超市货架补货训练数据集

内部数据采集通过手柄遥操作完成，覆盖多相机视觉流、机器人本体状态及操作指令，各传感器数据经时间戳同步后统一存储。采集流程包含任务定义、遥操作录制、数据完整性校验、标注审核三个核心环节，并实施帧率验证、逻辑一致性检查及文件完整性校验三重质量管控。数据按任务ID与版本号结构化归档，保留原始ROS2话题消息以备回溯，符合物理安全与数据管理规范，适用于多场景下的模仿学习算法训练。

Internal data collection is conducted via handle-based teleoperation, covering multi-camera visual streams, robot body states, and operation commands. All sensor data is uniformly stored after being synchronized using timestamps. The data collection workflow encompasses three core stages: task definition, teleoperation recording, data integrity verification, and annotation review, and implements three-tier quality control measures including frame rate verification, logical consistency inspection, and file integrity verification. The data is archived in a structured format based on task ID and version number, with raw ROS2 topic messages retained for retrospective traceability. It complies with physical safety and data management specifications, and is applicable to training imitation learning algorithms across various scenarios.

搜集汇总

背景与挑战

背景概述

该数据集是针对超市货架补货场景的训练数据，通过手柄遥操作采集，包含多相机视觉流、机器人状态和操作指令，数据经过时间戳同步和质量管控。它采用结构化归档和原始ROS2消息保留，适用于多场景下的模仿学习算法训练，强调数据完整性和安全性。