FPV Labs · Documentation
Stera.
Open infrastructure for Physical AI.
The next generation of embodied models needs more than pixels. It needs motion, depth, pose, and environment — captured together, in sync, at scale. Stera turns the iPhone in your pocket into a spatial-data capture system, and gives you the open-source tooling to read, process, and export the result for training.
What is Stera?
An end-to-end data platform for Physical AI. You record on an iPhone Pro, the SDK reads the recording, runs refinement and detection passes, and exports a self-contained episode you can drop into a training pipeline.
Why open?
Robotics datasets are locked away or built on closed hardware. Stera runs on commodity iPhones, ships an open Python SDK, and exports standard formats (MCAP, HDF5, MP4, PLY) so the data lives outside any single vendor.
Who is it for?
Researchers collecting demonstration data for imitation learning, teams labeling real-world manipulation, and anyone who needs synchronized RGB, depth, pose, and hand tracking without building the stack from scratch.
What's in these docs
Stera is built around four stages. Each section of the docs covers one.
01CaptureRecord with the Stera App. Review and approve on the Dashboard.02ProcessThe Python toolkit for reading, refining, and visualizing recordings.03EvaluateOne call generates an interactive HTML report with per-stream metrics and a health score.04ExportBundle a session into an episode directory ready for training.
Jump in
Install the SDKpip install stera-sdk, system deps, optional extras.QuickstartA 25-line pipeline: read a recording, blur faces, track hands, export.Record your first sessionMount, calibrate, hit record. From zero to first MCAP in five minutes.Output formatsWhat Stera writes — MCAP in, HDF5 + MP4 + PLY out.QC a recordingEvaluate(session).show() — one call to an interactive HTML report.