Traditional Modular and End-to-End Methods in Vehicle Trajectory Planning and Decision Control in Complex Scenario: Review

Modular traditional autonomous driving and end-to-end autonomous driving have their own characteristics and play an important role in different scenarios in autonomous driving. The comprehensive performance of these two methods is compared and evaluated systematically in this paper. Modularity Traditional autonomous driving achieves high controllability and interpretability by breaking the system into multiple independent functional modules. However, the efficiency of information transfer between modules is low, and local optimal problems may occur when dealing with complex scenes. The end-to-end autonomous driving system realizes direct mapping from perception to control through deep learning, showing strong global optimization capabilities and the potential to deal with complex scenarios, but also faces black box problems and dependence on a large number of labeled data. This paper discusses the advantages and disadvantages of these two methods in practical applications, and suggests possible future research directions, including the integration of modular and end-to-end methods, improving the interpretability and security of the system, and improving data efficiency and system generalization. Taken together, modular traditional autonomous driving and end-to-end autonomous driving can achieve a safer and more efficient autonomous driving system by combining their respective advantages.