煤矿巷道、采掘工作面等作业区域具有典型的半结构化或非结构化环境特征，且 GPS 无法 在煤矿井下直接应用，亟需构建适用于煤矿井下移动机器人的自主定位系统方案，解决机器人精确 定位、状态估计等问题。 针对上述问题，提出了一种基于激光惯性的融合 SLAM 算法，实现了移动 机器人在煤矿井下实时输出稳健的 6DOF 状态估计和全局一致的同步定位与地图构建。 该算法由 前端迭代卡尔曼滤波和后端位姿图优化 2 部分组成。 该方法首先在前端，将传感器数据经过预处 理，构建了观测模型和预测模型，建立了迭代卡尔曼滤波器，结合机器人先验位姿经过预测和观测 的状态传播过程，使其状态更新后的后验位姿更加准确，如此循环迭代得到了基于紧耦合的激光惯 性里程计，增强了机器人在这种非结构环境下的鲁棒性。 其次在后端，部署了关键帧的选取策略， 以限制状态估计的数量，满足其在大尺度场景下实时性的要求。 同时，在优化框架中添加了地面约 束和回环检测，优化了相邻关键帧之间的相对位姿，以确保全局地图的一致性，从而进一步提高了 机器人 6DOF 状态估计的整体精度。 最后，分别在公开数据集和自采数据集上验证了该算法的性 能。 实验结果表明:针对煤矿井下这种特殊的非结构环境，与现有的激光 SLAM 算法相比，提出的 算法使机器人具有更高的精度、实时性和鲁棒性，有效降低了系统的累积误差，保证了所构建地图 的全局一致性。

The operation areas such as coal mine roadway and mining working face have some typical semi⁃structured or unstructured environment characteristics， and GPS cannot be directly applied in underground coal mines. Therefore， there is an urgent need to build an autonomous positioning system for coal mine mobile robot to solve the problems of its precise positioning and state estimation in underground coal mine. To solve these problems， a LiDAR⁃ Inertial SLAM algorithm is proposed to achieve a real⁃time output of robust six degrees of freedom (6DOF) state esti⁃ mation and globally consistent simultaneous localization and mapping (SLAM) for robot in underground coal mines. It consists of two parts: front end iterative Kalman filtering and back end pose graph optimization. Firstly， on the frontend， an iterative Kalman filter is established to construct a tightly coupled based LiDAR⁃Inertial Odometry (LIO). The state propagation process for the a priori position and attitude of a robot， which uses predictions and observations， increases the accuracy of the attitude and enhances the system robustness. Secondly， on the back end， the key frame selection strategy is deployed to meet the real⁃time requirements for large⁃scale scenes. Moreover， loop detection and ground constraints are added to the optimization framework， thereby further improving the overall accuracy of the 6DOF state estimation. Finally， the performance of the algorithm is verified using a public dataset and the dataset col⁃ lected. The experimental results show that for the special environment of underground coal mine， compared with the existing LiDAR⁃SLAM algorithm， the proposed algorithm makes the robot have higher accuracy， real⁃time performance and robustness， effectively reducing the cumulative error of the system and ensuring the global consistency of the con⁃ structed maps.