Author: Ankit Agarwal

Publisher:

Published: 2020

Total Pages:

ISBN-13:

Heightened public interest in Unmanned Aerial Systems (UAS) has led recently to a rapid increase in both the number and diversity of small- to medium-sized vehicles in the public airspace. With many of these UAS boasting autonomous capabilities such as hands-free flying and obstacle avoidance, safe and accurate autonomous localization and navigation remains critically important. Various technologies have been developed to solve the problem of accurate localization in an unknown airspace, but highly accurate vision-based navigation solutions continue to see rapid development due to the added challenges posed by indoor navigation. Namely, the lack of a reliable GPS connection in indoor environments proves challenging for precise maneuvering, and many of the highest-fidelity alternatives to GPS-based localization are heavy, expensive, and difficult to implement. Growing consumer and commercial adoption of Virtual and Augmented Reality technologies has led to a sharp increase in the number of compact localization solutions available to the public, and the capabilities of these devices conveniently make them choice candidates in solving the challenges of accurate indoor navigation. In the present study, a UAS navigation solution using the Intel RealSense T265, a commercially available Visual-Inertial Odometry (VIO) device, is developed and presented for the purpose of characterizing indoor localization performance. The goal of the study is to determine whether the localization fidelity of a compact and inexpensive VIO solution is sufficiently high to support safe and reliable autonomy of small indoor aerial vehicles. Position and heading data from the T265 are analyzed in their raw form and also after correction using an Extended Kalman Filter (EKF). These data are gathered by way of a hand-carry test, and are compared to ground truth measurements obtained via a Vicon motion capture system. Additionally, a closed-loop flight test is performed outside of a motion capture room for concept validation purposes and to evaluate the convergence and command tracking capability of the EKF-based navigation system. Results from hand-carry testing examined both the raw data from the T265 and the combined data using the EKF. Localization estimates from the device gathered immediately after initialization are highly inaccurate, but the raw data improves significantly as the VIO device continues to operate and gather information about its environment. The device may indeed prove sufficiently accurate for precision maneuvering applications, but only once it has been running for some time. These findings also suggest that the device may perform well when combined with additional sensors (such as LiDAR) that can "correct" the initial pose estimates and reduce the time required to provide an accurate solution. Further localization improvements may also be achievable with varied software configurations. The performance of the Extended Kalman Filter during the closed-loop flight is also evaluated, and while the EKF does not significantly improve position estimates while the raw device data is still inaccurate, it shows smoothing of noisy T265 measurements and generally precise trajectory following capabilities. Future work to extend this characterization shall involve testing the performance of the device across varying flight envelopes, and especially for longer durations.