Tracking of large crowds with a swarm of aerial robots

Searching for and tracking multiple moving targets has many civilian applications including search and rescue in disaster scenarios, environmental monitoring, air traffic control and surveillance. The use of multi-robot systems is advantageous in these scenarios due to their parallelism and distributed sensing as well as their ability to dynamically adapt their spatial distribution according to target movements. Swarm robotic systems in particular are suitable for real-world applications due to their high scalability, flexibility and robustness. This work considers a realistic scenario a swarm of aerial robots tracking a large crowd in an urban environment. Since people in a crowd or animals in a herd tend to move towards common destinations collaboratively, it is intuitive that it would be more efficient to track clusters of targets in such scenarios. This thesis introduces a distributed search and tracking algorithm where the robots first identify target clusters and then decide on a cluster to follow. Through numerical experiments using a simulated evacuation scenario, it is demonstrated that the algorithm is able to successfully track large crowd groups without using any prior information about their positions or movements. For two very different simulated crowd patterns the algorithm demonstrated equally good performance values (over 80% coverage over time), indicating the algorithm's ability to handle variations in the target movements. The algorithm also showed good performance (over 70% coverage over time) over a range of different parameter values which showed that the algorithm can run on different hardware configurations.