Outputs: Continuous-value action vectors mapped to the robot's degrees of freedom

Validated across loco-manipulation, tabletop manipulation, and dexterous bimanual tasks on Unitree G1, Bimanual Manipulator YAM, and AGIBot Genie 1.

The central research that has been used for GR00T N1.7 is EgoScale — pre-training on 20,854 hours of human egocentric video spanning 20+ task categories, from manufacturing and retail to healthcare and home environments. This is a significant step up from the few thousand hours of robot teleoperation data used to train N1.6.

The intuition: humans and robots share similar embodiments — two hands, a first-person viewpoint, a world full of objects to manipulate. Training on sensorized human video (ego cameras, wrist cameras, hand tracking) gives the model rich manipulation priors without requiring every behavior to be demonstrated on a physical robot first. It moves pre-training beyond what teleoperation can scale to.