汽车工程学报2025，Vol.15Issue(5)：621-637,17.DOI:10.3969/j.issn.2095-1469.2025.05.01
基于鸟瞰图融合表示的端到端自动驾驶系统研究综述
A Comprehensive Review of End-to-End Autonomous Driving Based on Bird's-Eye-View Fusion Representations
摘要
Abstract
End-to-end autonomous driving has attracted significant attention from both academia and industry due to its high level of integration and data-driven nature.The Bird's-Eye-View(BEV)feature representation,with its global perspective and geometric consistency,has emerged as the core approach for feature fusion and representation in today's end-to-end autonomous driving systems.Most existing surveys on end-to-end autonomous driving primarily focus on data-driven paradigms such as reinforcement learning and imitation learning.In contrast,this paper explores feature representation,systematically reviewing different BEV feature encoding methods and their impact on overall system performance.Furthermore,we analyze the leading feature-lightweighting strategies to improve BEV perception efficiency and summarize the key performance metrics.Finally,the paper discusses the challenges that BEV-based end-to-end autonomous driving systems still face.关键词
自动驾驶/端到端系统/鸟瞰图/多模态融合/数据驱动Key words
autonomous driving/end-to-end system/bird's-eye-view(BEV)/multi-modal fusion/data-driven分类
信息技术与安全科学引用本文复制引用
汪博文,王亚飞,罗阿彤,王振豪,孙家铭..基于鸟瞰图融合表示的端到端自动驾驶系统研究综述[J].汽车工程学报,2025,15(5):621-637,17.基金项目
国家自然科学基金项目(52372417,52072243) （52372417,52072243）
