Yotta
Data labelling, quality, and inspection for versatile humanoid robotics datasets. We ensure the quality is high and useful labels are included to provide a strong base infrastructure for training or data curation and exploration.
Open pipeline: yotta-mcap
· Python 3.11+
Where we spend our time
A narrow scope, executed end-to-end: labels you can trust, inspection you can repeat, and artifacts downstream teams can adopt.
-
Labelling
Structured labels and taxonomies suited to humanoid teleop, video, and state–action traces so teams can iterate without losing provenance.
-
Quality & inspection
Timeline-native review, validation, and QA passes so segments stay consistent, non-overlapping, and defensible for partners and internal ML.
-
Training-ready handoff
Normalized exports and sidecar workflows that slot into training stacks, curation tools, and exploration notebooks without one-off glue code.
Technical walkthrough (tooling)
The walkthrough is a hands-on pass over the open pipeline we use day to day: ingest, sidecar annotations, validation, and export. If you want the README-level detail first, start with the docs link in the header or footer.
-
MCAP-native inspection
Understand recording bounds, topics, and structure before labelling starts.
-
LeRobot → review timelines
Ingest Hugging Face–style snapshots and align them with Foxglove-friendly MCAP for scrubbing and QA.
-
Validated sidecars
Keep raw recordings intact while annotations and exports evolve with your taxonomy.
Works with your stack
- MCAP Robotics recordings
- Foxglove Timeline review
- Hugging Face LeRobot-style data
- Python 3.11+ CLI workflows
Book a technical walkthrough
Fill in a few details, then open your email client to message our team. Nothing is sent from this page until
you send the message yourself. For one-click submit from the browser instead, set PUBLIC_FORMSPREE_FORM_ID
at build time (see website/.env.example).