TartanGround: Large-Scale Simulated Ground Robot Perception and Navigation Dataset (CMU AirLab)
- Provider
- AirLab, Carnegie Mellon University, with ETH Zurich (Patel et al.)
- License
- CC-BY-4.0 (source)
- Size
- 34,673 files
- Stored at
- Hugging Face (https://huggingface.co/datasets/theairlabcmu/TartanGround)
- Access
- Free and public. No account with the data provider is needed.
- Category
- Robotics
- Source
- https://arxiv.org/abs/2505.10696
About this dataset
TartanGround is a large multi-modal dataset from the CMU AirLab for perception and autonomy of ground robots, collected in 63 photorealistic simulation environments grouped as indoor, nature, rural, urban, industrial/infrastructure and historical scenes. It has 878 trajectories and over 1.44 million samples from three simulated robot platforms: omnidirectional (P0xxx), differential-drive (P1xxx) and quadruped legged robots (P2xxx). Each sample carries RGB stereo camera pairs facing six directions, depth maps, semantic segmentation, optical flow, stereo disparity, LiDAR point clouds, IMU, 6-DoF ground-truth poses, 3D semantic occupancy voxel maps and, for the quadruped runs, proprioceptive data. Supported tasks include semantic and open-vocabulary occupancy prediction, visual SLAM, neural scene representation, bird's-eye-view prediction and navigation. The repository is organized by environment folder (for example ForestEnv, GreatMarsh, Downtown, NordicHarbor) with per-modality zip archives, about 34,700 zips and 17.9 TB in total (measured 2026-11-21); download selectively with huggingface_hub allow_patterns. Released under CC-BY-4.0; cite Patel et al., TartanGround, arXiv:2505.10696 (2025).
How to load this dataset in Python
from huggingface_hub import hf_hub_download
path = hf_hub_download(repo_id="theairlabcmu/TartanGround", filename="AbandonedCable/AbandonedCable_rgb_pcd.zip", repo_type="dataset")
print(path)Data structure
Files
Example file: AbandonedCable/AbandonedCable_rgb_pcd.zip
Analyze this dataset without downloading it
On Scigantic, TartanGround: Large-Scale Simulated Ground Robot Perception and Navigation Dataset (CMU AirLab) 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 Robotics datasets
- Arena GR 1 Manipulation PlaceItemCloseDoor Task
- FoMo Dataset
- PhysicalAI SimReady Warehouse 01
- PhysicalAI Robotics Open H Embodiment
- Dexwm
- Everyday Manipulation 3D: 1,513 Raw iPhone LiDAR RGB-D Egocentric Clips of Two-Handed Tasks with ARKit Camera Poses (CaryX AI)
- PhysicalAI Robotics Manipulation Kitchen
- MOSAIC: Multi-Source Human Motion Capture and Unitree G1 Retargeted Motions for Humanoid Motion Tracking and Teleoperation (BAAI)