Ego2Robot: Scalable Robot Data Synthesis from Egocentric Human Data

Ego2Robot presents a scalable pipeline for converting egocentric human manipulation videos into robot training data through action retargeting, robot-arm visual synthesis, and multi-level quality curation. The authors report 18,561 hours of synthesized data across 15 robot morphologies, using both curated datasets and in-the-wild videos. Evaluation extends RoboTwin2.0 with disentangled perturbations in visual appearance, scene layout, embodiment morphology, and task semantics. Joint pretraining with synthesized and real robot data consistently improves out-of-distribution generalization across multiple perturbation types, with reported validation on real-robot deployment.