Computationally Adjustable Autonomy
Henry H. Hexmoor, Brian McLaughlan · 2001
Abstra t. Reasoning about autonomy is an integral omponent of ollaboration among omputational units of distributed systems. This paper introdu es an agent-level algorithm that allows an agent to ontinuously update its autonomy with respe t to re urring asyn hronous problems with the aim of system-wide ollaboration e ien y. The algorithm is demonstrated in a relevant s enario involving NASA spa e station-based Personal Satellite Assistants, whi h an handle dynami situation management that frustrates global ollaboration proto ols. Key words. Agents, Autonomy, portable satellite assistant. 1. Introdu tion. Computer- ontrolled systems feature prominently in large-s ale proje ts urrently un-der development by the military, ommer ial, and s ienti agen ies. Examples of these proje ts in lude the US military's Network-Centri Warfare do trine, IBM's Autonomi Computing initiative, and NASA's spa e station proje t. As these systems have in reased in omplexity, self-governing omponents have ome to feature prominently in their design and ontrol. This hange in paradigm from dire t human ontrol to indire t human oversight has for ed designers to address issues involving the autonomy of these sub-systems. Autonomy is dened and used in multi-agent system resear h [6, 7, 11, 12, 13 ℄ and other dis iplines in luding so iology [10 ℄ and philosophy [14, 15℄. It is important in multiagent intera tions sin e it relates the abilities of an agent to its freedoms and hoi es. The understanding and quanti ation of an agent's autonomy is required