Fulfilling Collective Obligations in Human-Agent Teams using KAoS Policies


Dr. Jurriaan van Diggelen


Whereas an individual obligation describes what must be done by a particular individual, a collective obligation describes what must be done by a team of agents. For example, we may state that an astronaut team must collectively ensure safety of its team members, without specifying to which particular individual this obligation applies. This is very useful as carrying out a rescue operation to ensure safety is typically not something which can be done by an individual agent alone, but requires teamwork in order to be fulfilled. In this talk, I will discuss several organizational structures which realize that a team of agents carries out a collective obligation. On one end of the spectrum, the organization can be set up very hierarchically, making the team leader bear all responsibility of the collective obligation. On the other end of the spectrum, the organization contains no leaders at all, making each team member equally responsible for the collective obligations of its team. We have developed a set of KAoS policies which describe how agents must behave to establish to desired type of cooperation. Some of these policies concern invariant aspects of teamwork, such as how to behave when a leader is present, how to ensure that actions are properly coordinated, and how to delegate actions. Other policies can be switched on and off to regulate how hierarchical the team will be. We will show a simple prototype of a Mars-mission scenario which demonstrates the use of these KAoS policies.

Dr Jurriaan van Diggelen studied Cognitive Artificial Intelligence. He received his PhD degree on research on semantic interoperability in multi-agent systems at the Intelligence Systems group at Utrecht University, the Netherlands. Since 2007 he works as a Postdoc at Utrecht University and TNO Defence, Security & Safety. His current research concerns Ubiquitous computing, Context awareness, Crisis management, and Agent Organisations. His main interests are agent communication,
ontologies, ambient intelligence and the semantic web. From June 2st until September 2st, 2008, he works at IHMC on Agent Organisations and Human Machine Teams with Jeff Bradshaw and Matt Johnson. More information and a list of publications can be found on: http://people.cs.uu.nl/jurriaan/