NYU VINN
收藏资源简介:
该数据集由纽约大学(NYU)的研究团队创建,旨在支持视觉模仿学习的研究。数据集包含三个主要任务的数据:推动物体、堆叠物体和打开门。其中,推动物体和堆叠物体任务的数据分别包含约 750 和 930 条演示轨迹,而打开门任务的数据包含 71 条训练演示和 21 条测试演示。数据来源为人类专家通过遥操作机器人收集的视觉演示数据,涵盖了多种背景和物体的场景。数据集的创建过程基于遥操作工具收集的视频数据,通过结构化运动(SfM)方法重建三维运动轨迹,提取动作标签。该数据集的应用领域主要集中在机器人视觉模仿学习,旨在提高机器人在复杂视觉环境中的任务执行能力,特别是在数据量有限的情况下实现更好的泛化性能。
This dataset was created by a research team at New York University (NYU) to support research in visual imitation learning. It contains data for three core tasks: object pushing, object stacking, and door opening. Specifically, the datasets for the object pushing and object stacking tasks include approximately 750 and 930 demonstration trajectories respectively, while the door opening task dataset has 71 training demonstrations and 21 test demonstrations. The data originates from visual demonstration data collected by human experts via teleoperated robots, covering scenarios with diverse backgrounds and objects. The dataset is constructed based on video data collected through teleoperation tools, where 3D motion trajectories are reconstructed and action labels are extracted using the Structure from Motion (SfM) method. The main application scope of this dataset focuses on robotic visual imitation learning, aiming to enhance robots' task execution capabilities in complex visual environments, particularly to achieve better generalization performance under limited data conditions.




