Data Fusion of RADAR and LIDAR for Robot Localization Under Low-Visibility Conditions in Structured Environments

Optical range sensors such as LiDAR and range cameras have become the most common devices for robot localization and navigation tasks. However, their performance can be degraded by meteorological hazards, such as fog, smoke, or rain. This paper proposes a new method to combine information from LiDAR sensors and low-cost RADAR sensors in structured 2D environments, in order to ensure the availability of useful information in low-visibility conditions due to smoke. Our method makes use of a novel DBScan-Line segmentation for clustering the measurements from the LiDAR sensor, and then it establishes correspondences between these clusters and the measurements from the RADAR sensors. The method has been extensively tested in field experiments with artificial smoke, and the results benchmarked against raw sensors and a state-of-the-art fusion method. Moreover, the fused measurements have been integrated into a localization method, which was able to robustly localize a ground platform in dense fog.