Implementing collective obligations in human-agent teams using kaos policies

Abstract. Obligations can apply to individuals, either severally or collectively. When applied severally, each individual or member of a team is independently responsible to fulfill the obligation. When applied collectively, it is the group as a whole that becomes responsible, with individual members sharing the obligation. In this paper, we present several variations of teamwork models involving the performance of collective obligations. Some of these rely heavily on a leader to ensure effective teamwork, whereas others leave much room for member autonomy. We strongly focus on the implementation of such models. We demonstrate how KAoS policies can be used to establish desired forms of cooperation through regulation of agent behavior. 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 enabled or disabled to regulate the degree of autonomy of the team members. We have implemented a prototype of a Mars-mission scenario that demonstrates varying results when applied across these different teamwork models