Multi-agent implementation of asymmetric protocol for bilateral negotiations (extended abstract)

We present a practical implementation of a FIPA-compliant multiagent system, written in Java using the ABLE agent platform, in which the agents can negotiate the exchange of multi-attribute goods in reasonable time and with high efficiency. The architecture employs separate modules for messaging and decision logic. Negotiation protocols are expressed as conversation policies in XML and are highly reconfigurable; agents can also conduct a metanegotiation to decide which protocol to use. We believe this provides a flexible and powerful platform for both real-world negotiation of complex goods, and for research into protocol design and strategy design. We test the system using an asymmetric protocol to negotiate a two-attribute good, and using the simplest possible “ZeroIntelligence” (ZI) negotiating strategies. A key aspect of the protocol is a facility to obtain mutually beneficial modifications to an initial agreement. We find that the ZI agents reliably reach final agreements in a moderate number of rounds that are surprisingly close to Pareto-optimal contracts. These results, combined with ongoing studies of more sophisticated strategies, suggest that the protocol may be more generally useful in negotiating realistic goods with intelligent strategies.