de Miguel Miguel Ángel, García Fernando, Armingol José María
Intelligent Systems Laboratory, Universidad Carlos III de Madrid, Av. de la Universidad, 30, 28911 Leganés, Madrid, Spain.
Sensors (Basel). 2020 Jun 2;20(11):3145. doi: 10.3390/s20113145.
This paper proposes a method that improves autonomous vehicles localization using a modification of probabilistic laser localization like Monte Carlo Localization (MCL) algorithm, enhancing the weights of the particles by adding Kalman filtered Global Navigation Satellite System (GNSS) information. GNSS data are used to improve localization accuracy in places with fewer map features and to prevent the kidnapped robot problems. Besides, laser information improves accuracy in places where the map has more features and GNSS higher covariance, allowing the approach to be used in specifically difficult scenarios for GNSS such as urban canyons. The algorithm is tested using KITTI odometry dataset proving that it improves localization compared with classic GNSS + Inertial Navigation System (INS) fusion and Adaptive Monte Carlo Localization (AMCL), it is also tested in the autonomous vehicle platform of the Intelligent Systems Lab (LSI), of the University Carlos III de of Madrid, providing qualitative results.
本文提出了一种方法，该方法通过对概率激光定位（如蒙特卡洛定位（MCL）算法）进行改进来提高自动驾驶车辆的定位精度，即通过添加卡尔曼滤波后的全球导航卫星系统（GNSS）信息来增强粒子的权重。GNSS数据用于提高地图特征较少区域的定位精度，并防止出现机器人被劫持问题。此外，激光信息可提高地图特征较多且GNSS协方差较高区域的定位精度，使得该方法可用于GNSS特别困难的场景，如城市峡谷。该算法使用KITTI里程数据集进行了测试，结果证明与经典的GNSS + 惯性导航系统（INS）融合以及自适应蒙特卡洛定位（AMCL）相比，它提高了定位精度。该算法还在马德里卡洛斯三世大学智能系统实验室（LSI）的自动驾驶车辆平台上进行了测试，给出了定性结果。
