Implementing Real-Time DOV Compensation: A Practical Approach Using a MLP on an NPU Embedded in an INS

This paper explores the impact of gravity disturbances on INS accuracy and presents a method for real-time compensation during the navigation process. By utilizing data from the precise gravity model, EGM2008, a novel approach to compensate for the DOV in real time on the INS’s built-in NPU was introduced. This method predicts gravity disturbances while traveling for platforms on both land and water, utilizing the MLP technique. To predict these gravity disturbances, four distinct MLP models, MLP1~MLP4, were designed and their supervised learning results were compared using HMSE and RMSE. This comparative analysis allowed us to identify that the MLP4 model exhibited the best performance. In order to validate the proposed method, MLP4 was implemented inside the NPU and the measured execution time was 1.041 ms. The field test was conducted with real-time execution of the MLP4 model on the NPU of the INS. The results of this field test clearly demonstrated the effectiveness of the proposed approach in enhancing position accuracy. Over the course of a 2 h field test, it was evident that employing the proposed method improved position accuracy by a notable 27%.