July 15, 2026 HumanScale: Egocentric Human Video Can Outperform Real-Robot Data for Embodied Pretraining 系统研究表明,经过精心筛选和标注的拟人视角(egocentric)人类视频作为具身基础模型预训练数据源,可显著超越遥操作真实机器人数据 embodied-learning vision-language world-action-model