Robotics Studio Open
Every serious robotics team has a private dataset review tool.
It might be a Streamlit app. It might be a notebook. It might be a custom desktop wrapper around a folder of videos and parquet files. The shape is always similar: list episodes, play videos next to joint state, mark success or failure, write a few tags, export the subset that training should see next.
The workflow is standard. The tooling is not.
Robotics Studio Open is AuraOne's local-first IDE for reviewed teleop and VLA datasets. It opens LeRobot, RLDS, OpenX, HDF5, ROS bag, and folder-of-mp4-jsonl captures. It keeps RGB, depth, joint state, force/torque, audio, and language streams synchronized in one scrubber. It lets reviewers mark action phases, interventions, failures, sensor anomalies, and training decisions.
It also clusters repeated failures, runs VLA robustness probes, generates embodiment cards, and exports reviewed subsets to Hugging Face Hub or local disk.
The integrated robotics OSS surface
Robotics Studio Open brings six AuraOne robotics projects into one daily surface:
lerobot-quality-gatesfor dataset health checksrobot-recovery-benchfor intervention and recovery metricsvla-robustness-kitfor perturbation probesembodiment-cardfor robot and dataset cardsrobotics-reviewkitfor review schemas and taxonomyfailure-galleryfor shareable failure cases
The standalone CLIs remain. The studio is the product surface for daily review.
Local first by design
Robot datasets are large, private, and operationally sensitive. Robotics Studio Open is not a hosted browser editor. It is a local-first desktop app and CLI. The sidecar index, thumbnails, review tags, sensor anomaly notes, and saved views stay with the dataset unless the user explicitly exports them.
That local-first boundary also keeps the commercial line clean. Robotics Studio Open is single-user and MIT licensed. Multi-reviewer queues, hosted dataset storage, approval chains, audit ledgers, RBAC, SSO, and managed reviewer pools belong to Robotics Studio Cloud, Robotics Studio Enterprise, and AuraOne Robotics Programs.
What the first workflow looks like
Open a dataset. Browse episodes without loading every video into memory. Scrub the synchronized streams. Tag a failure. Cluster similar failures. Run sensor QA. Generate an embodiment card. Export a reviewed subset to Hugging Face or create a privacy-previewed AuraOne Programs intake packet.
That is the loop robotics teams keep rebuilding internally.
Now it has a public surface.
Watch the caption-only launch demo at /open/robotics-studio/demo-caption-only.mp4.
Launch the hosted preview at robotics-studio.auraone.ai, or start from the product page at Robotics Studio Open.