GeNCA : un modèle général de négociation de contrats entre agents. (GeNCA: a general contract-based negotiation between agents model)

When numerous agents interact, conflicts may arise. To solve them, different ways canbe used such as cooperation, voting systems or negotiation. The one which interests us in thisthesis is contract-based negotiation over resources. The aim of our work is to conceive a general negotiation model (called GeNCA : GenericNegotiation of Contracts API), and to give an implementation of it. This allows a user wishing todevelop a negotiation application not to have to do the whole job but to have a model that willfacilitate his work. In order to conceive such a model, we begin by studying the different types of negotiationmost used. The analysis of these negotiations leads to the collect of their common points, andprovides us a basis to conceive our model. Thanks to the collect of these common points andthe study of the progress of these negotiations, we show that it is possible to conceive a generalnegotiation protocol, parameterable, which offer the possibility to make counter-proposals andwhich can describe the progress of a particular negotiation. We propose a general negotiation model using this protocol and having a management ofnegotiations mechanism, which allows to negotiate contracts on disjoint sets of resources inparallel, and to negotiate contracts having conflicts on resources sequentially. This model alsoallows to automatically renegotiate contracts that cannot be met any longer. Our proposition is based on a three-level architecture, that separates the communicationpart between agents, the negotiation part and the negotiation strategy part of an application.As a matter of fact, the way agents communicate doesn’t play a role in the way negotiationis made, and different communication ways can be used in a same application executed ondifferent environments. We equally show that it is important to separate the negotiation strategy from the two otherlevels, to allow a user to choose which negotiation strategy he will use without disturbing theremaining of the application. Moreover, the negotiation strategy is intrinsically linked to thenegotiation application, and it is obvious that negotiating a ton of potatoes is not the same asnegotiating a slot-time for an appointment, nor the same as negotiating the exclusive use of ashared resource for an hour. Our model has been implemented by a Java API also called GeNCA. In order to validateour model, we used our API to achieve different applications such as an auction system, anappointment taking system and a system to negotiate the choice of a restaurant for a commontrip.