Multi-Agent Negotiation using Combinatorial Auctions with Precedence Constraints

We present a system for multi-agent contract negotiation, implemented as an auctionbased market architecture called MAGNET. A principal feature of MAGNET is support for negotiation of contracts based on temporal and precedence constraints. We propose using an extended combinatorial auction paradigm to support these negotiations. A critical component of the agent negotiation process in a MAGNET system is the ability of a customer to evaluate the bids of competing suppliers. Winner determination in standard combinatorial auctions is known to be difficult, and the problem is further complicated by the addition of temporal constraints and a requirement to complete the winner-determination process within a hard deadline. We introduce two approaches to the extended winner determination problem. One is based on Integer Programming, and the other is a flexible, multi-criterion, anytime bid evaluator based on a simulated annealing framework. We evaluate the performance of both approaches and show how performance data can be used in the agent’s deliberation-scheduling process. The results show that coarse problem-size metrics can be effectively used to predict winner-determination processing time.