Measuring and Modeling the Impact of Dynamic Trust in Automated Vehicles on Driver Behavior

机动车碰撞每年造成35,000多人死亡，近300万人受伤。自动驾驶汽车技术已经成为一种很有前途的机制，可以防止这些碰撞，增加个人机动性，降低排放。由于公众对自动驾驶汽车安全性的担忧日益增加，特别是在自动系统与人类驾驶员之间的控制转移过程中，这些承诺的兑现受到了限制。在这些互动过程中，人类驾驶员和自动系统之间的信任是一个核心问题。先前对人机信任的研究已经确定，人机系统的安全和性能需要经过校准的信任，即人类驾驶员对自动化系统的信任与系统的能力相匹配的状态。由于信任测量和方法的局限性，自动驾驶汽车的信任校准是一个难以捉摸的挑战，无法阐明技术设计决策对信任和驾驶员行为的影响。该项目将通过增进对人-自动化信任的理解，促进科学进步，促进国民健康。具体来说，该项目将解决现有信任措施、模型信任和驾驶员行为的局限性，并确定纳入信任校准模型的自动驾驶汽车如何影响动态信任和驾驶行为。该方法将提供指导方针和技术设计建议，可以显著减少与车辆碰撞相关的人员死亡和伤害。这项工作的广泛影响包括本科生和研究生课程的开发，为德州农工大学（Texas a&m University）代表性不足的本科生提供重点研究机会，以及为当地高中生提供学生拓展活动。该项目将分为三个阶段，旨在：(1)利用自动驾驶汽车交互过程中神经激活的实时测量，开发一种新的客观动态信任测量方法；(2)沿着动态信任谱建立驾驶员与自动驾驶汽车交互的模型；(3)通过基于信任校准干预的驾驶仿真实验验证信任测量和驾驶员行为模型。该措施的开发将得到人类受试者研究数据的支持，在该研究中，模拟自动驾驶车辆中的驾驶员将遇到一系列现实驾驶场景，这些场景旨在调节对系统的信任，同时收集生理、神经、行为和主观测量。神经学数据将用回归和连通性分析方法进行分析，以确定与生理和主观测量验证的信任状态的相关性。相关物将用于训练和测试驾驶员信任的新过程模型和对扩展现有驾驶员决策和控制框架的自动化事件的响应。模型和措施将在第二次驾驶模拟研究中进行验证，包括信任校准干预。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Motor vehicle crashes cause over 35,000 deaths and almost 3 million injuries per year. Automated vehicle technologies have emerged as a promising mechanism to prevent these crashes, to increase personal mobility, and to lower emissions. Delivery on these promises has been limited by a growing public concern over the safety of automated vehicles, particularly during transfers of control between automated systems and human drivers. Trust between human drivers and automated systems is a central concern during these interactions. Prior research in human-automation trust has established that the safety and performance of human-machine systems requires calibrated trust—a state where a human driver’s trust in an automated system matches the system’s capabilities. Trust calibration in automated vehicles is an elusive challenge because of limitations in trust measurement and methods that illuminate the impact of technology design decisions on trust and driver behavior. This project will promote the progress of science and advance the national health by advancing an understanding of human-automation trust. Specifically, the project will address the limitations of existing trust measures, model trust and driver behavior, and determine how autonomous vehicles that incorporate trust calibration models can influence dynamic trust and driving behavior. The approach will provide guidelines and technology design recommendations that could significantly reduce the human lives lost and injuries associated with vehicle crashes. Broader impacts of the work include undergraduate and graduate course development, focused research opportunities for underrepresented undergraduates at Texas A&M University, as well as student-leg outreach activities to local high school students.The project will consist of three phases designed to (1) develop a novel and objective measure of dynamic trust using real-time measures of neural activation during AV interactions, (2) model driver interactions with automated vehicles along the spectrum of dynamic trust, and (3) validate the trust measure and driver behavior model with a trust calibration intervention-based driving simulation experiment. The measure development will be supported by data from human subjects studies in which drivers in a simulated automated vehicle will encounter a series of realistic driving scenarios designed to modulate trust in the system while physiological, neurological, behavioral, and subjective measures are collected. Neurological data will be analyzed with regression and connectivity analysis methods to identify correlates with trust states validated by the physiological and subjective measures. The correlates will be used to train and test novel process models of driver trust and responses to automation events that extend existing driver decision-making and control frameworks. The models and measure will be validated in a second driving simulation study including a trust calibration intervention.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.