Zitong Shan

Open Egocentric Data Lowers Barriers for Robot Learning

Smartphone capture and a modular processing toolchain turn 2,000 hours of human manipulation video into structured supervision for embodied models.