Natura-1 reportedly uses human hand footage to control robot hands

A post describes Nature Will's Natura-1 as trained on 1 million hours of first-person footage of people using their hands, then used to control real robot hands. It says the model handles fine two-handed control after a grasp, new tasks without extra training, and long sequences requiring memory and recovery from mistakes. The author argues human footage offers a faster route to scale than collecting robot demonstrations, provided the skills transfer to robot hands.