Device, a vision-language-action (VLA) foundation model designed to run locally on robot hardware. The model features ...

The new training method doesn't use sensors or onboard control tweaks, but a single camera that watches the robot's movements ...