转运机器人多层级融合定位数据
收藏资源简介：
该数据集面向转运机器人多源融合定位精度提升的研究需求建设，核心背景是在实际工业转运场景中，单一传感器定位方案存在明显的场景局限性：激光雷达在低特征场景易漂移，视觉定位受光照与遮挡影响大，磁导航易受金属工件干扰，而多源传感器融合定位通过整合各传感器的优势特征，弥补单一传感器的不足，是实现工业场景下轨道对接高精度定位的核心技术路径。其资源来源于前三个数据集的激光雷达、视觉、磁传感器原始数据，通过多层级融合算法处理生成，确保了数据来源的一致性与关联性，为融合算法研发提供了基础数据支撑。产生方法经过严格的标准化设计，主要分为三个核心步骤：首先开展各传感器数据的时间同步处理，基于ROS的时间同步机制，将激光、视觉、磁传感器的数据时间戳统一到同一时间基准，确保时间同步误差≤5ms，为多源数据的时空配准奠定基础；其次进行传感器外参校准，采用手眼标定方法精准获取激光雷达、双目相机、磁传感器相对于机器人本体坐标系的外参矩阵，校准精度误差≤0.1cm，保障不同传感器数据在空间坐标系上的一致性；最后基于各传感器在不同工况下的定位精度表现，建立精度评估模型，动态分配各传感器的融合权重，通过加权融合算法计算机器人的精准位姿。采集过程覆盖了低特征场景、光照波动、电磁干扰、轻微遮挡等多种典型工况，全面验证融合算法在不同场景下的性能。主要内容包含融合后的机器人位置坐标、姿态角、各传感器坐标系与机器人本体坐标系的转换矩阵、各传感器的置信度评估信息、融合过程日志、原始传感器数据与融合结果的对比数据等。数据格式为ROS标准.bag格式，数据量为73.49MB，包含近50组不同工况下的多源融合定位数据序列。
This dataset was developed to address the research needs of improving multi-source fusion positioning accuracy for industrial transfer robots. The core background lies in the obvious scene limitations of single-sensor positioning schemes in actual industrial transfer scenarios: LiDAR is prone to drift in low-feature scenes, visual positioning is greatly affected by lighting and occlusion, and magnetic navigation is easily disturbed by metal workpieces. Multi-source sensor fusion positioning, which integrates the advantageous features of each sensor and compensates for the shortcomings of single sensors, is the core technical path to achieve high-precision track docking positioning in industrial scenarios. This dataset is generated by processing the raw LiDAR, vision and magnetic sensor data from the first three datasets via a multi-level fusion algorithm, ensuring the consistency and correlation of data sources and providing basic data support for the development of fusion algorithms. Its generation process has undergone strict standardized design, which is mainly divided into three core steps: First, time synchronization processing of sensor data: Based on the ROS time synchronization mechanism, the timestamps of LiDAR, vision and magnetic sensor data are unified to the same time benchmark, with the time synchronization error ≤ 5 ms, laying a foundation for spatial-temporal registration of multi-source data; Second, sensor extrinsic parameter calibration: The hand-eye calibration method is used to accurately obtain the extrinsic matrices of LiDAR, binocular camera and magnetic sensor relative to the robot body coordinate system, with a calibration accuracy error ≤ 0.1 cm, ensuring the consistency of different sensor data in the spatial coordinate system; Finally, pose calculation via weighted fusion: Based on the positioning accuracy performance of each sensor under different working conditions, an accuracy evaluation model is established to dynamically allocate the fusion weights of each sensor, and the accurate pose of the robot is calculated through the weighted fusion algorithm. The data collection process covers multiple typical working conditions such as low-feature scenes, lighting fluctuations, electromagnetic interference and slight occlusion, comprehensively verifying the performance of the fusion algorithm in different scenarios. The main content includes fused robot position coordinates, attitude angles, transformation matrices between each sensor's coordinate system and the robot body coordinate system, confidence evaluation information of each sensor, fusion process logs, comparison data between original sensor data and fusion results, etc. The data format is ROS standard .bag format, with a total data volume of 73.49 MB, containing nearly 50 sets of multi-source fusion positioning data sequences under different working conditions.
