{"success":true,"data":{"author":"Raul Castilla-Arquillo, Miguel Olivares-Mendez","bucketAccessType":"public","bucketName":"raulcastar/mare-lunar-rover-dataset","bucketPrefix":"","category":"Astronomy & Space","cloudProvider":"huggingface","createdAt":"2026-11-18T00:06:04.272Z","description":"5,040 synthetic samples recorded by a lunar rover (Leo Raph) driving autonomously across eight procedurally generated 500 m x 500 m terrains in the MARE simulator (Unreal Engine 5.7), under low sun elevations of 3 to 35 degrees with long shadows. Each sample pairs a 640 x 480 central navigation camera (with and without lens glare), a stereo pair with 20 cm baseline, a rear camera, an event camera (346 x 260), a SPAD single-photon camera (256 x 256) and a 1 Hz lidar scan with pixel-accurate ground truth: planar depth (uint16 mm and float32 m), per-pixel object classes (sky, rover, sun disc, sun glare, landscape, crater, crater rim, wheel track, boulder, small rock) and shadow classes that separate the rover's own shadow from terrain shadow. Splits are by terrain: 3,872 train, 583 validation, 585 test. Shards are WebDataset tar files loadable with datasets or webdataset; the repo also holds per-episode metadata and intrinsics, the terrain DEMs with crater and rock lists, and MCAP rosbags of odometry, IMU and tf. Licensed CC BY 4.0; cite the MARE paper (Castilla-Arquillo and Olivares-Mendez, 2026).","displayTitle":"MARE Lunar Rover Dataset: Synthetic Multisensor Frames with Depth, Object and Shadow Ground Truth","huggingFaceUrl":"https://huggingface.co/datasets/raulcastar/mare-lunar-rover-dataset","id":"9b02ae1b-5f53-5bd5-a633-4fc860d63262","license":"CC-BY-4.0","licenseEvidenceUrl":"https://huggingface.co/datasets/raulcastar/mare-lunar-rover-dataset","objectCountEstimate":179,"probeStatus":"ok","referenceUrl":"https://arxiv.org/abs/2610.28465","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"raulcastar/mare-lunar-rover-dataset\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"raulcastar/mare-lunar-rover-dataset\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=raulcastar, license:cc-by-4.0, size_categories:1K<n<10K, downloads=412"],"format":"mixed","generatedAt":"2026-11-18T00:09:28.137Z","layout":{"example":"episodes/map_000_ep00_frames.csv","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"frame,stamp,events_stamp,spad_stamp,lidar_stamp,noglare_stamp,rear_stamp,x,y,z,qx,qy,qz,qw\r\n0,1792745421.918,1792745421.851334,1792745421.918,,1792745421.918,1792745421.918,452.97746,79.07168,198.78469,0.02174,-0.0575,-0.53096,0.84516\r\n1,1792745423.918,1792745423.884667,1792745423.884667,1792745423.518001,1792745423.918,1792745423.918,452.96582,79.08891,198.78306,0.02356,-0.05666,-0.54785,0.83432\r\n2,1792745425.918,1792745425.851334,1792745425.851334,1792745425.518001,1792745425.918,1792745425.918,452.93453,","encoding":"utf8","key":"episodes/map_000_ep00_frames.csv","truncatedAt":512}],"sampleObjects":["bags/map_000_ep00.mcap","bags/map_000_ep01.mcap","bags/map_000_ep02.mcap","bags/map_000_ep03.mcap","bags/map_001_ep00.mcap"],"sidecars":[{"bytes":"---\nlicense: cc-by-4.0\npretty_name: MARE Lunar Rover Dataset\nsize_categories:\n- 1K<n<10K\ntask_categories:\n- image-segmentation\n- depth-estimation\n- object-detection\ntags:\n- lunar\n- moon\n- planetary-robotics\n- rover\n- space\n- synthetic\n- simulation\n- shadows\n- event-camera\n- spad\n- lidar\n- stereo\n- unreal-engine\nconfigs:\n- config_name: default\n  data_files:\n  - split: train\n    path: data/train/*.tar\n  - split: validation\n    path: data/validation/*.tar\n  - split: test\n    path: data/test/*.tar\n---\n\n# MARE Lunar Rover Dataset\n\n5,040 synthetic samples from a lunar rover driving autonomously across eight\nprocedurally generated terrains, recorded in the\n[MARE simulator](https://github.com/raulcastar/MARE-Lunar-Simulator)\n([paper](https://arxiv.org/abs/2610.28465)). Every sample pairs the rover's\ncameras with pixel-accurate ground truth: depth, object classes and shadows,\nincluding the rover's own shadow, under low lunar sun angles.