Devices
Mono, stereo, wrist, custom rigs, and future contact capture devices will all become part of one capture system.
Vision
Human embodiment data, collected at fleet scale.
OpenRobot builds capture devices, a central unit, and an SDK for turning new physical-AI data requirements into synchronized human embodiment datasets.
Request Device Sample01 / The loop
Fleet Learning Moves Beyond Roads.
Tesla FSD compounded because cars already had human drivers, sensors, and a fleet path that turned real driving into training data. The useful analogy is the loop: a real-world activity, an instrumented capture surface, and model updates fed by what happened outside the lab.
Physical AI needs a similar loop for human work: head perspective, hand action, object contact, timing, recovery, and context captured at fleet scale without making collection intrusive.
02 / Research to fleet
Hardware That Follows Research.
When a robotics paper proves that a new signal matters, the next training run should not wait quarters for a one-off rig. The requirement has to become data spec, hardware spec, and deployed capture devices fast enough to feed the next world/action-model training run.
The compounding move is hardware that follows research. OpenRobot owns the device iteration path, so we can design, manufacture, qualify, and update capture hardware around new data requirements while making the collection path as non-invasive as possible.
The OpenRobot stack
View devicesCentral unit
A shared system for sync, power, metadata, session control, and multi-device capture.
SDK coming soon
Capture APIs, synced-streams.