01
Browse robot data samples
Switch between LeRobot, RLDS, HDF5, ROS bag, and mp4/jsonl collections without turning the app into a notebook.
ROBOTICS STUDIO OPEN · APACHE-2.0 · LOCAL DATASET REVIEW IDE
Review robot datasets without uploading the robot.
A desktop workbench for LeRobot, RLDS, OpenX, HDF5, ROS bag, and mp4/jsonl captures. Scrub episodes, tag failures, probe policies, and export reviewed manifests and dataset cards. The robot data stays on your machine. Nothing is pooled. It opens the same teleop, VLA, and failure episodes Human Data OS Robotics delivers rights-cleared, reviewed, and signed — here you inspect them yourself.
After a competitor lost four terabytes — including who its workers were — nobody wants tooling that uploads the robot. This one never does.
Local
build proof required
Nothing
uploaded or pooled
Apache-2.0
read the source
02
Review the episode
Scrub synchronized camera, depth, joint, force, and language streams with review actions close to the timeline.
03
Cluster the failures
Group repeated failure modes, inspect representatives, split or merge clusters, and keep training-readiness visible.
04
Export a reviewed subset
Write an export manifest, a dataset card, and consent and license references to local disk. A reviewer can open the bundle without an account.
THE PRODUCT SURFACE
A desktop editor for the data only the robot can give.
Open a collection, scrub the run, cluster the failures, export the reviewed subset. The same review the robot data needs, run on your own infrastructure.
REVIEW WORKBENCH
Episode review that feels like a desktop editor.
Open a collection, scrub the run, tag the failure. The desktop build and browser preview do the same review work, on your machine.
EPISODE SCRUB
Every stream stays in the same review loop.
Camera, depth, joint state, force, language, phase, and intervention notes stay visible while the reviewer tags the exact failure.
FAILURE INTELLIGENCE
Cluster review is a first-class workflow.
Representative failures, readiness scores, dominant tags, and cluster actions are presented as review primitives.
EXPORT MANIFEST
Every export leaves a portable file.
Each export writes a manifest, a dataset card, and consent and license references to local disk — a folder a reviewer can open and diff without an account.
SOURCE AND RELEASE PROOF
Source listed. Release proof required.
Apache-2.0 source is listed on GitHub. Package, release, checksum, desktop trust, and platform proof are required before install or binary availability claims are marketed.
License
Apache-2.0 — source link listed; license proof required.
Source link listed
Source
github.com/auraoneai/robotics-studio-open
Source link listed
Install (macOS)
Package release proof required before install commands are marketed.
Release proof required
Desktop artifact
Binary artifact, signing, and notarization proof required.
DMG proof required
Checksum
Checksum proof required before checksum claims are marketed.
Checksum proof required
Browser preview
Hosted preview proof required before browser availability is marketed.
Preview proof required
Platforms
Desktop platform proof required.
Desktop trust proof required
Changelog
Release and changelog proof required.
Platform proof required
DEMO
Watch the local review loop. Then read the docs.
The docs read like engineering references: the capture schema, the exact export formats, and the teleop server. Open is where you inspect the episodes locally. Human Data OS Robotics is where they arrive non-pooled — signed delivery, RLDS, OpenX, and HDF5, with the provenance the EU AI Act asks for in August 2026. Bring it in when the work becomes shared: shared authorship, approval queues, governance across a team. Until then, the tools are yours.
CAPTIONED DEMO · LOCAL REVIEW WORKFLOW
ROBOTICS STUDIO OPEN
Review the robot data. Keep the code.
Review the source snapshot and inspect the robot-data examples without uploading anything. The episodes are sample Human Data OS Robotics review records; here you inspect the format yourself. When the work becomes a shared dataset across a team, the OS handles non-pooled delivery and provenance — and you keep the weights.