Real and Simulated Grasp Data

该数据集由KUKA LBR iiwa机器人收集的成功抓握数据组成，涵盖了真实世界和模拟环境。数据集包含了具有不同视觉属性和大小的物体图像，以及相应的抓握动作。此外，该数据集支持对以物体为中心的表示进行测试，提供了从肩上视角的单目RGB观察，这使得可以在没有空间监督的情况下评估抓握策略和物体定位。该数据集的规模包括15,000个模拟抓握和43.7万个真实世界抓握。其任务涵盖了物体召回、物体定位以及针对抓握任务的强化学习。

This dataset consists of successful grasping data collected by a KUKA LBR iiwa robot, covering both real-world and simulated environments. It includes images of objects with varying visual properties and sizes, paired with their corresponding grasping motions. Additionally, this dataset supports testing of object-centric representations, providing monocular RGB observations from a shoulder-mounted perspective, which enables evaluation of grasping strategies and object localization without spatial supervision. The dataset has a scale of 15,000 simulated grasps and 437,000 real-world grasps. Its supported tasks include object recall, object localization, and reinforcement learning for grasping tasks.