Read original
HuggingFace Daily PapersYe WangPapers90

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.

Why it's worth reading

Robot foundation models remain constrained by the scale and diversity of real demonstrations. Ego2Robot is timely because it combines 18,561 hours of synthetic data with cross-morphology and disentangled OOD evaluation, making its data-generation and validation choices especially important to inspect.

Tags

机器人学习VLA数据合成模仿学习操作策略分布外泛化RoboTwin2.0具身智能