robot-learning-team43/so101_HQ_merged_diverse_pos_pi05_relative

该数据集是一个用于机器人学习的多模态数据集，专注于机器人控制任务。它包含233个episodes，总计255,511帧，帧率为30fps。数据集特征包括：动作（action）和观察状态（observation.state），均为6维浮点向量，对应机器人的6个关节位置（肩部平移、肩部提升、肘部弯曲、腕部弯曲、腕部旋转和夹爪位置）；前视摄像头图像（observation.images.front），分辨率为480x640，3通道（RGB），以30fps的av1编码视频格式存储；此外还有时间戳、帧索引、episode索引、索引和任务索引等元数据。数据以parquet文件组织，视频以mp4格式存储，总数据大小约100MB，视频大小约200MB。数据集由LeRobot工具创建，适用于机器人控制和强化学习研究，任务类别为robotics，标签为LeRobot。This dataset is a multimodal dataset for robotics learning, focusing on robot control tasks. It contains 233 episodes with a total of 255,511 frames at a frame rate of 30fps. The dataset features include: action and observation.state, both as 6-dimensional float vectors corresponding to the positions of 6 robot joints (shoulder pan, shoulder lift, elbow flex, wrist flex, wrist roll, and gripper); front camera images (observation.images.front) with a resolution of 480x640, 3 channels (RGB), stored as 30fps av1-encoded videos; and metadata such as timestamp, frame index, episode index, index, and task index. The data is organized in parquet files, with videos in mp4 format, total data size approximately 100MB and video size approximately 200MB. The dataset was created using the LeRobot tool and is suitable for robotics control and reinforcement learning research, with task category robotics and tag LeRobot.