CausalWorld
收藏资源简介:
CausalWorld是由苏黎世联邦理工学院等机构创建的一个机器人操作基准数据集,旨在促进因果结构和转移学习研究。该数据集包含丰富的环境参数,如机器人和物体的质量、颜色、大小等,允许用户对这些参数进行干预,从而设计不同难度的训练和评估分布。数据集中的任务涉及使用一组给定的块构建3D形状,从简单的单个对象操作到复杂的结构构建,涵盖了广泛的操作技能。CausalWorld不仅支持模拟环境中的学习,还允许将训练策略转移到真实世界,为研究因果学习和机器人操作提供了强大的平台。
CausalWorld is a robotic manipulation benchmark dataset created by ETH Zurich and other institutions, aiming to advance research on causal structure and transfer learning. This dataset includes a rich set of environmental parameters, such as the mass, color, and size of robots and objects, enabling users to intervene on these parameters to design training and evaluation distributions with varying levels of difficulty. Tasks in the dataset involve constructing 3D shapes using a set of given blocks, ranging from simple single-object manipulation to complex structural building, covering a wide spectrum of manipulation skills. CausalWorld not only supports learning in simulated environments but also allows the transfer of trained policies to the real world, providing a powerful platform for research on causal learning and robotic manipulation.




