一种基于激光雷达的同步定位建图与自主导航算法研究

SLAM and Autonomous Navigation Algorithm Based on 2D-Lidar

摘要: 同步定位与建图(Simultaneous Localization and Mapping, SLAM)是自动驾驶领域的关键技术之一。基于差异化的应用场景、传感器和算法，主流SLAM实现方法分为视觉SLAM、激光SLAM和多传感器SLAM等，其性能的优化方向主要基于传感器性能，算法策略以及估计方法。室内场景下，视觉SLAM方法因视觉传感器的视野范围有限且对光线明暗，性能受限于环境变化。多线激光雷达因价格昂贵，不适宜广泛应用。多传感器融合的方案，因部分视觉传感器采集环境特征信息丰富而存在系统运行的时延问题且融合设计复杂度高。针对以上问题，本文提出了一种基于单线激光雷达，感知算法与全局路径规划算法融合的SLAM与自主导航方案，设计软件与硬件结合的系统框架，所实现的运动系统在室内场景下高速获取环境信息，能够获取良好的建图性能并完成低时延自主导航。首先，本文系统利用IMU对单线激光雷达采集的回波数据做畸变矫正，对激光雷达数据采用Gmapping算法处理，并对各传感器的坐标系关系做精确计算和转换，而后采用Djikstra算法作为全局路径规划算法来实现运动系统的自主导航。最终通过实验，利用ROS系统，NVIDIA-Jetson-Nano，IMU、2D-Lidar等，在实际环境中实现了算法模型与系统构建，室内场景中建图效果良好，机器车在同步定位与建图过程中自主导航，验证了本文方案及系统的可靠性与有效性。

Abstract: Simultaneous Localization and Mapping (SLAM) is a key technology in autonomous driving. Based on different applications, hardware sensors and algorithms, the SLAM implementation methods are mostly divided into Visual SLAM, Laser SLAM and multi-sensor SLAM. The optimization direction of SLAM performance is mainly based on sensor performance, algorithm and estimation method. In the indoor scene, the performance of the Visual SLAM method is limited by the environment due to the limited field of vision sensor and its sensitivity to light. 3D-Lidar is not suitable for wide application for the high price. The multi-sensor fusion scheme has the problem of the complexity of fusion design and has operation delay due to the rich environmental characteristic information collected by some vision sensors. To solve the problems above, this paper proposes a SLAM scheme based on 2D-Lidar, the fusion of sensing algorithm and global path algorithm, and designs a system framework combining software and hardware. The realized system can obtain environmental information at high speed in indoor scenes and complete low-delay autonomous navigation with good mapping performance. First, the system uses IMU to correct the distortion of the Lidar data which is processed by the Gmapping algorithm. The system realized in this paper performs accurate calculation and conversion of the coordinate system relations of each sensor. After that, the Djikstra algorithm is adopted as the global path planning algorithm. Finally, ROS systems, NVIDIA-Jetson-Nano, IMU, 2D-Lidar, etc. are used to realize the construction of the algorithm model and system in the actual environment. The diagram construction effect in the indoor scene is good, the complete map of the experimental environment is obtained and the motion system positioning is accurate, which verifies the reliability and effectiveness of the scheme and system in this paper.