The Acacia project focused on the link between agents and some classes of problems, with an application on traffic accident analysis where the experts - psychologists, infrastructure engineers, vehicle engineers - perform a task of modelling and of diagnosis.
We proposed a model of cognitive agent for knowledge acquisition [12], in order to model the human and artificial agents involved in the development of a knowledge-based system. This model of agent allows to describe individual aspects (resources, KADS model of expertise) and social aspects (e.g. interaction points, cooperation modes, conflict solving methods, organizational structure of a composite agent...). Knowledge acquisition then consists of identifying the adequate agents involved in the problem solving, of creating in the acquisition tool the corresponding artificial agents and to determine their characteristics. An agent can represent one or several experts or, even, a subpart of an expert. We proposed techniques of management of conflicts between such agents, based on different strategies [13,10] and we implemented an agent editor relying on our model of agent.
We applied our model of agent to an application of traffic accident analysis [9]. We showed how this model of agent, combined with an analysis of collective case studies performed cooperatively by experts can help to specify a cooperative expert system [11].
We studied how the library of generic models offered by CommonKADS can be exploited in order to build generic agents for different classes of problems (diagnosis, repair...). For construction of an expertise model, the knowledge engineer must identify the adequate problem class and problem solving methods. Then he can use our library of generic agents having partially known expertise models, according to the class of problem (for example, generic agents for diagnosis).
We proposed a model of cooperation among agents [33], we proposed an architecture of cognitive agent, implemented upon BB1, and we developed RACE, a platform based on facilitators agents, for realisation of multi-agent applications [32]. This platform comprises primitives of communication allowing to agents on distributed different sites to exchange information. We also offer an agent-oriented language CTALK and an infrastructure composed of specialized facilitators agents (communication, services, control, data sharing) enabling the cooperation inter-agents. This platform, implemented in C/C++, aims at enabling the rapid prototyping of multi-agents applications and their execution on distributed machines.