move_clutter_camera_goal
收藏资源简介:
该数据集基于LeRobot框架创建,专为机器人操控学习任务设计。数据包含机器人执行多任务(共5个任务)的演示数据,涵盖631个episode,总计481915帧。数据形式包括12维动作指令(如目标夹爪相机位置、锚点位置、腕部偏移、手指速度等)、20维机器人状态观测(如速度、位置、夹爪压力、激光测距仪、目标力、障碍物方位等),以及三路224x224彩色视频流(夹爪相机、锚点相机0、锚点相机1),视频编码为AV1,帧率30fps。此外,还包含时间戳、帧索引、episode索引、任务索引以及锚点姿态(12维旋转和平移)。数据文件以Parquet格式存储,视频文件以MP4格式存储。机器人类型为stringman。数据集适用于模仿学习、强化学习等机器人操控任务,如移动杂乱物体、到达目标等场景。
This dataset is created based on the LeRobot framework and is specifically designed for robot manipulation learning tasks. The data contains demonstration data of a robot performing multiple tasks (5 tasks in total), covering 631 episodes with a total of 481,915 frames. The data format includes 12-dimensional action commands (e.g., target gripper camera position, anchor position, wrist offset, finger velocity, etc.), 20-dimensional robot state observations (e.g., velocity, position, gripper pressure, laser rangefinder, target force, obstacle orientation, etc.), and three 224x224 color video streams (gripper camera, anchor camera 0, anchor camera 1) encoded in AV1 at 30fps. Additionally, it includes timestamps, frame indices, episode indices, task indices, and anchor poses (12-dimensional rotation and translation). The data files are stored in Parquet format, and video files are stored in MP4 format. The robot type is stringman. The dataset is suitable for robot manipulation tasks such as imitation learning and reinforcement learning, including moving cluttered objects, reaching targets, and other scenarios.
数据集概述:naavox/move_clutter_camera_goal
基本信息
- 许可证:Apache-2.0
- 任务类别:机器人学(Robotics)
- 创建工具:基于 LeRobot 框架(https://github.com/huggingface/lerobot)创建
- 机器人类型:stringman
- 数据集规模:总计 631 个片段(episodes),481,915 帧,5 个任务
- 数据划分:训练集包含全部 631 个片段(0:631)
数据格式
- 帧率:30 FPS
- 数据文件:Parquet 格式,存储于
data/*/*.parquet - 视频文件:MP4 格式,使用 AV1 编码,分辨率 224x224,RGB 三通道
- 文件大小:数据文件约 100MB,视频文件约 200MB
特征结构
动作(Action)
12 维浮点数组,包含:
- 目标夹爪相机坐标(goal_gripper_cam_x/y/z)
- 两个锚点目标坐标(goal_anchor_0_x/y/z, goal_anchor_1_x/y/z)
- 腕部偏移(wrist_offset)、手指速度(finger_speed)及片段结束标志(episode_end)
观测状态(Observation State)
20 维浮点数组,包含:
- 速度信息(vel_x/y/z、wrist_speed、finger_speed)
- 夹爪位姿(gripper_pos_x/y/z、spin、finger_angle)
- 传感器数据(激光测距仪、手指压力、腕部角度、目标力)
- 环境目标方位(hamper_bearing/distance、toybox_bearing/distance、trashcan_bearing/distance)
观测图像(Observation Images)
三个同步视频流,均为 224x224x3:
gripper_camera(夹爪相机)anchor_camera_0(锚点相机 0)anchor_camera_1(锚点相机 1)
辅助信息
- anchor_poses:12 维浮点数组,表示两个锚点的位姿(旋转和平移)
- 时间戳:timestamp(float32)
- 索引字段:frame_index、episode_index、index、task_index(均为 int64)
引用信息
该数据集的 BibTeX 引用信息暂未提供(标记为 [More Information Needed])。




