AuraOne Open / Robotics Studio / Export formats

Export a portable record of the reviewed subset.

Robotics Studio exports local manifests, QA reports, and data cards. Source media is never uploaded silently.

At a glance

Robotics Studio docs

Local-first
Preview
robotics-studio.auraone.ai
Source
github.com/auraoneai/robotics-studio-open
Data
User datasets remain local
Inputs
JSON and JSONL manifests

Documented export formats

Use the current local export formats.

Exports are designed to preserve review decisions while staying portable. A team should be able to leave with the subset, the manifest, the data card, and the evidence needed to reproduce the decision.

Reviewed subset

.json

Selected episode references with their review state and source context.

Large collections

.jsonl

Line-delimited reviewed records for larger episode and clip collections.

Quality reports

.json

Failure-cluster and sensor-QA results linked to the reviewed records.

Evidence archive

.zip

A local archive containing the manifests, the reports, and the review metadata.

Media

path references

Source video, depth, and action files stay local unless the operator packages them.

Large media

Exports stay on your machine.

The exporter writes JSONL records, reports, and a checksum file into a local archive. It reads episode metadata, not source media, and it sends nothing to a hosted service.

Artifact manifest

checksum + source

checksums.sha256 lists the SHA-256 of every file in the archive.

External delivery

operator controlled

Move the exported archive through your approved transfer or object-storage workflow after local review.

Handoff file

Every export explains itself.

The manifest lists every payload in the archive with its role, its size, and its SHA-256 checksum, along with the consent the user gave before export. RLDS, OpenX, HDF5, and LeRobot targets include a JSONL metadata bridge alongside the archive manifest.

{
  "$schema": "https://schemas.auraone.ai/open-studio/intake-packet/v1.json",
  "manifest_version": "1.0.0",
  "product": "robotics-studio-open",
  "created_at": "2026-05-14T21:22:00Z",
  "consent": {
    "user_acknowledged_preview": true,
    "user_acknowledged_transport": true
  },
  "payload_manifest": [
    {
      "path": "payload/reviewed_subset_manifest.json",
      "role": "robotics_reviewed_subset_manifest",
      "sha256": "535553111de1...",
      "size_bytes": 18422
    },
    {
      "path": "payload/sensor_qa_report.json",
      "role": "robotics_sensor_qa_report",
      "sha256": "9b1f04c2a7e3...",
      "size_bytes": 2310
    }
  ]
}

Delivery manifest

Buyer delivery is tracked per object.

In a managed program, the delivery manifest lists each accepted clip and export artifact with its checksum. Buyer access uses a signed, expiring link, and recall or deletion workflows are coordinated per program by the AuraOne delivery team.

Managed exports ship as RLDS, OpenX, HDF5, BVH, JSON, Parquet, or CSV. Other formats, including MCAP, ROS bag, LeRobot, and Zarr, are scoped per program.

Trajectory tables

One row per episode step, in Parquet or CSV.

Managed programs can export trajectories as a flat step table. Each row is one step of one episode. Parquet and CSV carry the same columns.

Parquet

.parquet

Apache Parquet. Dataset context is also written to the file's key-value metadata.

CSV

.csv

RFC 4180 CSV, encoded as UTF-8, with a header row.

Context columns

per row

dataset_id, dataset_name, robot_platform, trajectory_id, session_id.

Step columns

per row

episode_index, step_index, frame_t, frame_payload_json, action_t, action_payload_json.

Payload columns

stringified JSON

frame_payload_json and action_payload_json hold each payload as a JSON string. Parse them after load.

AuraOne Programs

The intake export is explicit.

The open-source build exports portable local archives with zero external network dependencies, ready to transfer through your approved pipeline.