三层箱体双臂定点抓取训练数据集

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 synchronized using timestamps and then uniformly stored. The data collection workflow consists of three core stages: task definition, teleoperation recording, as well as data integrity verification and annotation review. Three-tier quality control is implemented, including frame rate verification, logical consistency inspection, and file integrity verification. The data is archived in a structured manner based on task IDs and version numbers, with original ROS2 topic messages preserved for retrospective analysis. It complies with physical safety and data management standards, and is suitable for imitation learning algorithm training across diverse scenarios.

搜集汇总

背景与挑战

背景概述

该数据集专为双臂工业机器人在三层箱体定点抓取场景设计，适用于工业分拣和仓储物流中的视觉识别与协同规划算法训练。它包含40万条通过手柄遥操作采集的数据，覆盖多相机视觉流和机器人状态，并经过帧率验证和逻辑一致性检查等质量管控，每日更新以支持模仿学习。数据处理采用伪匿名化和加密等安全措施，确保原始信息无法还原，适用于解决深层遮挡识别和双臂协同抓取规划等复杂问题。