A Case Reflective Negotiation Model
Project Award Date: 06-16-1999
Teams of agents representing independent, and often competitive, resource constraints must negotiate with one another to ensure that good enough solutions are provided soon enough to be useful. Negotiations concerning a single resource allocation are challenging enough, but a real solution to the real problem must consider how to conduct negotiations and allocate multiple resources to multiple consumers under a wide range of constraints. The problem is further complicated by the fact that while the teams are autonomous, they are not independent. This is due to the interaction of allocation problems with the consumption of time and computer resources during the negotiation process (which can affect the resources available to consumers).
This work uses case-based reasoning (CBR) to recall an appropriate negotiation strategy that the program knows has worked in the past, and this strategy defines the evaluation metrics, and the real-time constraints. The actual negotiating process is performed in a heuristic fashion and guided by CBR and utility theory. Multivalued-utility theory relates satisfaction and preference criteria to the possible negotiation strategies and selects the one that optimizes these criteria while minimizing risky behavior by the autonomous negotiating agent.
We also use a real-time system presenting a sophisticated programming model enabling reflective and adaptive behavior on the part of the agents. Reflective behavior refers to the action of the agents, which must be controlled, in part, in reaction to system conditions that they themselves change. Support for reflection enables teams of negotiating agents to monitor and control their own progress as well as to adapt their future behavior in response to previous performance.
Another aspect of support for reflection and adaptation takes the form in an adaptable feedback profile. The agent specifies the type of information it requires, which the system then provides. The reflective interface also supports negotiation among different teams for computational resources.
Faculty Investigator(s): Costas Tsatsoulis (PI), Douglas Niehaus, Jerry James, Jerry James (PI)
Student Investigator(s): Jacob Woltersdorf, Huseyin Sevay, Michael Frisbie, Praveen Sirivolu, Michele VanDyne, Kalyana Koka, Ping Yang, Brad Johnsmeyer, Chia-chi Chang, Dushyanth Nataraj
Primary Sponsor(s): U.S. Air Force