Handheld, UGV
CAVERS
Multimodal cave SLAM
Cave recordings from handheld and rover platforms combining RGB-D, thermal, LiDAR and IMU with partial external motion-capture truth.
Access links
Primary places to inspect, download, read about, or reproduce the dataset.
Quick facts
- Vehicle type
- Handheld, UGV
- Environment
- Indoor
- Data origin
- Real-world
- Sensor count
- 5
- Ground truth
- Available
- Calibration
- Parameters Only
- Annotations
- Not reported
Sensor overview
High-level sensor availability before the detailed sensor records below.
Ground truth
Reference scope, method, rate, coverage, and provenance.
| Scope | Method | Reference system | Count | Rate | Coverage |
Reference provenance
|
|---|---|---|---|---|---|---|
| 6DoF | Motion capture | OptiTrack | 10 infrared cameras | 120 Hz | Partial · Loc Diablo1, Loc Diablo2, Loc Diablo3, Loc Diablo4, Loc Diablo5, Loc Diablo6, Loc Diablo7, Loc Diablo8, Loc Handheld1, Loc Handheld2, Loc Handheld3, Loc Handheld4, Rec Handheld1 |
Reference provenance
External
|
|
||||||
Calibration and synchronization
Reported calibration level, reproducibility signals, and supporting notes.
- Level
- Parameters Only
- Processed parameters
- Available
- Raw calibration data
- Not reported
- Calibration targets
- Heated 12-by-9 glass checkerboard
Parameters only
Processed calibration parameters are reported, but raw calibration data is not reported.
Notes
MATLAB intrinsics, rig CAD and factory D435i extrinsics; recording calibration images does not by itself establish raw-calibration release.
Known or intentional limitations
Reported constraints and characteristics to check before using the dataset.
Motion capture covers thirteen of twenty-four sequences in Sala Dosel; it is absent from other cave recordings. The approximately 5 mm calibration residual is not a whole-trajectory accuracy bound.
Image resolution depends on task: localization and reconstruction RGB/depth differ. The combined 200 Hz inertial output interpolates native 62 Hz acceleration.
The thermal sensor is described inconsistently as near infrared despite an 8–14 micrometre band. Software timestamps leave acquisition delays; RTAB-Map estimates are not added as independent truth.
Cameras
Camera 1
- Model
- Intel RealSense D435i
- Setup
- Rgbd
- Count
- 1
- Effective cameras
- 2
- Modality
- Rgbd
- Resolution
- Not reported
- Rate
- 15 Hz
- Shutter
- Not Reported
- Lens type
- Not Reported
- HFOV
- Not reported
- VFOV
- Not reported
- Notes
- Localization RGB/depth 640x480; reconstruction RGB 1920x1080 and depth 848x480. One fixed resolution would be misleading.
Camera 2
- Model
- Optris PI640i
- Setup
- Mono
- Count
- 1
- Effective cameras
- 1
- Modality
- Thermal
- Resolution
- 640 x 480
- Rate
- 6 Hz
- Shutter
- Not Reported
- Lens type
- Not Reported
- HFOV
- Not reported
- VFOV
- Not reported
- Notes
- Approximate rate. Source says near-IR but gives 8–14 micrometres; normalized thermal images and floating-point temperatures are distinct products.
IMUs
D435i integrated IMU
- Model
- Not reported
- Count
- 1
- Accel / gyro
- 62 / 200 Hz
GNSS
LiDAR
Velodyne VLP-16
- Model or name
- Velodyne VLP-16
- Count
- 1
- Dimensions
- 3
- Rate
- 5 Hz
- Channels
- Not reported
- Range
- Not reported
Radar
Additional sensors
Annotations
Reported annotation availability and task support.
- Availability
- Not reported
- Format
- Not reported
- Class count
- Not reported
Citation
Citation key and BibTeX kept at the end of the page for reference.
@misc{franchini2026cavers,
author = {Giacomo Franchini and David Rodríguez-Martínez and Alfonso Martínez-Petersen and Carlos J. Pérez-del-Pulgar and Marcello Chiaberge},
title = {CAVERS: Multimodal SLAM Data from a Natural Karstic Cave with Ground Truth Motion Capture},
year = {2026},
url = {https://arxiv.org/abs/2604.15052v2}
}
Dataset corrections
Notice a specific issue after reviewing the page?
Report an inaccuracy