桌面纸巾整理训练数据集

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 consists of three core phases: task definition and teleoperation recording, data integrity verification, and annotation review, and implements three-tier quality control measures including frame rate verification, logical consistency check, and file integrity verification. The data is archived in a structured manner based on task ID and version number, with original ROS2 topic messages retained for retrospective analysis. It complies with physical safety and data management specifications, and is applicable to the training of imitation learning algorithms in diverse scenarios.

搜集汇总

背景与挑战

背景概述

该数据集是一个专门用于训练服务机器人在桌面环境中进行纸巾整理任务的数据集，包含40万条通过手柄遥操作采集的多传感器数据，每天更新。它适用于模仿学习算法，旨在解决纸巾盒小目标识别、盒盖开启动作规划、柔性材料抓取稳定性等具体问题，数据经过严格的质量校验和加密处理，确保安全性和不可逆性。