CloudMAV: Unlocking Range and Compute with Wifi Connectivity in Micro Aerial Vehicles

Abstract

Due to their small size, micro aerial vehicles (MAVs – under 100 g) are ideal for agile exploration of a range of environments, particularly indoor ones. This size, however, greatly limits their payload capacity and affects their sensing and computational resources. Recent work demonstrates remarkable progress miniaturizing sensors, compute, and navigation algorithms to fit onboard MAVs, but the tasks that they can accomplish remain fundamentally constrained by their small size. Scaling laws mean that state of the art robotics autonomy requires ever increasing computation which simply cannot be achieved onboard MAVs. Typical MAV offboard computation uses a low latency line-of-sight radio link, but the prospects for non-line-of-sight cloud-based links via Wifi or 5G are becoming increasingly attractive as wireless technology improves. In this thesis I present CloudMAV, a Wifi-enabled navigation stack for MAVs. CloudMAV is designed from the ground up to support a reliable control link and a low latency video stream for offboard monocular navigation. I demonstrate that a Wifi network provides sufficient bandwidth to support a video stream and control with moderate latency and throughput. Through the use of Wifi, CloudMAV is able to navigate wherever it has connection and can access virtually unlimited computational resources in the cloud.