An open-hardware, head-mounted stereo-inertial capture device for egocentric data collection in the wild - under USD 200 in commodity parts - released with 550 hours of stereo data captured across 100+ environments.
We are open-sourcing Stera 2.0: a capture stack for collecting high-fidelity multimodal RGB-D data from commodity mobile phones. The mobile app, native recording engine, upload backend, and white-label tooling let any lab own its own capture stack.
Our production-grade system for fine-grained action labeling of raw robot and human egocentric video. Vidur sets a new state of the art on WGO-Bench and beats Gemini 3.5 Flash on EgoExoLearn at lower cost.
Benchmarking ARKit pose for Stera against a 24-camera Vicon system and over hour-long sessions: metric to within centimetres, sub-degree, and drift-bounded under 1 cm across full-home captures.
A full-stack platform for high-fidelity multimodal data capture and Stera-10M open-source dataset collected on commodity hardware.
Why the sensor-level choice between rolling and global shutter matters deeply for egocentric SLAM, and what can be done about it.
Why long-horizon state tracking in egocentric manipulation is harder than it looks, and why it matters for robot learning.
FPV Labs is building the full-stack infrastructure for capturing, processing, and transferring human experience into machine knowledge.