Independent Validation of an AI-Based Inertial Measurement Unit-Based Odometry Solution for Localization of Small Uncrewed Aircraft Systems

Le résumé fourni par la source

Dead reckoning using inertial measurement units (IMUs) can be improved by leveraging artificial intelligence/machine learning (AI/ML) approaches using a process called inertial odometry (IO). IO can be used independently or in conjunction with other sensing modalities to improve navigation solutions in situations where global navigation satellite systems (GNSS) is either degraded or denied. Many IO-based solutions presented in literature are pedestrian-based algorithms which fail to generalize to uncrewed aircraft systems (UAS) due in part to the highly dynamic and nonlinear motion of flight. However, due to the rising use of UAS, the number of IO algorithms presented in literature designed for UAS has increased. Though these algorithms have shown state-of-the-art results consistent with many other IO approaches proposed in literature, they have (1) not been independently validated, and (2) only trained and tested either in the simulation or indoor environment. Current knowledge indicates that the capability and limitation of IO-based navigation solutions for use on UAS platforms in outdoor environments is unknown. This paper presents an independent validation of one such UAS, IO-based algorithm by evaluating its performance on a dataset captured outdoors. Furthermore, this paper outlines the data collection procedure and post processing routine required for a custom hexarotor multicopter equipped with an IMU capable of logging high-speed data (Field Ops Remote Tracker, PNI Sensor, Windsor, CA) and GNSS with real-time kinematic (RTK) corrections for ground-truth positional data. Preliminary results indicate network sensitivity to heading direction and velocity, performing best when heading aligns with course and when trained on a wide range of flight speeds.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.