Symbiotic Adaptive Multisimulation

Abstract

Systems characterized by non-linear interactions among diverse agents often exhibit emergent behavior that may be very different from what the initial conditions of these systems would suggest. Traditional simulation techniques that rely on accurate knowledge of these conditions typically fail in these cases. The goal of Symbiotic Adaptive Multisimulation (SAMS) is to enable robust decision making in real-time for these problems. Rather than rely on a single authoritative model, SAMS explores an ensemble of plausible models, which are individually flawed but collectively provide more insight than would be possible otherwise.

The insights derived from the model ensemble are used to improve the performance of the system under study. Likewise, as the system develops, observations of emerging conditions can be used to improve exploration of the model ensemble. In essence, a useful coevolution between the physical system and SAMS occurs. In this thesis, an outline of the core techniques of SAMS is provided. In addition, a parallel simulation application for the study of autonomous Unmanned Aerial Vehicle (UAV) teams was developed. Experimental results from this application are presented and their implications for further study are discussed.