Multi-Agent Simulation and Analysis of Surveillance Systems
Résumé
The Army is looking for ways to reduce the number of Soldiers exposed to the hostile war environment by using advanced technologies. These technologies will modernize the equipment the Army is using to conduct varying missions such as reconnaissance operations. The Combat Capabilities Development Command’s (CCDC) Army Research Laboratory (ARL) has been researching and developing multi-agent simulations of humans and UAVs for this purpose. The multi-agent simulation is a 3km x 3km field with 100 points of interest. The simulation's standard agents are an operator and three distinct UAVs: fixed-wings, flapping wings, and quadcopters. The performance metrics measured in the simulation are duration, the time it takes the UAVs to photograph and classify the imagery; noise, duration and decibel level of exposure at evenly spaced points throughout the field; and accuracy, the proportion of images the operators identify correctly. Within the simulation, varying the number of agents will have a measurable effect upon these performance metrics. To understand this effect, the team developed a full-factorial design of experiments (DOE). The human operators and fixed-wing unmanned aerial vehicles ranged between one to ten and the quadcopters and flapping wing ranged from zero to ten. Using the results of the DOE, we conducted a parametric analysis to determine the breakpoint for the metrics and developed an interactive R shiny application. The team conducted the parametric analysis by holding two of the factors constant and varying the other two. We determined that, after seven operators, accuracy did not have significant improvement, and in order to achieve an optimal accuracy rating, the number of UAVs should be less than 13. In terms of duration and noise, we determined that six quadcopters were the optimum number. After six quadcopters, there was no improvement in duration and the noise in the simulation reached possible detection levels. The team developed the interactive R shiny multi-attribute decision-making (MADM) optimization tool using Simple Additive Weighting (SAW). The purpose of the tool is to output the optimal combinations of human operators and UAVs based on how the user weighs these system-level attributes: of number operators, total number of UAVs, accuracy, noise, and duration. The project will facilitate answering the following issues for analysis: 1) Which of the four factors has the most significant impact on the performance metrics?, 2) How does the interaction between the factors affect the performance metrics?, and 3) What is the optimal mixture of human operators and UAV platforms?.
| Origine | Fichiers produits par l'(les) auteur(s) |
|---|---|
| Licence |