Abstract: Surveillance robot equipped with multiple sensors can autonomously navigate in industrial environments, performing tasks such as personnel standard wear detection, monitoring hazardous areas, and fire and smoke detection, thereby ensuring safety production. However, its application in complex dynamic industrial environments remains challenging, such as in locomotive maintenance workshops, where the temporary placement of parts alters environmental features and passability, affecting the robot’s autonomous localization and navigation capabilities. Therefore, this paper proposes an enhanced autonomous localization and navigation system for surveillance robot. By stratifying the 3D LiDAR scan point cloud according to the height of semi-dynamic objects, the system captures static environmental features for laser SLAM, and perceives semi-dynamic obstacles for obstacle avoidance planning. Furthermore, in robot motion planning, real-time perception of surrounding obstacle density and proximity scales the velocity constraints of the robot, enhancing its safety navigation capabilities. In workshop environments with semi-dynamic objects, experiments show that the quadruped robot with the proposed system can recover localization within 6.4s when it is not initialized at the origin. Furthermore, in multiple navigations to the same target pose, the maximum repeat precision error was ±0.011m and maintains an average minimum distance of 0.56m to obstacles. Compared to baseline methods, this approach exhibits advantages in robot autonomous localization and navigation.

Key words: surveillance robot, autonomous localization and navigation, path planning, motion planning, obstacle perception, complex industrial environment

摘要： 安全巡视机器人配备多种传感器，可在工业现场进行自主运动，执行人员穿戴标准检测、危险区域盯控、烟雾火灾识别等任务，保障安全生产。然而，其在复杂动态环境的运用仍然具有挑战性，如机车检修车间中，临时堆放的物品改变了环境特征与通过性，影响了机器人自主定位导航能力，因此，本文提出了针对性的改进方案。首先，将三维激光雷达扫描点云以半动态物体高度为界限分层，上半层包含环境静态特征，用于激光同步定位与建图，下半层包含半动态障碍物信息，用于实时规划避障。接着，在机器人运动规划算法中，实时感知障碍物密度与接近程度，缩放机器人速度约束，提升其安全通行能力。最后，使用四足机器人在受到半动态物体干扰环境中进行了实验，结果表明：当其在非原点初始化时，能在内6.4s恢复定位；当多次执行到达地图中某一点的导航任务时，其重复精度误差最大为±0.011m，且过程中与障碍物最短距离平均为0.56m ，本文方法在复杂动态场景定位导航性能指标上均优于现有方法。

关键词: 巡视机器人, 自主定位导航, 路径规划, 运动规划, 障碍感知, 复杂工业现场