Alkatt/eval_LAVLA_Async_VI_100M_40k_b20_whiteblackcube_v2_2026_05_27_white_latency_bench_v3
收藏资源简介:
该数据集使用LeRobot创建,专注于机器人任务,具体针对so_follower类型机器人。数据集包含10个完整的情节,总计8468帧,数据以parquet文件形式存储,视频文件为MP4格式,帧率为30fps。数据集结构包括动作特征(如肩部平移、肩部提升、肘部弯曲、腕部弯曲、腕部旋转和夹爪位置,共6个浮点值)、观测状态特征(与动作相同的6个关节位置)、三个摄像头的图像观测(每个摄像头提供480x640分辨率、3通道的视频数据),以及时间戳、帧索引、情节索引、索引和任务索引等元数据。数据集总数据文件大小为100MB,视频文件大小为200MB,适用于机器人控制和学习任务。
This dataset was created using LeRobot and focuses on robotics tasks, specifically for the so_follower robot type. It contains 10 total episodes with 8468 frames, stored in parquet files, and video files in MP4 format at 30fps. The dataset structure includes action features (e.g., shoulder pan, shoulder lift, elbow flex, wrist flex, wrist roll, and gripper positions, with 6 float32 values), observation state features (same 6 joint positions as actions), image observations from three cameras (each providing video data at 480x640 resolution with 3 channels), and metadata such as timestamp, frame index, episode index, index, and task index. The total data file size is 100MB, and video file size is 200MB, suitable for robotics control and learning tasks.




