Raw data from the three V3 flights reported in the dissertation. Each flight followed an identical four-leg lawn-mower pattern (4 lanes × 25 m, 8 m altitude, ~128 s duration, ~289 m path length) over the same procedurally-generated 97-obstacle orchard world. The only varying parameter was Gazebo scene lighting.
| Flight | Directory | Sun elevation | Label |
|---|---|---|---|
| A | flight_A_noon/ |
60° | Baseline — noon |
| B | flight_B_dusk/ |
20° | Reduced contrast — dusk |
| C | flight_C_dawn/ |
10° | Low contrast — dawn |
gps_trajectory.csv— ground-truth trajectory from PX4 via pymavlink, 20 Hz. Columns: timestamp, N (m), E (m), D (m), velocity components, attitude (roll/pitch/yaw).cloud.pcd— dense RTAB-Map point cloud. Binary PCD format. Can be viewed with PCL tools or CloudCompare.features.png— feature-tracker visualisation showing the KLT-tracked feature points during flight.gazebo_scene.png— screenshot of the Gazebo scene at mission time.rtabmap_cloud.png— rendered view of the dense point cloud in RViz.metrics.json— pipeline throughput statistics (camera Hz, feature tracker Hz, VINS odometry Hz).fly.log— mission execution log fromfly_mission.py.summary.json— flight summary (duration, lanes, cruise altitude).best10.mp4— short video clip of the flight.
- Flight A: 6.47 m endpoint drift, 2.18 M RTAB-Map points
- Flight B: 7.86 m endpoint drift, 2.01 M RTAB-Map points
- Flight C: 34.45 m endpoint drift, 2.22 M RTAB-Map points
The 5.3× drift amplification (A → C) with near-identical point cloud density confirms the combined-architecture thesis: VINS-Fusion degrades with lighting, RTAB-Map does not.
See ../analysis/ for derived metrics and per-flight plots.