随着自动驾驶技术趋于成熟和共享出行理念深入人心，自动驾驶出租车有望成为未来居民重要出行方式之一。与此同时，全球掀起研究和发展自动驾驶出租车的浪潮。但是，自动驾驶出租车相关的用户需求与交通管理问题尚属研究空白，亟需体系化的解决方案。本论文基于问卷调研、机器学习、双层规划等研究方法，探究自动驾驶出租车大范围应用的关键影响因素，为其科学管理提供理论支撑。首先，提出扩展的技术接受模型（TAM），探究自动驾驶出租车接受度的影响因素及作用机理。将经典TAM模型中的外部变量扩展为显式表达的感知信任、政府支持、社会影响和感知娱乐因素，利用结构方程模型（SEM）验证模型假设，并建立机器学习回归模型预测用户的自动驾驶出租车接受度。结果表明，与SEM相比，支持向量机（SVM）的预测精度最高提升7.9%，并通过累积局部效应（ALE）图刻画各因素对接受度的一阶效应和二阶效应，可服务完善自动驾驶出租车功能。其次，构建与现有交通出行方式耦合的自动驾驶出租车出行场景，研究用户对自动驾驶出租车的出行选择行为。耦合建模考虑出行目的、时段、费用、时间和接驳距离等多场景因素，以及自动驾驶出租车使用经历、共享出行经历和心理因素。建立机器学习多分类模型预测用户出行选择行为，准确率比混合Logit模型提升13.0%，并通过ALE图刻画各因素的非线性影响机理，可促进提升自动驾驶出租车较其它出行方式的竞争力。最后，建立多模式交通系统下自动驾驶出租车费率定价的双层规划模型，阐述自动驾驶出租车分担率的影响机制。基于经济、时间、安全性和舒适性成本构造广义出行函数，并利用超级交通网络描述多模式交通系统。以自动驾驶出租车平台利润最大化和路网随机用户均衡为上下层目标的双层规划模型，利用遗传-迭代加权法（GA-MSA）求解，分析安全系数、时间价值系数和运维成本与分担率间的关联关系，可协助扩大自动驾驶出租车长途出行和短途接驳的应用范围。本研究从自动驾驶出租车与现有交通模式的融合发展的新视角出发，构建多方式、多因素耦合场景下的理论模型，并提出一种基于机器学习模型的用户接受度和出行选择影响因素分析方法，克服传统机器学习模型可解释性差的缺点，研究影响自动驾驶出租车的用户决策行为和交通组织管理的关键因素。研究结果为自动驾驶出租车的广义接受度、出行选择行为、分担率的关键影响因素提供理论模型的支撑，在推动自动驾驶技术的应用方面具有较大的实际意义。

As the autonomous driving technology and shared mobility tends to be mature, the robo taxi is expected to become one of the future important travel modes for residents. Meantime, researchers and companies have paid more and more attention to robo taxis. However, there is still a research gap in user demand and traffic management related to robo taxis, and systematic solutions are urgently needed. This thesis explores the key influencing factors of the large-scale application of robo taxis by applying questionnaire survey, machine learning algorithms and bi-level programming. The research could provide theoretical support for scientific management of robo taxis.Firstly, an extended technology acceptance model (TAM) is proposed to explore the psychological factors and mechanism that affect the acceptance of robo taxis. External variables in traditional TAM model are expanded to an explicit expression of perceived trust, policy support, social factor and perceived entertainment. Structural equation modeling (SEM) analysis is used to validate the hypothesis of extended TAM and machine learning regression models are applied to predict users’ acceptance of robo taxis. The results show that the accuracy of support vector machine (SVM) is up to 7.9% higher than SEM. Accumulated local effect (ALE) plots are applied to describe the first-order effect and second-order effect of factors on users’ acceptance, which could be applied for the enhancement of improving the robo taxis’ functionality. Secondly, a scenario that coupled the robo taxis with the existing travel model is constructed to study the user‘s travel choice behavior of robo taxis. Multiple factors, including travel purpose, time, cost, time and connection distance, are considered in the coupling modeling, as well as robo taxi use experience, shared travel experience and psychological factors. Machine learning multi-classification models are applied to predict users’ travel choice behavior and the accuracy is 13.0% higher than mixed Logit models. ALE plots are applied to describe the nonlinear effects of factors, which could promote the competitiveness of robo taxis compared with other travel modes. Finally, a bi-level programming model of robo taxi pricing under a multi-mode transportation system is established to illustrate the influence mechanism of the robo taxis sharing rate. The upper objective of bi-level programming model is to maximize the profit of the robo taxis platform. The lower objective is to realize stochastic user equilibrium condition. The genetic algorithm - method of successive average(GA-MSA) is used to solve the proposed model. The relationship between safety coefficient, time value coefficient, operation and maintenance cost and sharing rate are analyzed, which could help expand the robo taxis application scope of long-distance travel and short switching.From the new perspective of the integrated development of robo taxis and existing traffic modes, a theoretical model of the multi-mode and multi-factor coupling scenario is constructed in this thesis. A machine learning-based methodology for explaining users’ acceptance and travel choice is also put forward, which can overcome their difficulty in explaining prediction results. The key factors affecting the decision-making behavior of the robo taxis user and the traffic organization management are studied. The results could provide theoretical model support for key factors affecting generalized acceptance, travel choice behavior, and demand and management, which has practical significance in promoting the applications of the autonomous driving technology.