Multi-modal Data-driven Positioning and Navigation Technology for Autonomous Driving

Against the backdrop of the rapid development of artificial intelligence and sensor technology, autonomous driving technology is gradually moving from the laboratory to reality. However, due to the complex and uncertain changes of the real environment (extreme weather, signal blockage, dynamic interference, etc.), it brings serious challenges to autonomous driving location and navigation technology. Single-modal perception and decision-making schemes are not capable of handling these challenges. Consequently, it has become an inevitable choice to make driving more robust and reliable by utilizing multi-modal information. To this end, this report reviews and analyzes the current key technologies for multimodal data-driven unmanned vehicle positioning and navigation. First, by combining the development history of unmanned driving, the report makes clear the background and significance of this research. Second, some fundamental theories, including Kalman filtering, dynamic Bayesian networks, and convolutional neural networks, are introduced. After that, frontier methods such as MixedFusion, tightly coupled SLAM, MDSTF, DeepInteraction++, RoboTron-Drive, and CCTP-Net were analyzed and their innovation in ideas, technical routes, and advantages and disadvantages in performance were compared from the perspective of positioning and navigation. Finally, it summarizes problems and progress of this research topic, to provide theoretical basis and analysis framework for future research.