\n\n![A sample: RGB, glare-free RGB, depth, lidar depth, object and shadow labels, rear camera, SPAD and event camera with their labels](images/sample_1.png)\n\n## What is in a sample\n\n| Field | Format | Content |\n|---|---|---|\n| `rgb_central.png` | RGB 640×480 | Central navigation camera (90° horizontal FOV), with the camera's lens flares and bloom |\n| `rgb_central_noglare.png` | RGB 640×480 | The same rendered frame without bloom or lens flares (91 % of samples) |\n| `rgb_left.png`, `rgb_right.png` | RGB 640×480 | Stereo pair, same instant, 20 cm baseline |\n| `depth.png` | uint16, mm | Ground-truth planar depth of the central camera; 0 = sky or beyond 65.5 m |\n| `depth_m.npy` | float32, m | The same, unrounded; `inf` = sky |\n| `object.png` | uint8 ids | Object class per pixel of the central camera (classes below) |\n| `shadow.png` | uint8 ids | Shadow class per pixel of the central camera (classes below) |\n| `rear_rgb.png` | RGB 640×480 | Rear camera, looking back (90° FOV); sees the wheel tracks (93 % of samples) |\n| `rear_object.png`, `rear_shadow.png` | uint8 ids | Its object and shadow labels (91 %) |\n| `events.png` | RGB 346×260 | Event-camera frame (positive/negative events) |\n| `spad.png` | grey 256×256 | SPAD (single-photon) camera image |\n| `events_object.png`, `events_shadow.png`, `events_depth_m.npy` | on the event grid | The central labels and depth resampled onto the event camera's pixels |\n| `spad_object.png`, `spad_shadow.png`, `spad_depth_m.npy` | on the SPAD grid | The same on the SPAD's pixels; 255 / NaN where the SPAD sees beyond the central camera |\n| `lidar.npy` | float32 (rings, columns, 4) | Lidar scan: x, y, z, intensity in the lidar frame (x forward, y left, z up); NaN = no return |\n| `lidar_depth.png`, `lidar_depth_m.npy` | uint16 mm / float32 m | That scan projected into the central camera (sparse; 0 = no point), corrected for the rover's motion between scan and frame |\n| `json` | dict | Episode, frame, map, split, timestamps of every sensor, rover pose (map frame), sun azimuth and elevation |\n\nA field is missing from a sample when that sensor had no frame close enough in\ntime (event, SPAD and rear camera within 0.1 s, lidar within 0.5 s; the lidar\nruns at 1 Hz).\n\n**Geometry.** The central, SPAD and event cameras share one position,\norientation and 90° horizontal field of view; the SPAD and event labels are the\ncentral ones resampled, not separate renders. Intrinsics (`camera_info`) and the\nclass legends are in `episodes/<episode>.json`; the camera and lidar poses on\nthe rover are in the bag's `/tf_static`.\n\n### Object classes (`object.png`, `rear_object.png`)\n\n| id | class | id | class |\n|---|---|---|---|\n| 0 | unlabeled | 6 | crater |\n| 1 | sky | 7 | crater rim |\n| 2 | rover | 8 | wheel track |\n| 3 | sun (disc) | 9 | boulder (radius ≥ 0.5 m) |\n| 4 | sun glare (the sky washed out around the sun) | 10 | small rock |\n| 5 | landscape | 255 | no label (SPAD rows outside the central camera) |\n\nPixel shares in the central camera: landscape 69.5 %, sky 23.8 %, crater rim\n2.7 %, small rock 1.9 %, boulder 1.2 %, crater 0.7 %, wheel track 0.1 % (13.6 %\nin the rear camera). ","encoding":"utf8","key":"README.md","name":"README.md","size":8230,"truncated":true}],"sourceObjectKey":"episodes/map_000_ep00_frames.csv","starterCell":"import pandas as pd\ndf = pd.read_csv(\"hf://datasets/raulcastar/mare-lunar-rover-dataset/episodes/map_000_ep00_frames.csv\")\ndf.head()","starterSamplePath":"bags/map_000_ep00.mcap","totalObjects":179},"summary":"5,040 synthetic samples recorded by a lunar rover (Leo Raph) driving autonomously across eight procedurally generated 500 m x 500 m terrains in the MARE simulator (Unreal Engine…","title":"MARE Lunar Rover Dataset: Synthetic Multisensor Frames with Depth, Object and Shadow Ground Truth","updatedAt":"2026-11-18T04:02:08.940Z","visibility":"public","websiteUrl":"https://github.com/raulcastar/MARE-Lunar-Simulator"}}