eval_grab_cube_v4
收藏资源简介:
该数据集是一个用于机器人学习的交互式数据集,使用 LeRobot 工具创建。数据集专注于一个具体的机器人操作任务(推测为抓取立方体,根据数据集名称推断,但 README 未明确说明)。数据集包含 4 个完整的任务执行序列(episodes),共计 3593 个时间步(frames)。数据以多模态形式组织,核心特征包括:机器人的 6 维关节位置动作指令(action)、对应的 6 维关节位置状态观测(observation.state)、以及来自三个不同视角(侧视、前视、顶视)的 RGB 图像观测(observation.images.side/front/top)。所有图像分辨率为 640x480,以 15 FPS 的 H.264 编码视频格式存储。此外,数据还包含时间戳、帧索引、序列索引等元数据。数据集以 Apache-2.0 许可证发布,适用于机器人模仿学习、强化学习、视觉运动策略学习等研究任务。
This dataset is an interactive dataset for robot learning, created using the LeRobot tool. It focuses on a specific robot manipulation task (inferred as grasping a cube based on the dataset name, but not explicitly stated in the README). The dataset contains 4 complete task execution episodes, totaling 3593 time steps (frames). The data is organized in a multimodal format, with core features including: 6-dimensional joint position action commands for the robot (action), corresponding 6-dimensional joint position state observations (observation.state), and RGB image observations from three different perspectives (side, front, top) (observation.images.side/front/top). All images have a resolution of 640x480 and are stored in H.264 encoded video format at 15 FPS. Additionally, the data includes metadata such as timestamps, frame indices, and episode indices. The dataset is released under the Apache-2.0 license and is suitable for research tasks such as robot imitation learning, reinforcement learning, and visual-motor policy learning.




