Negotiation in Semi-cooperative Agreement Problems

In this paper we introduce the Semi-Cooperative Extended Incremental Multiagent Agreement Problem with Preferences (SC-EIMAPP). In SC-EIMAPPs, variables arise over time. For each variable, a set of distributed agents gain utility for agreeing on an option to assign to the variable. We define semi-cooperative utility as an agent's privately owned preferences, discounted as negotiation time increases. SC-EIMAPPs reflect real world agreement problems, including meeting scheduling and task allocation.We analyze negotiation in SC-EIMAPPs theoretically. We note that agents necessarily reveal information about their own preferences and constraints as they negotiate agreements. We show how agents can use this limited and noisy information to learn to negotiate more effectively. We demonstrate our results experimentally.