Mobile ALOHA Whole-Body Mobile Bimanual Teleoperation Dataset
Provider：Stanford University (Zipeng Fu, Tony Z. Zhao, Chelsea Finn)
Mobile ALOHA adds a wheeled base to ALOHA for whole-body mobile manipulation; the official TFDS release has 276 episodes under CC-BY 4.0 data licensing.
At a glance
- Readiness
- L1 · Rigid-body ready
- Rationale
- Manipulation L1: a wheeled mobile base plus a whole-body teleoperation interface are added to the stationary ALOHA bimanual rig to collect kitchen/home whole-body mobile-manipulation demos (the official Open X / TFDS release holds 276 episodes, ~50 per task); these are real demos rather than simulation-loadable 3D models, hence L1 as a real behavior data layer, consistent with DROID and ALOHA.
- Scale
- The official Open X / TFDS release has 276 episodes with three RGB cameras (overhead plus two wrists at 480x640), a 14-D joint state, a 16-D action (dual 7-DoF arms plus grippers and base linear/angular velocity) and per-step language instructions; about 50 human demos per task across kitchen/home skills such as sauteing, opening a two-door cabinet, taking an elevator and rinsing a pan (Episode count is the 276 of the official Open X / TFDS release (aloha_mobile); the separate '~50 demos per task' phrasing describes collection density and is a different basis, so neither is reconciled. The base is an AgileX Tracer AGV; the full rig weighs about 75kg with a roughly 12-hour battery. Checked 2026-10-28.)
- License
- CC BY — The official Open X / TFDS data release (aloha_mobile) is CC-BY 4.0; the companion code repository (tonyzhaozh/mobile_aloha) is MIT and covers code only. Community LeRobot mirrors and secondary catalogs sometimes mix the two, so defer to the official TFDS dataset card 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
Mobile ALOHA extends the stationary ALOHA rig with an AgileX Tracer wheeled base and a whole-body teleoperation interface in which the operator walks behind the robot and backdrives both arms and the base at once. The design targets mobility near walking speed, stability under heavy household objects, simultaneous whole-body control and untethered onboard power and compute, at a build cost well below prior mobile-manipulation platforms.
Readiness assessment
The official Open X / TFDS release packages 276 episodes with three RGB cameras at 480x640, a 14-D joint state, a 16-D action (dual 7-DoF arms plus grippers and base velocities) and per-step language instructions, so it shares the RLDS loader path with Open X-Embodiment and DROID. Co-training the mobile data with static ALOHA episodes raised success rates substantially on whole-body tasks such as wiping wine, calling an elevator, using a cabinet, rinsing a pan and cooking shrimp.
Practical notes
The data release is CC-BY 4.0 while the code repository is MIT and covers code only; community LeRobot mirrors sometimes mix the two. The rig weighs about 75kg with roughly 12 hours of battery, and the operator tethers at the waist and detaches the leader arms with four screws for autonomous execution. Because the dataset is small, most users co-train it with static ALOHA or larger corpora rather than train from it alone.
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.