AIVIO: Closed-loop, Object-relative Navigation of UAVs with AI-aided Visual Inertial Odometry

This research enables real-time object-relative navigation using only an IMU and an RGB camera, ideal for payload-constrained robots. It employs a deep learning-based object pose estimator trained solely on synthetic data and optimized for companion board deployment. By fusing object-relative pose measurements with IMU data, the system achieves precise localization for tasks like power pole inspection. Real-world experiments validate its performance, with an example flight shown in the supplementary video.