为梳理复杂交通场景下智能驾驶决策的研究脉络,本文以现有研究成果为基础,从场景分类、决策算法及决策指标三个核心维度展开系统性归纳与分析。依据影响因素的差异,将复杂交通场景划分为复杂车辆行为场景、高密度交通流场景及复杂结构道路场景三类,并剖析各类场景的风险特征;结合场景特性,将决策算法分为学习型、博弈型与端到端型,对比分析不同算法在实际应用中的优势与局限;从安全性、舒适性、高效性三个层面整理决策指标,明确不同场景下的核心关注指标。通过上述分析,本文总结当前研究的重点方向,同时指出三大研究难点,即复合复杂场景决策研究不足、自进化决策机制缺失及决策评估指标不统一。据此,本文提出未来三大研究方向,包括多目标决策模型构建、自学习型决策算法开发及基于认知的决策研究,旨在为复杂交通场景及未来困境场景下的智能驾驶决策研究提供理论支撑与实践参考。

To sort out the research context of intelligent driving decision-making in complex traffic scenarios, this paper conducts a systematic induction and analysis from three core dimensions: scenario classification, decision-making algorithms, and decision-making indicators based on existing research results. According to the differences in influencing factors, complex traffic scenarios are divided into three categories: complex vehicle behavior scenarios, high-density traffic flow scenarios, and complex structural road scenarios, and the risk characteristics of each type of scenario are analyzed. Combined with scenario characteristics, decision-making algorithms are classified into learning-based, interactive, and end-to-end types, and the advantages and limitations of different algorithms in practical applications are compared and analyzed. Decision-making indicators are sorted out from three aspects: safety, comfort, and efficiency, and the core focus indicators under different scenarios are clarified. Through the above analysis, this paper summarizes the key directions of current research, and at the same time points out three major research difficulties, namely insufficient research on decision-making in compound complex scenarios, lack of self-evolutionary decision-making mechanisms, and inconsistent decision-making evaluation indicators. Based on this, the paper proposes three future research directions, including the construction of multi-objective decision-making models, the development of self-learning decision-making algorithms, and research on cognition-based decision-making, aiming to provide theoretical support and practical references for intelligent driving decision-making research in complex traffic scenarios and future dilemma scenarios.