Input
What you provide
- Robot sessions or dataset request
- RLDS, OpenX, HDF5, BVH, JSONL, video, labels, or Parquet
- Task brief and robot skill
- Safety, consent, license, or use constraints
App Data / Robotics
Robotics Dataset Prep
Turn robot runs into a dataset your team can use.
Send robot sessions, dataset requirements, task context, and delivery rules. The app organizes the runs, separates failed or unsafe sessions, and prepares the dataset handoff. Ready to scope now. Start with one real batch and get back the output your team can use.
Available nowAvailable now. Start with one robotics dataset.
Input, work, and output
Work
What happens
- Organizes episodes by task, source, and export format
- Separates questionable captures and missing context
- Checks delivery files, checksums, and dataset card state
- Carries failure modes into the next collection run
Output
What you receive
- Usable robot episode set
- Dataset card and delivery file
- Checksums and artifact list
- Failure cases for the next collection batch
Decision flow
Move from intake to review and handoff with clear owners at every step.
01
Scope the dataset
Define the robot skill, dataset name, category, and export format.
The request becomes an app run.
02
Load the sessions
Attach the robot runs and labels your team wants prepared.
Episodes stay tied to the request.
03
Separate bad runs
Questionable captures are held out instead of silently entering the dataset.
Missing context is visible.
04
Deliver the dataset
Approved runs ship with delivery files, checksums, and failure memory.
The dataset is easier to inspect.
What you receive
Robotics Dataset Prep deliverables
- Application
- Robotics Dataset PrepAvailable now
- Buyer
- Robotics AI and data teamsProgram owner
- Input
- robotics sessionRLDS / OpenX / HDF5 / BVH / JSONL handoff / MP4 or MOV + data card / CSV labels / Parquet
- Review
- Robotics data ownerOrganizes episodes by task, source, and export format Separates questionable captures and missing context Checks delivery files, checksums, and dataset card state Carries failure modes into the next collection run
- Output
- Usable robot episodes, data card, handoff file, checksums, and failure cases for the next test run.Checked episode data, consent and license records, delivery files, and failure memory your robotics team can keep.
- Scope
- $75K-$250K typical engagementTypical engagement
What the workflow supports
| Outcome | Work | What you receive | Program fit |
|---|---|---|---|
| Input is ready | Robot sessions or dataset request RLDS, OpenX, HDF5, BVH, JSONL, video, labels, or Parquet Task brief and robot skill Safety, consent, license, or use constraints | Source, access, format, validation, project rules, and responsible owner. | Best for teams with a defined source, owner, and delivery goal. |
| Work can advance | Organizes episodes by task, source, and export format Separates questionable captures and missing context Checks delivery files, checksums, and dataset card state Carries failure modes into the next collection run | Stage, owner, checks, exceptions, reviewer notes, and required next action. | Your team can see blockers, owners, and next actions throughout the workflow. |
| Output can be released | Usable robot episode set Dataset card and delivery file Checksums and artifact list Failure cases for the next collection batch | Sources, criteria, reviewers, decision, manifest, version, and audit history. | The final package is configured for the application, project, and review requirements. |
Good fit
What this application is for
- VLA training data prep
- Teleop batch cleanup
- Robot skill dataset handoff
- Failed-run analysis
Boundary
What this application is not
- Controlling robots in real time
- Guaranteeing a policy will perform in the field
Availability
Available nowAvailable now for scoped dataset batches with clear source files and delivery format.