eval_results
收藏资源简介:
该数据集使用LeRobot框架创建,是一个面向机器人学任务的多模态数据集,核心内容是机器人follower的交互数据。数据以parquet文件格式存储,并包含对应的mp4格式视频文件(视频帧率为30fps)。数据集结构包括:1) 动作(action):一个6维浮点向量,表示机器人六个关节(肩部平移、肩部抬升、肘部弯曲、腕部弯曲、腕部旋转、夹爪)的目标位置;2) 观测状态(observation.state):一个6维浮点向量,表示上述六个关节的当前实际位置;3) 视觉观测(observation.images.camera1 和 camera2):来自两个摄像头的RGB视频流,每个视频帧的分辨率为480像素(高)x 640像素(宽)x 3通道;4) 元数据索引:包括时间戳(timestamp)、帧索引(frame_index)、回合索引(episode_index)、数据索引(index)和任务索引(task_index),用于标识和组织数据序列。该数据集适用于机器人模仿学习、强化学习、状态估计、多传感器融合等研究,提供了机器人的关节状态、控制指令和同步的视觉观测。
This dataset, constructed using the LeRobot framework, is a multimodal dataset tailored for robotics tasks, with its core content being the interaction data of a robotic follower. The data is stored in Parquet file format, accompanied by corresponding MP4 format video files with a frame rate of 30 FPS. The dataset structure includes: 1) Action: a 6-dimensional floating-point vector representing the target positions of the robot's six joints (shoulder translation, shoulder elevation, elbow flexion, wrist flexion, wrist rotation, and gripper); 2) Observation State (observation.state): a 6-dimensional floating-point vector representing the current actual positions of the aforementioned six joints; 3) Visual Observations (observation.images.camera1 and camera2): RGB video streams from two cameras, with each video frame having a resolution of 480 (height) × 640 (width) × 3 channels; 4) Metadata Index: including timestamp, frame_index, episode_index, data index (index), and task_index, which are used to identify and organize data sequences. This dataset is applicable to research in robot imitation learning, reinforcement learning, state estimation, multi-sensor fusion and other fields, providing the robot's joint states, control commands and synchronized visual observations.




