Abstract
In multi-agent coordination, the bilateral negotiation is an important and essential mechanism for agents to resolve the conflicts and enhance the global performance. However, since agents are usually rational, namely, they are always self-interested and seeking for their maximal expected utility payoff, they will tend to be protecting their own profit and utility during the negotiation. In addition, since the autonomous agents might have to choose the proper concessions and alternatives based on various multi-attribute utility functions during negotiation under bounded number of negotiation messages, it may lead to very inefficient and prolong process before both agents can reach their joint agreement. We assume rational agents are cooperative negotiators under the bounded number of messages. Namely, they are motivated to reach compromised agreement within the time bound in order to get as close to optimal solution as possible. Since agents could observe and learn from other agents’ negotiation proposal, agents could adapt during negotiation in order to speed up the negotiation process. Therefore we incorporate a learning mechanism into agents during the negotiation by using a simple perceptron learning method. We show by experiments that the learning agents can reach their joint agreement much faster than non-learning agents. Besides, the learning agents could reach the joint-agreements that always lie on the Pareto optimal frontier.