A Critical Analysis of the Effects of Organization on Performance and Robustness of Multiagent Systems

Michael Schillo
German Research Center for AI (DFKI)

Friday, 31st May 2002
0930 - 1100

RMIT University Function Room (near Kaleide Theatre)
Building 8, level 2, RMIT
Swanston Street, Melbourne.

Abstract:

In task-assignment multiagent systems we assume agents to engage in interaction with other agents to distribute tasks according to costs, competency, maybe even task load. As a consequence, a complicated web of external commitments between the agents and a number of commitments internal to the agents are created. Although we can expect rational agents to only make new commitments as long as they fit in with previous ones, we have to face the fact that in dynamic environments agents can fail (due to loss of communication, failure of computing devices etc.). Such failure will leave other agents with an important problem: they may have committed to other agents to do a task which they assigned to a failling agent. In this case we have to provide mechanisms to deal with the possibility that the agent cannot do the task, as this would cause an internal resource conflict. Not doing the task will result in an external commitment conflict with the agent which issued the task.

We deal with this interesting aspect of multiagent robustness, by showing how different organizational forms can help to reassign tasks under these circumstances. By using several performance measures, we discuss their advantages and disadvantages for this kind of conflict resolution.

Biosketch:

Michael Schillo received an MSc in natural language processing from the University of Leeds, UK and holds a Diploma in informatics from Saarland University, Germany. He is currently affiliated with the German Research Center for AI (DFKI), where he is pursueing a PhD on robustness in multiagent systems.