AuraOne / Resources / AuraOne Open

Robotics Studio Open: Local VLA Dataset Review

Review declared robotics episode metadata, preserve unknown values and export local evidence with Robotics Studio Open’s explicit input and delivery boundaries.

Robotics Studio Open cockpit showing declared episode metadata and the local review workspace
Published
May 14, 2026
Reviewed
October 25, 2026
Author
AuraOne Open team
Category
AuraOne Open
Reading
4 min

Robotics teams need to understand which episodes were reviewed, what failed and what evidence supports a delivery. Physical AI needs more than raw video: intent, provenance and uncertainty must survive the review.

Robotics Studio Open 0.2.0 is a local-first interface for reviewing declared robot episode metadata and producing inspectable evidence artifacts. Its versioned README, checked for this article update on October 25, 2026, distinguishes the visual app from a native robotics engine. The GitHub release provides a signed and notarized macOS Apple Silicon DMG under the MIT license; the browser and source-build workflows have their own runtime boundaries.

Which inputs the 0.2.0 review app supports

The visual app works from metadata supplied by the user or bundled fixture. It does not infer missing measurements or prove that a referenced video was decoded.

Input or actionVersion 0.2.0 boundary
JSON manifest or JSONL episode recordsOpens supported declared metadata for review. Missing values remain unknown.
Bundled fixtures/sample-so101Contains 96 deterministic synthetic metadata variants from three documented seed scenes; no real robot media or third-party demonstrations.
Parquet, HDF5, ROS bag, DB3 or MP4 binary inputRejected by the visual source build with an adapter error; no native decoder is connected.
Sensor QA and local VLA probeUses declared QA/rate fields and deterministic local mock trials; it does not establish physical-model performance.
Local artifact exportProduces a ZIP containing applicable manifests, review records, cards, reports and SHA-256 checksums.
Hugging Face or AuraOne Programs deliveryUnavailable network destinations in this source build; no authentication or upload is implied.

Standalone engines and CLIs are separate projects. Their format adapters do not establish adapter support in the visual app. Confirm capabilities against the selected release’s adapter documentation rather than assuming support from a format name.

Review one synthetic episode and inspect the output

Start from the versioned quickstart and the bundled synthetic fixture. This is a safe first example because it contains no customer capture or real robot recording.

  1. Open the hosted example or verified desktop release and select the bundled SO-101-style sample.
  2. Inspect an episode’s declared duration, streams, outcome and provenance. Identify which values are present and which are unknown; a missing measurement must not become an inferred pass.
  3. Add a failure/review decision in the local session. Inspect the declared QA summary and, if used, label the deterministic probe result as mock evidence.
  4. Export the selected local evidence ZIP. Open it and compare the manifest, review decision, applicable QA report and checksums with what the interface showed.

Input: synthetic episode metadata. Expected artifact: a local evidence archive whose provenance and review decisions can be inspected. Acceptance: the selected episode and decision are represented, missing values stay unknown, and checksums identify the exported files. This does not test a real robot, train a VLA model or deliver a dataset to a remote customer.

The product overview explains the review surface. For a real capture and delivery engagement, the Robotics Dataset Release Pilot describes a separate managed workflow; the local export is not an automatic handoff into that program.

Keep dataset rights and delivery separate from review

Robot records may contain proprietary scenes, participant information or restricted source material. Review only what the program is authorized to inspect and export. A checksum identifies bytes; it does not establish the rights to distribute them or certify their usefulness for training.

The app’s local-first contract matters here. Imported metadata and review state remain local to the selected runtime unless the user deliberately exports them. Dataset records are not attached to ordinary documentation links. The visual source build does not create a hosted dataset database or upload to the disabled network destinations.

Install the surface you actually need

Download the Robotics Studio Open v0.2.0 release and verify its artifact using the release instructions, or inspect the hosted example. The versioned README distinguishes the JavaScript manifest-validator companion from the visual app; the companion does not bundle native adapters or robotics media.

Use the first session to verify the input, unknown-state handling and exported artifact. If a real dataset needs a binary adapter or remote delivery, scope that requirement separately rather than treating a successful synthetic fixture review as proof of support.