《F l e x i b l e social laws》.pdfVIP

F l e x i b l e s o c i a l l a w s * W i l l Brigg s and Dian e Coo k Department of Computer Science and Engineering University of Texas at Arlingto n Arlington , TX 76019 U . S . A . wbriggs@sirius.uta.edu, ocok@centauri.uta.edu Abstrac t Although communication is generally consid- ered to dominate over processing cost in dis- tributed systems, the problem of communica- tion cost in multiagent planning has not been sufficiently addressed. One method for re- ducing both communication cost and planning time is the use of social laws. Social laws, however, can be too restrictive, limiting sound- ness. Flexible social laws can enable multia- gent systems to reap the benefits of reduced Durfee, 1994] shows a model in which agents use deci- communication cost and planning time (except sion theory to decide when to communicate, and [Dur- in the worst case), without limiting soundness fee and Montgomery, 1991] reduces the cost of com- (although optimality may be degraded). By munication by sending abstract messages. Previous analyzing the performance, we show that this work on social laws [Moses and Tennenholtz, 1990; model can make multiagent planning exponen- 199l], that is, constraints on what actions an agent may tially more efficient without limiting its appli- take, has shown a method for reducing communication cability.