We present a generic negotiation architecture that uses Multi-attribute Utility Theory (MAUT) principles to reach agreements that satisfy multiple interdependent objectives. The architecture is built by giving a constraint optimization formulation to the MAUT principles and by using a constraint optimization solver to find the best 'deals' from an agent's local perspective. These are then proposed to other agents via a second component that supports conversational interactions among agents. When received proposals are disjoint from what an agent can currently accept, we provide a systematic constraint relaxation protocol that allows agents to generate the next acceptable 'deal'. This protocol ensures that in the end the Pareto optimal deal will be found, if one exists. The approach is built on top of our Negotiation Engine, a generic architecture for coordination and negotiation that integrates local reasoning, in the form of propositional constraint optimization, with interaction, in the form of conversational exchanges. The system is fully operational, being currently used to automate negotiations in the electronic components domain.
展开▼
