DexMimicGen Datasets

该数据集名为DexMimicGen，它基于少量的人类演示生成双手灵巧操作任务的合成数据集。这些数据集涵盖了多种操作行为，包括整理抽屉、穿线和部件组装等任务，并展示了在DexMimicGen数据集上训练时，策略性能的显著提升。规模上，该数据集从60个源人类演示中生成了21,000个示范，覆盖了各种操作任务。其研究任务专注于双手灵巧操作。

The dataset, named DexMimicGen, generates synthetic datasets for bimanual dexterous manipulation tasks based on a small number of human demonstrations. It covers a diverse range of manipulation behaviors, including tasks such as drawer tidying, threading, and part assembly, and demonstrates that training policies on the DexMimicGen dataset yields significant performance improvements. In terms of scale, this dataset generates 21,000 demonstrations from 60 source human demonstrations, spanning various manipulation tasks, with its research focus centered on bimanual dexterous manipulation.