High-Level Control of UAV Swarms with RSSI Based Position Estimation

Abstract: Search and rescue operations can greatly benefit from the use of cooperative swarms of autonomous UAVs in order to investigate areas and collect information about the position of a missing person. In this thesis, UAV swarm algorithms are investigated where collisions are prevented both between agent pairs and between agents and static obstacles. The swarm consists of low-cost collaborative fixed-wing aircraft with communication constraints. A decentralized swarm behavior is first developed when the system is assumed to provide accurate positions of all aircraft. Further, the agents estimate their position by the use of RSSI measurements. All agents are equipped with communication devices and broadcast radio signals and measure the received signal strength in order to estimate the distance to other swarm members. These estimates are further used to develop a multilateration algorithm, where each agent estimates its own position by using distance estimates from a minimum of three nearby agents. By adding a dynamic model of the aircraft kinematics, a more accurate estimation is provided which takes account for false position estimates. The autonomous swarm is simulated in a 2-D environment in MATLAB. The agents make decisions in real-time, where their movements are controlled by potential fields and pheromone levels. Repulsive potentials are used to prevent collision and attractive potentials are applied to form a cluster of UAVs, such that all members stay within communication range. The swarm is also attracted to unexplored areas of the environment. When the true UAV positions are provided, the developed potential field algorithm did show promising results in terms of controlling the swarm. No collisions occurred between agent pairs or agents and obstacles. The agents did not go out of bounds and the swarm was robust as it was able to handle the loss of individual members. For the approach of RSSI based position estimates, further development of the swarm behavior was needed. The receiver sensitivity of the communication devices limits both the maximum distance between agents and their difference in roll angle. When individual failures occurred, or when an obstacle obstructed the path of the swarm, there was not always enough RSSI measurements available to perform the multilateration algorithm. In combination with the dynamic model of the aircraft kinematics, the resulting algorithm produced position estimates with a mean error of approximately 9 meters. No significant difference was found regarding the efficiency of scanning the area when the positions were estimated by RSSI values compared to when the positions were known. However, victims may go undetected when using estimated positions if the position error results in UAVs believing they have visited certain areas they have not yet scanned.