Adjusting the Autonomy of Collections of Agents in Multiagent Systems

Abstract

A topic of recent interest to researchers is designing multiagent systems that allow agents to reason about adjusting their autonomy, determining when and to whom control of decision making should be transferred (e.g. [2, 6])). What is lacking, however, is a method for integrating these individual adjustable autonomy algorithms into a cohesive solution for the delegation of tasks for the society. In this paper, we first discuss one approach that employs a central coordinating agent in order to not only adjust the levels of autonomy, but also ensure that there is coordination of this adjustment across all agents in the system. Fully autonomous agents elect to have their autonomy adjusted when faced with unexpected events that they are unable to resolve. The coordinating agent revokes the autonomy of other agents in the system, temporarily, in order to address these events. We discuss how this strategy for the adjustment of autonomy of agents is well suited for multiagent systems operating in soft real-time environments. We then present a model for agent-initiated adjustable autonomy that reasons not only about decision-making delegation, but also about interaction in order to make more informed decisions. Coordination of decision-making delegation amongst agents is addressed by a locking mechanism, while the provision for interaction allows run-time monitoring of the degree of bother of a potential resource/entity, resulting in possible refinements to the selection of entities to which decision making is delegated.