Mimicgen environments
收藏arXiv2025-09-30 收录
下载链接：
https://flowbasedpolicies.github.io/
下载链接
链接失效反馈官方服务：
资源简介：
该数据集包含了1000个合成的演示数据，这些数据是从原始数据集中改编的10个原始演示生成的，以便与ActionFlow模型兼容。原始数据集提供的是单个观测向量，并在世界坐标系中表示动作位移。经过改编后，数据集在世界坐标系中分别表示物体姿态和动作姿态。规模上，该数据集包含了从10个原始演示扩展的1000个演示。该数据集的任务是评估机器人操作任务。
This dataset consists of 1,000 synthesized demonstration samples, which are generated by adapting 10 original demonstrations sourced from the raw dataset to ensure compatibility with the ActionFlow model. The raw dataset originally provided single observation vectors and represented action displacements in the world coordinate system. Following adaptation, this dataset respectively represents object poses and action poses in the world coordinate system. In terms of scale, this dataset includes 1,000 demonstrations expanded from the initial 10 original ones. This dataset is designed for evaluating robotic manipulation tasks.
