Robogym

该数据集在Robogym环境中实现了非对称自我对弈的训练和评估，其中包含了随机放置的ShapeNet物体作为初始状态分布。此外，该数据集支持对未见任务泛化能力的评估，并包含了不同复杂程度的环境。任务涵盖了机器人操作的各种任务，包括推动、翻转、捡起以及堆叠积木。

This dataset implements asymmetric self-play training and evaluation within the Robogym environment, where randomly placed ShapeNet objects serve as the initial state distribution. Furthermore, this dataset supports evaluation of generalization performance on unseen tasks and encompasses environments with varying complexity levels. The included tasks cover a wide range of robotic manipulation operations, including pushing, flipping, picking up, and stacking blocks.