ALOHA Low-Cost Bimanual Teleoperation Dataset
Provider：Stanford University (Tony Z. Zhao et al.)
ALOHA is a low-cost bimanual teleoperation platform and demonstration dataset: ~10 fine tasks with ~50 human demos each, four RGB views, 14-DoF joint at 50Hz, under MIT.
At a glance
- Readiness
- L1 · Rigid-body ready
- Rationale
- Manipulation L1: the assets are human teleoperation demonstrations collected on a low-cost bimanual rig (about 10 fine-manipulation tasks, ~50 demos each), not simulation-loadable 3D models; with four RGB cameras, 14-DoF joint plus two gripper-width channels at 50Hz, they feed ACT or diffusion-policy training directly, hence L1 as a real behavior data layer, consistent with DROID and BridgeData V2.
- Scale
- About 10 fine bimanual manipulation tasks, ~50 human teleoperation demos each (static tasks in the official ACT repo), four RGB cameras, 14-DoF joint state plus two gripper widths at 50Hz; LeRobot-format sim/real mirror variants also exist (Task and demo counts follow the ALOHA project and ACT-repo convention (~10 canonical tasks, ~50 demos each) rather than a single fixed release; community LeRobot mirrors (lerobot/aloha_*) include sim transfer/insertion and static cups_open / mobile cabinet variants of differing sizes. Checked 2026-10-28.)
- License
- MIT — The original ALOHA repository (tonyzhaozh/aloha) carries an MIT LICENSE (Copyright 2023 Tony Z. Zhao), covering code and the official static-task data organization; community LeRobot-format mirrors (lerobot/aloha_*) often label apache-2.0 on the dataset card, but that is the LeRobot push_to_hub default rather than an upstream term. Defer to the original repository license before use.
- Access
- Open
- Formats
- other
- Runtimes
- other
- Physics properties
- Joint structureSensor attributes
- Domain
- Other
- Published
- 2026-10-28
- Updated
- 2026-10-28
Why it is listed
ALOHA (A Low-cost Open-source Hardware System for Bimanual Teleoperation) was introduced by Tony Zhao and collaborators at Stanford as the rig and dataset behind Action Chunking with Transformers (ACT). By pairing two 6-DoF ViperX-300 follower arms with two WidowX leader arms, a single operator teleoperates fine bimanual manipulation at 50 Hz with a natural posture, making tasks such as opening condiment cups, slotting batteries, and pouring sugar into coffee repeatable. The released data packs carry synchronized joint states, gripper widths and four RGB camera streams, and the HDF5 layout became a de-facto schema for bimanual imitation learning.
Readiness assessment
ALOHA begat the Mobile ALOHA, ALOHA 2 and ALOHA Unleashed line of datasets, and it remains the reference platform for many bimanual VLA fine-tune recipes including pi-zero and OpenVLA. LeRobot maintains Parquet mirrors that plug directly into Hugging Face streaming dataloaders, and the canonical ACT entrypoint reads HDF5 directly, so the data drops into both imitation-learning and VLA pipelines with minimal preprocessing.
Practical notes
The original repository license is MIT, covering code and the official static-task organization, while LeRobot-format mirrors commonly list apache-2.0 as a conversion default rather than an upstream term. Simulation variants are lower fidelity than the real-world packs, and within each real task the demonstration count is around fifty, so task diversity per variant is limited; treat the canonical static tasks as benchmark demos rather than a large pre-training corpus. The open hardware bill of materials and CAD let labs reproduce the exact rig and contribute their own task collections.
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.