TartanAir V2: multi-sensor synthetic dataset for SLAM, navigation and robot perception

Provider
Carnegie Mellon University AirLab (Wenshan Wang, Yaoyu Hu, Yuheng Qiu, Shihao Shen, Yorai Shaoul)
License
CC-BY-4.0 (source)
Size
7,697 files
Stored at
Hugging Face (https://huggingface.co/datasets/theairlabcmu/tartanair2)
Access
Free and public. No account with the data provider is needed.
Category
Computer Vision
Source
https://tartanair.org/

About this dataset

TartanAir V2 is the second release of Carnegie Mellon University AirLab's simulated dataset for SLAM, navigation and robotics, expanding TartanAir from 18 to dozens of photo-realistic environments ranging from mountains, caves and towns to stylish homes and the Moon, each with recorded trajectories designed to be challenging and realistic. Sensors are rendered for a full rig of cameras around the platform (front, back, left, right, top and bottom for a left and right camera) plus fisheye and equirectangular panoramas, with RGB images, depth, semantic segmentation, camera poses, optical flow, LiDAR, IMU with a noise model, event-camera data and occupancy maps. The tartanair Python package (pip install tartanair) downloads, iterates, customises camera models and evaluates trajectories. Measured 2026-11-14: 74 environment folders (for example AbandonedCable, AmericanDiner, CoalMine, CyberPunkDowntown, ForestEnv), each with Data_easy and Data_hard trajectory sets stored as per-modality, per-camera zip archives; one environment's Data_easy folder alone holds 53 zips and about 1.06 TB, and the Hub reports about 54 TB for the repository. Uses: visual SLAM, multi-camera and fisheye depth estimation, optical flow, semantic segmentation, event-based vision and sim-to-real perception. The TartanAir V2 dataset is licensed by CMU under CC BY 4.0 and the toolkit under MIT.

How to load this dataset in Python

from huggingface_hub import hf_hub_download
path = hf_hub_download(repo_id="theairlabcmu/tartanair2", filename="AbandonedCable/Data_easy/depth_lcam_back.zip", repo_type="dataset")
print(path)

Data structure

Files

Example file: AbandonedCable/Data_easy/depth_lcam_back.zip

Analyze this dataset without downloading it

On Scigantic, TartanAir V2: multi-sensor synthetic dataset for SLAM, navigation and robot perception opens in a hosted JupyterLab notebook with the data already mounted, so you can read files where they are stored instead of copying them first. Each notebook includes an AI assistant that can see the mounted files and write and run analysis code. Open the dataset page to launch a notebook, or browse the full catalog.

This page is also available as Markdown and as JSON.

Related Computer Vision datasets