Robust Multi-Sensor Fusion for Localization in Hazardous Environments Using Thermal, LiDAR, and GNSS Data

Abstract

Navigation for autonomous robots in hazardous environments demands robust localization solutions. In challenging environments such as tunnels and urban disaster areas, autonomous robots and vehicles are particularly important for search and rescue operations. However, especially in these environments, sensor failures and errors make the localization task particularly difficult. We propose a robust sensor fusion algorithm that integrates data from a thermal camera, a LiDAR sensor, and a GNSS to provide reliable localization, even in environments where individual sensor data may be compromised. The thermal camera and LiDAR sensor employ distinct SLAM and odometry techniques to estimate movement and positioning, while an extended Kalman filter (EKF) fuses all three sensor inputs, accommodating varying sampling rates and potential sensor outages. To evaluate the algorithm, we conduct a field test in an urban environment using a vehicle equipped with the appropriate sensor suite while simulating an outage one at a time, to demonstrate the approach's effectiveness under real-world conditions.