Sensible Agents: Demonstration of Dynamic Adaptive Autonomy

Multi-agent Systems (MAS) meld the research fields of Distributed Computing and Artificial Intelligence into a field called Distributed Artificial Intelligence (DAI). MAS fit nicely into domains that are naturally distributed and require automated reasoning to solve problems. Sensible Agents are one MAS designed for domains with a high level of dynamism and uncertainty. A central problem in MAS is finding the correct organizational structure for the agents (e.g. hierarchical, peer group, etc.) in which responsibilities to plan for and execute goals are allocated. In dynamic situations, it is unreasonable to expect a single organizational structure to be appropriate at all times. For proof of this, one needs only look at human management theory and practice. Human corporations often reorganize to face new environmental conditions. Sensible Agents attack this problem with Dynamic Adaptive Autonomy (DAA), which allows them to reorganize themselves during runtime to solve different problems in the face of a changing environment. Some specific research that has contributed to flexible, adaptive multi-agent coordination includes partial global planning (Durfee and Lesser, 1987), organizational selfdesign (Ishida et al., 1992), STEAM flexible teamwork (Tambe, 1997), and RETSINA matchmaking (Sycara and Pannu, 1998). However, these techniques do not specifically adapt agent planning-interaction styles.