Multi-Modality Localization of Autonomous Lighter-Than-Air UAVs in Real-Time

Advancements in sensing technologies for aerial robotics have supported vehicle operation across a diverse range of environments. While high-end sensors are relevant in many applications, there has been a growing interest in low-cost, attritable vehicles have seen increased use in hazardous scenarios. This vehicle type often operates in GPS-denied environments which makes localization more difficult. Regardless of environment, vehicles still require an understanding of their local position and orientation to perform mission-critical tasks such as path planning and routing. Current localization techniques rely heavily on noisy measurement data provided by internal measurement units (IMUs) deployed on the majority of aerial systems. This paper proposes an extended Kalman filter-based sensor fusion approach that combines a low-cost ten degree-of-freedom (10-DoF) IMU and camera with a lightweight, off-board computer vision pipeline designed so that a single compute platform can provide localization updates for a fleet of vehicles. The system is implemented on a small lighter-than-air vehicle used in the Defend the Republic aerial robotics competition. Experimental results in a competition environment demonstrate that the proposed approach achieves real-time localization with an average positional error of less than one meter, even when using extremely low-cost components.