融合码盘和激光雷达的里程计与建图

Wheel-LiDAR Odometry and Mapping for Autonomous Vehicles

摘要:

提出了一种用于自动驾驶汽车的低漂移、低延迟的里程计与高精度建图的算法。该方法融合了多种传感器的测量结果，包括车轮编码器、转向盘转角编码器、激光雷达及可选GPS等的测量结果。里程计算法由车轮里程计和激光里程计组成：前者基于车辆运动学模型，高频、实时估计位姿增量，用于点云去畸变和为后者优化位姿提供可用的初值；后者以较低的频率估计车辆的精确位姿变化，以补偿前者累计的误差，其核心是一种基于角度度量的两阶段特征提取方法。建图算法基于因子图，包含激光里程计因子、回环因子和可选GPS因子，通过增量平滑和建图算法优化全局轨迹，在线生成全局地图，其中GPS因子能够自动对齐GPS坐标系和里程计坐标系，逐步融合GPS测量值，解除了算法初始化过程对于GPS的依赖。所提出的方法在自动驾驶汽车平台数据集上进行了评估，并和已开源的部分相关工作进行对比，结果表明它具有更低的漂移率，在本文进行的最大规模的测试中达到了0.53%。相关代码以开源形式供交流参考（https：//github.com/Saki-Chen/W-LOAM）。

Abstract:

This paper， by proposing a wheel-LiDAR method of odometry and mapping（WLOAM）， using wheel encoder， steering encoder，LiDAR， and optional GPS for autonomous vehicles， estimates the low-drift pose at real-time and builds a high-accurate map. The odometry consists of the wheel odometry algorithm and the LiDAR odometry algorithm. The former estimates the 3-DOF ego-motion of LiDAR at a high frequency based on Ackermann steering geometry， whose resulting pose increment is applied in point clouds de-skewing and works as a fine initial guess for LiDAR odometry while the latter performs the 6-DOF scan-to-map LiDAR pose optimization at a relatively low frequency to compensate the pose error accumulated by the wheel odometry， whose core is a two-stage method with an angle-based metric for extracting features. The mapping method is based on the factor graph consisting of the LiDAR odometry factor， the loop closure factor， and the optional GPS factor， which is solved via incremental smoothing and mapping （iSAM） to produce a global map online. An auto-aligned-GPS-factor is proposed for fusing GPS measurement incrementally without explicit initialization. The proposed method was extensively evaluated on the datasets gathered from the autonomous vehicle platform and compared with related open-sourced works. The results show a lower drift rate， which reaches 0.53% in the largest test described in this paper. The implementation of the proposed method is open-sourced for communication （https：//github.com/Saki-Chen/W-LOAM）.