Oxford Osprey Dataset
We present the Oxford Osprey dataset, a LiDAR-Visual-Inertial (LVI) dataset captured with Osprey, an autonomous aerial mapping system capable of surveying large outdoor structures over multiple flights. The dataset provides aerial surveys of three separate large industrial sites, with a total ground coverage of 2528 square meters. The Osprey platform incorporates a custom sensor payload, called Frontier, onto a DJI M600 drone. The captured sequences contain LiDAR pointclouds, IMU measurements and colour images from three fisheye cameras.
Please cite:
Osprey: Multi-Session Autonomous Aerial Mapping with LiDAR-based SLAM and Next Best View Planning
Rowan Border, Nived Chebrolu, Yifu Tao, Jonathan D. Gammell, Maurice Fallon
IEEE Transactions on Field Robotics (T-FR) 2024
[Preprint] [PDF] [Video]
@article{Border2024,
author = {Rowan Border and Nived Chebrolu and Yifu Tao and Jonathan D. Gammell and Maurice Fallon},
journal = {IEEE Transactions on Field Robotics},
pages = {1-25},
title = {Osprey: Multi-Session Autonomous Aerial Mapping with LiDAR-based SLAM and Next Best View Planning},
year = {2024},
}
Sites
The field experiments in this dataset were conducted at the Fire Service College, Moreton-in-the-Marsh, Gloucestershire, UK. The facility is a training center for firefighters with several standalone buildings and experiments were conducted at three of these: a medium-sized two storey industrial building (Industrial Building A), a large five storey industrial building with several external fire escapes (Industrial Building B) and an even larger site containing multiple structures that is staged as a disaster scenario (Rig 5). Three types of experiment were performed at each site: a pilot-flown mission, an Osprey autonomous mapping mission and a survey-grade scan using a Leica BLK360.
Industrial Building A
Industrial Building A is a two storey building, 30x16x12m (LxWxH) in size, with a balcony surrounding the upper storey, which is partially enclosed on one side. There is a concrete fire escape on one corner of the building, a garbage chute (red) attached to one side, a concrete chimney and various metal structures on the roof.
| Operation | Mission | Flight | Title | Link |
|---|---|---|---|---|
| Pilot | 1 | 1 | ind_a_pilot_mission_1_flight_1 | link |
| Autonomy | 1 | 1 | ind_a_autonomy_mission_1_flight_1 | link |
| Autonomy | 1 | 2 | ind_a_autonomy_mission_1_flight_2 | link |
| Leica BLK360 | N/A | N/A | ind_a_leica_blk360 | link |
Industrial Building B
Industrial Building B is a five storey building, 27x20x21m (LxWxH) in size, with a balcony surrounding the fourth storey and two concrete fire escapes attached to different sides of the building. There is also an exposed metal fire escape with a garbage chute attached to the building, which is comprised of thin metal structures and proved challenging to map.
| Operator | Mission | Flight | Title | Link |
|---|---|---|---|---|
| Pilot | 1 | 1 | ind_b_pilot_mission_1_flight_1 | link |
| Autonomy | 1 | 1 | ind_b_autonomy_mission_1_flight_1 | link |
| Autonomy | 1 | 2 | ind_b_autonomy_mission_1_flight_2 | link |
| Leica BLK360 | N/A | N/A | ind_b_leica_blk360 | link |
Rig 5
Rig 5 is a large site, 52x29x27m (LxWxH) in size, which contains multiple structures and is staged as a disaster scenario. The site is comprised of a seven storey building next to a long three storey building with an exposed interior on one side and a metal fire escape at one end. There is a large rubble pile next to these buildings with several upright metal girders within it.
| Operator | Mission | Flight | Title | Link |
|---|---|---|---|---|
| Pilot | 1 | 1 | rig_5_pilot_mission_1_flight_1 | link |
| Autonomy | 1 | 1 | rig_5_autonomy_mission_1_flight_1 | link |
| Autonomy | 1 | 2 | rig_5_autonomy_mission_1_flight_2 | link |
| Autonomy | 2 | 1 | rig_5_autonomy_mission_2_flight_1 | link |
| Autonomy | 2 | 2 | rig_5_autonomy_mission_2_flight_2 | link |
| Autonomy | 2 | 3 | rig_5_autonomy_mission_2_flight_3 | link |
| Leica BLK360 | N/A | N/A | rig_5_leica_blk360 | link |
ROS Topics
Each sequence is available in both ROS1 bag and ROS2 database formats and contains the following topics:
| Topic | Frequency [Hz] | Description |
|---|---|---|
/tf |
100 | Frame transforms |
/tf_static |
N/A | Static transforms |
/hesai/pandar |
10 | LiDAR pointclouds |
/alphasense_driver_ros/imu |
400 | IMU measurements |
/alphasense_driver_ros/cam0/color/image/compressed |
20 | Front camera images |
/alphasense_driver_ros/cam1/color/image/compressed |
20 | Left camera images |
/alphasense_driver_ros/cam2/color/image/compressed |
20 | Right camera images |
Platform
The Osprey aerial platform is a DJI M600 drone with a custom sensor payload, called Frontier, that was developed at the Oxford Robotics Institute (ORI). The Frontier consists of (a) a Hesai QT64 LiDAR, (b) a SevenSense Core Research sensor module with an IMU and three 1.6 megapixel colour fisheye cameras, and (c) an Intel NUC computer. Communication with the operator is provided by a wireless mesh network comprised of (d) a Rajant BreadCrumb ES1 module on Osprey and (e) Rajant BreadCrumb DX2 modules distributed around a target site.
| Sensor | Type | Rate [Hz] | Characteristics |
|---|---|---|---|
| Hesai Pandar QT64 | LiDAR | 10 | 64 beams, 20m range, 104° vertical FoV |
| Alphasense Core Research | Cameras | 20 | colour, global shutter, 1440 × 1080 |
| Bosch BMI085 | IMU | 400 |
Calibration
The camera intrinsics and camera-IMU extrinsics were calibrated with Kalibr (Furgale et al., 2013). The camera-LiDAR extrinsics were calibrated with DiffCal (Fu et al., 2023). The sensors are all connected to the NUC via Ethernet and time synchronised with PTP.
All of the intrinsic and extrinsic sensor calibration parameters are available here.
Ground Truth
Ground truth trajectories were not captured for these field experiments. However, the trajectory estimated online by the VILENS (Wisth et al., 2023) odometry algorithm is provided for comparison. This trajectory is provided on the /tf topic as the base_vilens frame relative to the odom_vilens frame.
Survey-grade maps of each site were captured using a Leica BLK360. These only contain surfaces visible from the ground-based scanning locations of the BLK360. They can be used to evaluate the reconstruction accuracy and coverage of SLAM maps.
Contact
Please email any inquiries regarding the dataset to rborder.robots@gmail.com.
Acknowledgements
Thanks to Wayne Tubby, Matt Towlson and Arundathi Shanthini at ORI for building the Frontier sensor payload and mounting infrastructure for the DJI M600.
This research was funded by Horizon Europe through the Digiforest project [101070405] and by UK Research and Innovation and EPSRC through ACE-OPS: From Autonomy to Cognitive assistance in Emergency OPerationS [EP/S030832/1].