Positioning and Localization Sensor Data Sets for Aerial Autonomy Research

Summary

With the increasing share of airspace taken up by unmanned aerial systems (UAS) and small UAS (sUAS), autonomous planning and navigation algorithms are becoming crucial for safe airspace integration and beyond visual line of sight (BVLOS) operations. The development and validation of such algorithms require access to large-scale, multimodal datasets that capture the complexities of aerial navigation. While such benchmark datasets like the KITTI exist, they were primarily designed for ground-based navigation. Equivalent datasets for aerial navigation remain limited.

To address this gap, we present a multimodal dataset designed to serve as a benchmark for UAS and sUAS navigation and research. The dataset has been generated using a custom-built sensor pod that integrates a 64-beam LiDAR, a Honeywell n580 inertial navigation system (INS), a downward-facing laser range finder, and a pair of electro-optical (EO) cameras in a fixed-baseline stereo configuration. The dataset currently contains synchronized data from different flight paths over a crop field. The dataset is stored in standard formats (rosbags, CSV and image sequences) to provide compatibility with a wide range of frameworks. This dataset will be continuously expanded as additional flights are conducted, allowing future releases to capture more diverse environments and provide richer resources.