UNMANNED AERIAL VEHICLE (UAV) SWARM TACTICS: AN AGENT-BASED SIMULATION AND MARKOV PROCESS ANALYSIS

Executive Summary
With the advent of unmanned combat aerial vehicles, the present day may represent one of the most significant periods of change in the battle space since the first use of aircraft in warfare.

Unmanned aerial vehicles (UAVs) are increasingly tasked with more missions, including those originally done by manned fighter aircraft. The advantages are obvious: first of all, they save lives because pilots are not exposed in the battle space. UAVs also have the advantage (either presently or in the future) of being cheaper in procurement, operation, maintenance, and necessary ground personnel. Such benefits do not stop at simply replacing manned aircraft with UAVs. The potential applications are even more significant, such as using autonomous swarms of UAVs as the next evolution of aerial warfare. The drawback of these emerging technologies is that potential adversaries also recognize these advantages of unmanned systems, which can then pose a threat to allied forces, including saturation attacks with large numbers of weapons and/or unmanned systems. One possible countermeasure to such threats is a defensive UAV swarm.

The concept of air-to-air combat with unmanned combat aerial vehicles is still in its nascent stages, and this work proposes and explores a novel future concept in which swarms of UAVs combat the adversary’s UAV swarm. Though there is substantial scientific literature available which address aspects of UAV swarms such as self-organization, UAV system measures, or multi-UAV search and detection approaches, this thesis uniquely investigates the development of tactics specifically addressing swarm versus swarm engagements. Even if tactics in manned air-to-air combat have previously been discussed during the last century of naval aviation, we assume that the employment of UAV swarms distinctly offers new tactics and merits revision of existing ones. The development of these swarm tactics motivates this thesis, with the goal of identifying influential factors and providing a foundation for follow-on research.

First we define a simple scenario that drives the presented work, motivated by ongoing proof-of-concept. We then develop an agent-based simulation in a bottom-up process, in which each UAV is treated as an agent in a network that forms the swarm. The agent is endowed with a small set of rules concerning the agent’s motion, combat behavior, and its interaction with teammates. The behavior set itself can be extended to the needs of other research questions in swarm vs. swarm engagements. By construction, there is no central controller that manages the agents; rather, all UAVs are assumed to be vehicles acting autonomously. Design of the agents’ rule sets leads to emergent behaviors of the collective, which are observed in the simulation and can be evaluated across varying input parameters defined in this thesis. Statistical design of experiments in the form of a central composite design is used to scan the factor space for the identified parameters. Among other parameters, we vary the initial positioning of a swarm, the weighting factor between preferences of the swarm for offensive and defensive behavior, and the maximum number of assignments of friendly UAVs to each detected enemy UAV. Logistic regression on the responses of the Monte Carlo simulation runs show that all identified factors are nominally important to explain the model behavior, though the weight factors for each respective swarm are determined to be the most significant ones.

Though the presented scenario ...