A Study on Integrated Navigation Algorithm using Deep learning based Lidar Odometry and Inertial Measurement

Abstract

As lidar has become one of the primary sensors for autonomous robots, interest in lidar-based sensor fusion and navigation has increased. In this paper, we propose an integrated navigation algorithm that combines an inertial navigation system and deep learning-based lidar odometry. We first developed a deep learning neural network for lidar odometry estimation. The network can estimate the relative pose by using consecutive lidar scans as an input. We then designed an extended Kalman filter that uses inertial sensors for the prediction step and lidar odometry for the correction step. To demonstrate the estimation accuracy of the algorithm, we used both the KITTI dataset and a drone flight simulator.