Time Delay Error Online Correction of LiDAR-IMU System Through MSCKF Integrated DLRNN Method

When fusing the measurement data with different sampling frequencies from the light detection and ranging (LiDAR) and inertial measurement unit (IMU), their timestamps should be exactly aligned. However, in reality the timestamps of LiDAR and IMU are typically subject to different influences, which will inevitably generate the time delay error to reduce the accuracy and robustness of the LiDAR-IMU system. To this avail, this article proposes a new method that integrates the double layer recurrent neural network (DLRNN) and multistate constrained Kalman filter (MSCKF) to online correct the LiDAR-IMU time delay errors. In this new method, the MSCKF can improve the DLRNN training accuracy while in return the DLRNN can enhance the error estimating performance of the MSCKF. With this mutual improvement strategy, the time delay error can be precisely corrected in both the static and dynamic operation modes of the LiDAR-IMU system. The main contributions include: 1) Dual-information fusion is achieved between the DLRNN and MSCKF for accurate correction of the LiDAR-IMU time delay error; and 2) the proposed approach significantly improves the efficiency and accuracy of the time delay error correction in a real-time manner. Several experiments were carried out to evaluate the online correction performance of the proposed method. The experimental results demonstrate that the LiDAR-IMU time delay error can be accurately and quickly corrected, regardless the time-varying and unknown time delay. As a result, the positioning and navigation performance of the LiDAR-IMU system can be improved more appropriately for practical applications.