Dexterity-BEV：时空对齐机器人操作数据集

The Dexterity-BEV Dataset is a spatiotemporally aligned robot manipulation dataset designed for generalizable and cross-embodiment robot policy learning. It aligns multi-view visual observations, robot proprioceptive states, and actions within a shared 3D coordinate frame, and constructs canonical Bird’s-Eye-View representations to reduce variations caused by different camera configurations, robot embodiments, and data sources. Together with a standardized spatial and temporal alignment pipeline, the dataset enables consistent training and evaluation across heterogeneous robots, viewpoints, operators, and trajectory datasets.