CVPR 2024 dataset

Robust SLAM data for environments where perception fails

SubT-MRS combines synchronized RGB, LiDAR, thermal, and inertial sensing across caves, tunnels, urban spaces, and off-road terrain. TartanAir adds photorealistic simulation with dense ground truth for testing sim-to-real robustness.

Read the paper Download ROS bags Dataset tools

4 platform typesUGV, UAV, legged, and handheld

4 sensor familiesRGB, LiDAR, thermal, and IMU

Real + simulatedSubT-MRS and TartanAir sequences

Adverse conditionsDarkness, smoke, dust, snow, and fog

Dataset overview

Two complementary sources for robust localization research

SubT-MRS

A real-world, multi-robot dataset collected in subterranean, indoor, outdoor, and mixed environments with hardware-synchronized sensing.

Four RGB cameras, LiDAR, IMU, and thermal imaging
RC cars, legged robots, drones, and handheld motion
Geometric degradation, darkness, smoke, dust, and illumination changes

TartanAir

A photorealistic AirSim dataset designed for difficult visual motion estimation and sim-to-real transfer.

Indoor and outdoor environments across weather and seasons
RGB stereo, depth, optical flow, and semantic labels
Complex 3D motion, dynamic objects, and low illumination

Open the TartanAir dataset page

Synchronized multi-modal data collected across degraded environments.

SubT-MRS sensor modalities and environments — Platforms, sensing modalities, and environmental conditions represented in the release.

Downloads

Sequences, calibration, maps, and trajectories

Primary release

Download the same sequences as ROS bags or extracted folders. Ground truth is distributed separately.

ROS bagsGoogle Drive Extracted foldersGoogle Drive Ground truthGoogle Drive

Sequence	Source	Platform	Sensors	Condition	Resources
`Final_Challenge_UGV1`	SubT-MRS	UGV1	LiDAR, IMU	Geometry degraded	Video Extrinsics
`Final_Challenge_UGV2`	SubT-MRS	UGV2	LiDAR, IMU	Geometry degraded	Video Extrinsics
`Final_Challenge_UGV3`	SubT-MRS	UGV3	LiDAR, IMU	Geometry degraded	Video Extrinsics
`Urban_Challenge_UGV1`	SubT-MRS	UGV1	LiDAR, IMU	Geometry degraded	Video Extrinsics
`Urban_Challenge_UGV2`	SubT-MRS	UGV2	LiDAR, IMU	Geometry degraded	Video Extrinsics
`Laurel_Cavern`	SubT-MRS	Handheld	LiDAR, IMU	Underground cave	Video Extrinsics
`Lidar_factory`	TartanAir	Virtual	LiDAR, IMU	Snow	Calibration
`Lidar_ocean`	TartanAir	Virtual	LiDAR, IMU	Dynamic objects	Calibration
`Lidar_sewerage`	TartanAir	Virtual	LiDAR, IMU	Geometry degraded	Not listed

Sensor Fusion Challenge extension

Additional visual, thermal, inertial, and LiDAR sequences used by the sensor fusion track.

ROS bagsChallenge extension Extracted foldersChallenge extension Ground truthChallenge extension

View 7 sensor fusion sequences

Sequence	Platform	Sensors	Condition	Resources
`Multi_Floor`	SP1	LiDAR, RGB, IMU	Multi-floor	Video Extrinsics Intrinsics
`Long_Corridor`	RC2	LiDAR, RGB, IMU	Long corridor	Video Extrinsics Intrinsics
`BlockLiDAR`	SP1	LiDAR, RGB, IMU	Blocked LiDAR	Video Extrinsics Intrinsics
`BlockVisual`	SP1	RGB, IMU, thermal	Blocked visual/thermal	Video Extrinsics Intrinsics
`SmokeRoom`	RC7	RGB, thermal, IMU	Visual degradation	Video Extrinsics Intrinsics
`OutdoorNight`	SP1	RGB, thermal, IMU	Night illumination	Video Extrinsics Intrinsics
`FlashLight`	SP1	RGB, thermal, IMU	Changing illumination	Video Extrinsics Intrinsics

SuperLoc localization sequences

Evaluation data with ROS bags, calibration, ground-truth maps, and trajectories.

View 8 SuperLoc sequences

Sequence	Source	Platform	Sensors	Resources
`Cave01`	SuperLoc	Handheld	RGB, LiDAR, IMU	Bag Extrinsics Intrinsics Map Trajectory
`Cave02`	SuperLoc	Handheld	RGB, LiDAR, IMU	Bag Extrinsics Intrinsics Map Trajectory
`Cave03`	SubT-MRS	Handheld	RGB, LiDAR, IMU	Bag Extrinsics Intrinsics Map Trajectory
`Cave04`	SuperLoc	Handheld	RGB, LiDAR, IMU	Bag Extrinsics Intrinsics Map Trajectory
`Corridor01`	SubT-MRS	RC2	RGB, LiDAR, IMU	Bag Extrinsics Intrinsics Map Trajectory
`Corridor02`	SuperLoc	RC1	RGB, LiDAR, IMU	Bag Extrinsics Intrinsics Map Trajectory
`Floor01`	SubT-MRS	SP1	RGB, LiDAR, IMU	Bag Extrinsics Intrinsics Map Trajectory
`Floor02` (bonus)	SuperLoc	SP1	RGB, LiDAR, IMU	Bag Extrinsics Intrinsics MapTrajectory not listed

Initialization poses

Download pose configurations

Trajectory format

timestamp x y z q_x q_y q_z q_w

Point cloud conversion

Velodyne conversion tutorial

Platforms and sensors

Heterogeneous motion and synchronized sensing

Sensor platform and synchronized streams.

Dataset sensor specifications — Sensor specifications and platform configuration.

ICCV 2023 challenge archive

Closed October 15, 2023

Three evaluation tracks

The challenge evaluated visual-inertial, LiDAR-inertial, and multi-sensor fusion systems independently. Track pages remain available as an archive.

Visual-inertial LiDAR-inertial Sensor fusion

Citation and related work

SubT-MRS appeared at CVPR 2024

SubT-MRS Dataset

Zhao et al., "SubT-MRS Dataset: Pushing SLAM Towards All-weather Environments," CVPR 2024.

@InProceedings{Zhao_2024_CVPR,
  author    = {Zhao, Shibo and Gao, Yuanjun and Wu, Tianhao and
               Singh, Damanpreet and Jiang, Rushan and Sun, Haoxiang and
               Sarawata, Mansi and Qiu, Yuheng and Whittaker, Warren and
               Higgins, Ian and Du, Yi and Su, Shaoshu and Xu, Can and
               Keller, John and Karhade, Jay and Nogueira, Lucas and
               Saha, Sourojit and Zhang, Ji and Wang, Wenshan and
               Wang, Chen and Scherer, Sebastian},
  title     = {SubT-MRS Dataset: Pushing SLAM Towards All-weather Environments},
  booktitle = {CVPR},
  year      = {2024},
  pages     = {22647--22657}
}