Towards Fully-Autonomous Ultralight Drones

Autonomous drones have emerged as an exciting new technology which could revolutionize infrastructure inspection, military reconnaissance, and police surveillance. However, the vast majority of today's platforms are heavy, costly, and difficult to operate. This restricts them from use in many mission settings, such as in densely populated environments, where government regulation forbids autonomous opera- tion of heavy drones near people. Much of this weight comes from the onboard compute resources required for these drones to run the critical computer vision algorithms that provide situational awareness. In this dissertation, I show how autonomy can be induced on lightweight drones using edge computing, offloading high compute jobs to a network-proximal server. I demonstrate how this technique can lead to autonomous aircraft that fly much closer to the FAAs regulatory limits at acceptable performance cost. I also reveal a new operating system designed to unify the disparate landscape of drones under a single, easy-to-program API. I show how this can be leveraged to create heterogeneous collaborative drone swarms on commercial off-the-shelf hardware