Spatial reconstruction
Capture the actual place a robot will be deployed and rebuild it as a physically grounded simulation, so training and testing happen in the setting that matters.
Coming soon
01 / What it will do
How it will work.
- 01CaptureScan the deployment site with commodity sensors: phone, handheld scanner or robot-mounted cameras.
- 02ReconstructRecover geometry, layout and materials, then segment the space into objects that can be simulated.
- 03Vary and trainGenerate variations of the rebuilt site, from layout and clutter to lighting and object placement, and use them to train and test robot policies.
02 / Why it matters
Why we are building it.
A policy trained in a generic room learns a generic room. Reliability is earned in the setting where the robot will actually run, with its clutter, its lighting and its tolerances.
Early access
Want spatial reconstruction early?
We are working with a small number of teams as we build it. Join the waitlist to be among the first to use it.