Effects of Dynamically Weighting Autonomous Rules in an Unmanned Aircraft System (UAS) Flocking Model.

摘要

Within the U.S. military, senior decision-makers and researchers alike have postulated that vast improvements could be made to current Unmanned Aircraft Systems (UAS) Concepts of Operation through inclusion of autonomous flocking. Myriad methods of implementation and desirable mission sets for this technology have been identified in the literature; however, this thesis posits that specific missions and behaviors are best suited for autonomous military flocking implementations. Adding to Craig Reynolds' basic theory that three naturally observed rules can be used as building blocks for simulating flocking behavior, new rules are proposed and defined in the development of an autonomous flocking UAS model. Simulation validates that missions of military utility can be accomplished in this method through incorporation of dynamic event- and time-based rule weights. Additionally, a methodology is proposed and demonstrated that iteratively improves simulated mission effectiveness. Quantitative analysis is presented on data from 570 simulation runs, which verifies the hypothesis that iterative changes to rule parameters and weights demonstrate significant improvement over baseline performance. For a 36 square mile scenario, results show a 100% increase in finding targets, a 40.2% reduction in time to find a target, a 4.5% increase in area coverage, with a 0% attribution rate due to collisions and near misses.