Rigid-body ready Embodied / Robotics CC BY-NC (non-commercial) Sign-up / registration

AgiBot World Colosseo Large-Scale Robot Manipulation Platform Dataset

Provider：AgiBot Robotics / OpenDriveLab with Shanghai AI Laboratory and others

AgiBot World provides more than one million bimanual teleoperated trajectories (about 43.8TB) from 100 robots and 100+ replicated scenes, with visual-tactile and dexterous hand data under CC BY-NC-SA terms.

Get it from the source →

At a glance

Readiness
L1 · Rigid-body ready
Rationale
Manipulation L1: the Beta release holds 1,003,672 real bimanual teleoperated trajectories (about 43.8TB) from 100 robots, covering 100+ one-to-one replicated real scenes, 5 target domains and 200+ task types, with visual-tactile sensing and 6-DoF dexterous hand data available for direct VLA policy training; but the assets are real-world video and action logs rather than simulation-loadable 3D models, hence L1 as a real behavior data layer, consistent with how BridgeData V2 and RLBench are graded.
Scale
Beta release 1,003,672 trajectories (about 43.8TB), Alpha curated subset 92,214 trajectories (about 8.5TB), from 100 robots across 100+ one-to-one replicated real scenes, 5 target domains and 200+ task types (The Beta and Alpha figures come from the official GitHub README, where Alpha is a curated subset of Beta rather than an independently collected corpus. The dataset card additionally cites 2,976.4 hours of total duration. The official 2026 LeRobot-format release directory publishes no aggregate trajectory count, so no unverified total is quoted here. Checked 2026-10-26.)
License
CC BY-NC (non-commercial) — The License and Citation section of the official GitHub README states that all data and code in the repository are under CC BY-NC-SA 4.0, and the README badge plus the arXiv 2503.06669 paper page point to the same license. This entry maps that to cc-by-nc, but the actual grant includes a ShareAlike obligation and is therefore stricter than plain CC BY-NC, so the original terms must be read verbatim before commercial use. On Hugging Face, the AgiBot World Community License Agreement must be signed and gated approval granted, hence access is gated.
Access
Sign-up / registration
Formats
other
Runtimes
other
Physics properties
Collision geometryJoint structureSensor attributes
Domain
Other
Published
2026-10-26
Updated
2026-10-26

Why it is listed

AgiBot World Colosseo is positioned as a full-stack robot learning platform rather than a single dataset release, combining large-scale bimanual data, a foundation model and a challenge ecosystem. The collection spans 100 robots and more than 100 one-to-one replicated real-life scenarios across 5 target domains, with the hardware including visual tactile sensors, a 6-DoF dexterous hand and mobile dual-arm platforms, which puts it among the few open corpora that pair manipulation data with contact sensing on dexterous hardware.

Readiness assessment

The Beta release is the full dataset at 1,003,672 trajectories, roughly 43.8TB, while Alpha is a curated 92,214-trajectory subset of it at about 8.5TB, so teams can start small without leaving the same distribution. Data is published in a WebDataset-style layout with task catalogs that map each episode to an end-effector type and an action-text description, and the project also open-sources GO-1, the foundation model pretrained on this corpus, together with fine-tuning scripts and both full and lightweight checkpoints.

Practical notes

Licensing is the main constraint: the repository states that all data and code are under CC BY-NC-SA 4.0, which permits non-commercial use but imposes share-alike obligations, and it is worth noting that a derivative model may inherit those conditions. Downloads run through Hugging Face under a community license agreement that requires a name, affiliation and contact details, so the process is gated rather than anonymous. The dataset is organized by task and robot, and the accompanying task spreadsheet is the authoritative index for which end-effector and scene each episode belongs to.

Disclaimer: license and usage terms follow the official source — verify them yourself before use. This site indexes and links only; it does not host or mirror any asset file.

License and usage terms are those of the official source — verify before use. No asset file is hosted or mirrored on this site.