Deep Learning-Based Strategies for Integrated Autonomous Navigation: A Review

Abstract

Multi-sensor integrated navigation/positioning systems (MSINPS) play a crucial role in achieving precise positioning and navigation in challenging environments. By combining data from various sensors such as Global Navigation Satellite System (GNSS), Inertial Measurement Units (IMUs), cameras, and Light Detection and Ranging (LiDar), MSINPS enhance accuracy and robustness, ensuring uninterrupted operation even in the event of sensor malfunction. This paper explores autonomous vehicle guidance for safe and efficient navigation, focusing on two paths of data fusion: sensor-based fusion and learning-based fusion. The effectiveness of MSINPS hinges on meticulous sensor selection and optimization to meet the requirements of autonomous navigation applications. Addressing noise, biases, and inconsistencies through data pre-processing is essential for robust data fusion. Advanced fusion algorithms leveraging machine learning and deep learning techniques exhibit superior accuracy and adaptability. Furthermore, integrating MSINPS with artificial intelligence (AI) opens new avenues for navigation and positioning enhancement. This integration can lead to improvements in sensor technology and solution outcomes. Striking a balance between computational complexity and available resources promotes efficient operation while implementing fault tolerance strategies ensures system robustness across diverse environments.