This work develops a real-time localization framework based on multi-sensor fusion for the autonomous navigation of mobile robots in challenging off-road environments. The proposed framework employs a hybrid coupling strategy at its front-end, which integrates Inertial Measurement Unit (IMU) data with Global Navigation Satellite System (GNSS) and LiDAR. As a consequence, GNSS Inertial Odometry (GIO) and Lidar Inertial Odometry (LIO) are created and these are crucial for achieving precise localization in real-time. The RGB-Height-Intensity descriptor is used to improve the precision of loop closure detection while maintaining search efficiency. This is achieved by integrating surface reflectance data from the LiDAR intensity channel and RGB data from camera images. Finally, in the back-end procedure of this framework, the accuracy of posture estimation is enhanced by the optimization of odometry and loop closure factors. Additionally, a framework of real-time localization and autonomous navigation is developed and implemented on a wheeled robot to verify the proposed real-time localization framework. The experimental results indicate the framework exhibited higher performance in off-road environments, where conventional localization techniques generally degrade.
