Onboard UAV Navigation Utilizing Limited Sensing

Abstract

Physical autonomous systems have come a long way in recent decades, and are becoming more prevalent in everyday life due to rapid development in many supporting fields of science and engineering. One may choose to view some of these modern developments through a “robotics emulating nature” lens, with robotics advances appearing to reflect some elements of biology in their design. Examples include recent advances in artificial muscles, the field of computer vision leveraging biology-inspired “neural” networks, and Boston Dynamics designing the commercial robo-dog, Spot. With this biological lens in mind, I asked myself what I found to be an interesting question: how would we design an aerial robot that uses only nature-like sensors? By restricting myself to using only sensors with a biological analog, I would design an autonomous flight system relying on two types of sensors: an inertial measurement unit (IMU), and cameras. This restriction imposed a large challenge in basic state estimation and control, which was solved by designing and building a custom flight controller that combines both inertial and visual measurements. By leveraging a lightweight onboard computer and appropriate computer vision algorithms, the visual input was processed into meaningful estimates of relative motion, which was combined with the IMU motion derivative measurements. The final product is a stable flying platform that can perform station-keeping, so long as there is sufficient visual fidelity. The report that follows is a detailed log of the entire process, including information on the original project intent which had to be pivoted from due to time constraints. Much of the planning was guided by COVID-19 restrictions, and some shortcuts had to be made in order to piece together a physical system with limited equipment. Overall, the project serves to explore challenges in aerial robotics in a unique, biology-centric way, with a primary focus on accelerating my own education in robotics. Enjoy!