bridge_v2_lerobot

该数据集是使用LeRobot框架创建的机器人数据集，专门针对widowx机器人。数据集包含大量机器人操作相关的数据，如观察状态（包括位置、姿态和夹持器状态）、动作指令、多摄像头图像（每帧256x256分辨率，3通道，5fps）、时间戳和各种索引信息。数据集结构庞大，包含53192个episodes、1999410帧、19974个任务和212768个视频，分为54个块，每个块1000个episodes。数据以parquet格式存储，视频采用av1编码。

This robotic dataset was developed using the LeRobot framework, specifically designed for the WidowX robot. It contains a substantial amount of data related to robotic manipulation, including observation states (comprising position, orientation, and gripper status), action commands, multi-camera images (256×256 resolution per frame, 3 channels, 5 fps), timestamps, and various indexing information. The dataset has a large-scale structure, consisting of 53,192 episodes, 1,999,410 frames, 19,974 tasks, and 212,768 videos. It is divided into 54 chunks, with each chunk containing 1,000 episodes. The data is stored in Parquet format, and all videos are encoded using the AV1 codec.

搜集汇总

背景与挑战

背景概述

bridge_v2_lerobot是一个基于LeRobot框架构建的机器人操作数据集，专门用于widowx机器人，包含大量episodes、帧和视频数据，覆盖观察状态、动作指令和多摄像头图像。数据集结构庞大，以parquet格式存储，视频采用av1编码，适用于机器人学习和模拟任务。