Correction Dataset
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/GeWu-Lab/Motion-based-Self-Reflection-Framework
下载链接
链接失效反馈官方服务：
资源简介：
该数据集包含了来自在线人工干预、离线人工标注以及专家演示的修正数据，旨在提升模型在动作修正方面的能力。为了平衡各种失败情况在模型训练中的比例，该数据集进行了筛选。具体数据规模包括3,644次在线干预、7,365次离线标注以及6,378次专家演示。这一系列数据收集工作的目的是为了机器人动作修正任务。
This dataset contains correction data sourced from online human interventions, offline human annotations, and expert demonstrations, aimed at improving the model's performance on action correction tasks. To balance the distribution of various failure cases during model training, this dataset has been filtered. The dataset specifically includes 3,644 online intervention instances, 7,365 offline annotation instances, and 6,378 expert demonstration instances. The entire data collection effort is oriented towards the robot action correction task.
