{"success":true,"data":{"author":"Collabora and the Computer Vision Group, TU Munich (Monado project)","bucketAccessType":"public","bucketName":"collabora/monado-slam-datasets","bucketPrefix":"","category":"Robotics","cloudProvider":"huggingface","createdAt":"2026-11-21T22:15:33.011Z","description":"The Monado SLAM Datasets are challenging egocentric visual-inertial (VI) recordings made with virtual-reality headsets to improve inside-out head tracking in Monado, the open-source OpenXR runtime, and to stress state-of-the-art VI-SLAM and visual-inertial odometry systems; the authors note the failure modes also matter for humanoid robotics. Three headsets were used: Valve Index (two 960x960 cameras at 54 fps, IMU at 1000 Hz), HP Reverb G2 (four 640x480 cameras at 30 fps, IMU at 1000 Hz) and Samsung Odyssey+ (two 640x480 cameras at 30 fps, IMU at 1000 Hz). Each headset folder has calibration recordings (camera targets, IMU calibration, magnetometer, and a long sequence for IMU Allan variance) and usage sequences such as playing VR games, with webm preview videos and per-headset documentation. Sequences are zip archives (some split .z01/.z02) with JSON and YAML calibration, about 528 GB in 375 files (measured 2026-11-21). The card also reports tracking results for the Monado fork of Basalt. Supported by Collabora with publication work at TU Munich; released under CC-BY 4.0 with a request to mention the original project; cite arXiv:2508.00088.","displayTitle":"Monado SLAM Datasets (MSD): Egocentric Visual-Inertial Recordings from Valve Index, HP Reverb G2 and Samsung Odyssey+ VR Headsets (Collabora)","huggingFaceUrl":"https://huggingface.co/datasets/collabora/monado-slam-datasets","id":"f393896d-d217-5b09-8d00-c0b8cd0280b7","license":"CC-BY-4.0","licenseEvidenceUrl":"https://huggingface.co/datasets/collabora/monado-slam-datasets","objectCountEstimate":375,"probeStatus":"ok","referenceUrl":"https://arxiv.org/abs/2508.00088","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"collabora/monado-slam-datasets\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"collabora/monado-slam-datasets\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=collabora, license:cc-by-4.0, downloads=5566"],"format":"mixed","generatedAt":"2026-11-21T22:31:28.844Z","layout":{"example":"M_monado_datasets/MG_reverb_g2/MGC_calibration/MGC01_camcalib01_1.zip","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"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","encoding":"hex","key":"M_monado_datasets/MG_reverb_g2/MGC_calibration/MGC01_camcalib01_1.zip","truncatedAt":512}],"sampleObjects":["M_monado_datasets/MG_reverb_g2/MGC_calibration/MGC01_camcalib01_1.zip","M_monado_datasets/MG_reverb_g2/MGC_calibration/MGC02_camcalib02_1.zip","M_monado_datasets/MG_reverb_g2/MGC_calibration/MGC03_camcalib13_1.zip","M_monado_datasets/MG_reverb_g2/MGC_calibration/MGC04_imucalib01_1.zip","M_monado_datasets/MG_reverb_g2/MGC_calibration/MGC05_imucalib02_1.zip"],"sidecars":[{"bytes":"All `camcalib` and `imucalib` datasets were recorded with a fixed exposure of\n3000μs and a gain of 32/255 on the OV7251 sensor looking at aprilgrids with\nsquare side of 3cm. For the magnetometer `magcalib` datasets the exposure/gain\nvalue adjusts automatically per camera as the rest of the datasets.\n\n1. MGC01-MGC07: Were recorded on 2024-03-16.\n2. MGC08-MGC14: Were recorded on some hours later after set 1, on a different\n   monado session.\n3. MGC15-MGC21: Were recorded on 2024-03-17 at the beginning of the monado\n   session that recorded all MGO datasets.\n4. MGC22-MGC28: Were recorded after finishing almost all MGO datasets on the\n   same session as set 3.\n\nThe Reverb G2 headset has 4 cameras:\n- 0: front-left\n- 1: front-right\n- 2: side-left\n- 3: side-right\n\n[Basalt calibration\ntools](https://gitlab.com/VladyslavUsenko/basalt/-/blob/master/doc/Calibration.md)\nwere used for generating the calibration files. The calibration of the 4 cameras\nwas done pairwise. Doing a stereo camera calibration and then an IMU calibration\nover the following three stereo pairs: 01, 02, and 13. Then the calibration file\nof pair 01 was used and extended with intrinscs and extrinsics of cameras 2 and\n3 from the pairs 02 and 13 (i.e., the 01 calibration had priority).\n\nMagnetometer data is only found in raw form since, at the time of recording, the\nmagnetometer units the headset reports was not yet known in the open-source WMR\nMonado driver. For magnetometer calibration you can look up this video guide:\nhttps://www.youtube.com/watch?v=MinV5V1ioWg&t=956s\n","encoding":"utf8","key":"M_monado_datasets/MG_reverb_g2/MGC_calibration/README.md","name":"README.md","size":1554,"truncated":false},{"bytes":"For unzipping the multipart dataset MGO12 you can either use the `7z` command\n\n```bash\n7z x MGO12_freemovement_long_session.zip\n```\n\nOr, the `unzip` command, but since it doesn't support multipart files, you will need create an intermediary file with `zip` like this:\n\n```bash\nzip -F MGO12_freemovement_long_session.zip --out MGO12.zip\nunzip MGO12.zip\n```\n","encoding":"utf8","key":"M_monado_datasets/MG_reverb_g2/MGO_others/README.md","name":"README.md","size":356,"truncated":false},{"bytes":"To be done.\n","encoding":"utf8","key":"M_monado_datasets/MG_reverb_g2/README.md","name":"README.md","size":12,"truncated":false},{"bytes":"calibration 1 and 4 have the side-cameras extrinsics weird, they don't work on basalt unless you swap axes and flip signs accordingly.\n","encoding":"utf8","key":"M_monado_datasets/MG_reverb_g2/extras/other_calibrations/README.md","name":"README.md","size":135,"truncated":false}],"sourceObjectKey":"M_monado_datasets/MG_reverb_g2/MGC_calibration/MGC01_camcalib01_1.zip","starterCell":"from huggingface_hub import hf_hub_download\npath = hf_hub_download(repo_id=\"collabora/monado-slam-datasets\", filename=\"M_monado_datasets/MG_reverb_g2/MGC_calibration/MGC01_camcalib01_1.zip\", repo_type=\"dataset\")\nprint(path)","starterSamplePath":"M_monado_datasets/MG_reverb_g2/MGC_calibration/MGC01_camcalib01_1.zip","totalObjects":375},"summary":"The Monado SLAM Datasets are challenging egocentric visual-inertial (VI) recordings made with virtual-reality headsets to improve inside-out head tracking in Monado, the…","title":"Monado SLAM Datasets (MSD): Egocentric Visual-Inertial Recordings from Valve Index, HP Reverb G2 and Samsung Odyssey+ VR Headsets (Collabora)","updatedAt":"2026-11-21T23:15:34.292Z","visibility":"public","websiteUrl":"https://monado.dev"}}