GSO

该数据集被用于训练在打包已知物体上的策略，并适应于打包未知物体。此外，它还旨在实现一个任务：在分布外泛化，即训练策略能够应对在训练期间未见过的物体。

This dataset is developed for training packing policies on known objects, while facilitating adaptation to packing unseen objects. Furthermore, it targets the task of out-of-distribution (OOD) generalization, which requires the trained policy to handle objects that were not encountered during the training process.