized the PhD Thesis of Louis Hugues and I now supervize that of Olivier Garcia (in collaboration with Thalès Airborne Systems) on learning by demonstration behaviors to a robot. More to come...
- MAAM (2000-2003) - I am responsible of some of the sofware AI developments in the MAAM project, lead by Dominique Duhaut. This is a fascinating subject and I really hope we'll be able to use self-organized techniques to achieve what is planned (see related interships here).
- Chain-Making robots (1993-1994 / 2003-..) - Simulated robots that are making chains. Heavily inspired on one hand by all the literature about ants' foraging and, on the other hand, by methodological concepts like incremental design, it is a funny way to test the reality of self-organization (in a simulated world). More to come soon (with a new implementation, and more detailed experiments).
- The Microb 1,2 & 3 projects (1995-1999) - Originally devoted to building teams of robots playing soccer against human players, it has moved in 1997 to building robots for the RoboCup and MiroSot competitions, and is now heading towards a more realistic study on the usefulness of robots' colonies in everyday life. The original project was called Microb (Making Intelligent Collective Robotics). Its first sequel, Microb2, aimed at building four different teams for the 98 RoboCup competition : small-size, middle-size, legged robots and simulation. Microb3 is the end of the trilogy, following the same goal for RoboCup'99.
- The Garden of Chances project (1995-1998)- I have supervised the work of Guillaume Hutzler (now at the University of Evry) on this artistical project initiated by the painter Bernard Gortais. By combining abstract art and artificial life techniques, they aim at providing new user interfaces for complex systems.
- The Data Gardens project (1998-2001) - A project targeted at building innovative user interfaces for hierarchising the information in dynamic, real-time and complex dataflows. I supervise, in collaboration with Bernard Hennion, at France Telecom R&D, the work of Valérie Renault on this subject.
- The Eco-Problem Solving framework (1991-1994) - EPS (Eco-Problem Solving) is an object-oriented, multi-agent problem-solving framework. It provides the programmer with a set of abstract classes that enable him to describe his problem as a set of reactive agents searching for satisfaction.
- Collective Problem Solving (1992-1998) - I've worked on some small CPS applications inspired from the collective activity of ants : collective sorting, foraging robots, chain-making robots (see above in the collective robotics section). Most of them were written in Smalltalk, but I recently discovered how powerful a simple language like StarLogo, from the MIT MediaLab, was. You can see some of my trials on the CPS/StarLogo page, as well as have a look at the StarJava project, a tentative port of StarLogo to Java I have supervised in 1999.
- The EMF framework - EMF (Etho-Modelling Framework) is an object-oriented, multi-agent simulation framework based on ethological concepts. It provides the programmer with a set of abstract classes for representing a 2D discrete environment and agents provided with a behavioral model described below and pheromonal-like means of communication. The abstract kernel is available in Smalltalk-80 (the Delphi & Java versions are provided with Manta, see above).
- The EMF agent behavioral architecture - It is a behavioral model close to that proposed by Konrad Lorenz (fixed actions patterns). It has been used in Manta, and, currently, for controlling the robots of Microbes.
- Ethological concepts applied to MAS design methodologies - Or: how to get rid of cognitive pyschology for designing intelligent systems ? ;-). Some insights about the usefulness of ethological concepts, methods and ideas applied to the design of artificial systems. Nothing really concrete, though, but some of these ideas (largely inspired by Dennett's work) have found their way in the previous projects.
- The Cassiopeia methodology - Or: how to be able to provide a software engineer with a kind of abstract levels similar to those employed in Object-oriented programming when dealing with multi-agent techniques. The question is complex (as is the model ;-).
- The Andromeda methodology - daughter of the previous one. A methodology that should help in clarifying the complex relationships between machine learning techniques and multi-agent systems. In collaboration with Jean-Daniel Zucker.