“…Specifically, unlike previous aerial SLAM datasets which collect only visual-inertial data, including EuRoC (Burri et al, 2016), UPenn Fast Flight (Sun et al, 2018), Zurich Urban MAV dataset (Majdik et al, 2017), UZH-FPV dataset (Delmerico et al, 2019), Blackbird dataset (Antonini et al, 2018), CLOUD dataset (Patel et al, 2020), and WildNav dataset (Gurgu et al, 2022), we collect hardware-synchronized LiDAR, camera, IMU, and GNSS data for multi-sensor fusion. Different from previous aerial SLAM datasets which contain LiDAR data but were collected in small-scale school campuses at low altitudes, including NTU-VIRAL dataset (Nguyen et al, 2022) and GRACO dataset (Zhu et al, 2023), our dataset includes 21 sequences, captured across a variety of environments including an aero-model airfield, an island, a rural town, and a valley at an altitude of higher than 80 m. In these sequences, the established flight speeds range from 3 m/s to 12 m/s, covering an area of from 94,000 m 2 to 577,000 m 2 in a single flight. Moreover, our sensor package, configured for downward-looking orientation as shown in Figure 2, imposes considerable challenges for SLAM due to the downward-looking viewpoints and large-scale diversified environments.…”