XDLL: Explained Deep Learning LiDAR-Based Localization and Mapping Method for Self-Driving Vehicles

Self-driving vehicles need a robust positioning system to continue the revolution in intelligent transportation. Global navigation satellite systems (GNSS) are most commonly used accomplish this task because of their ability accurately locate vehicle environment. However, recent publications have revealed serious cases where GNSS fails miserably determine position vehicle, for example, under bridge, tunnel, or dense forests. In work, we propose framework architecture explaining deep learning LiDAR-based (XDLL) models that predicts by using only few LiDAR points environment, which ensures required fastness and accuracy interactions between components. The proposed extracts non-semantic features from scans clustering algorithm. identified clusters serve as input our model, relies on LSTM GRU layers store trajectory convolutional smooth data. model has been extensively tested with short- long-term trajectories two benchmark datasets, Kitti NCLT, containing different environmental scenarios. Moreover, investigated obtained results contribution each cluster feature several explainable methods, including Saliency, SmoothGrad, VarGrad. analysis showed taking mean all an is enough obtain better compared first it reduces time consumption well. improved able absolute error below one meter sequences trajectories.