CHS: Smal: AI-Human Collaboration in Autonomous Vehicles for Safety and Security

自动驾驶汽车（AVs）有望提高交通安全性，但自动驾驶汽车中使用的最先进的人工智能（AI）技术仍然不够，这在涉及自动驾驶汽车的致命事故中是显而易见的。在可预见的未来，人类的输入和干预仍然是必要的，至少作为自动驾驶汽车的监控器或监督员。要正确监测自动驾驶汽车中罕见但可能致命的事件，需要高度警惕。在自动驾驶汽车中，需要对人类监督员保持警惕。该项目旨在通过人工智能系统和人类驾驶员之间的新型合作来克服这些挑战。该项目将产生有助于减少道路交通事故的算法和设计原则，并广泛适用于网络安全、国防和医疗保健等关键领域的其他相关智能系统。此外，该项目将支持跨学科研究生和本科生的多学科培训，开发课程模块，为学生提供跨学科经验，这对塑造地区和国家劳动力至关重要，并让代表性不足的学生参与STEM领域的研究生、本科生和学前至12年级。该项目通过开发一种可认知的人在环安全人工智能机制来解决安全关键挑战。该项目专注于自动驾驶，包括三个重点：(1)研究如何通过包含人工智能决策不确定性水平的辅助任务分配来保持人类驾驶员的警惕性；(2)开发一个容错的、对手感知的人工智能引擎，该引擎输出其决策中的不确定性水平，作为请求人工输入的基础。(3)基于人-被试数据构建的定量警觉性模型，开发基于警觉性的自适应任务分配方案，在线校准人类警觉性。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。

Autonomous vehicles (AVs) are promising to increase transportation safety and security, but the state-of-the-art artificial intelligence (AI) technologies used in AVs are still not sufficient, as evident in fatal crashes involving AVs. In the foreseeable future, human inputs and interventions will still be necessary, at least as a monitor or supervisor in AVs. Monitoring to correctly detect rare but potentially deadly events in AVs requires high levels of vigilance. The required vigilance taxes human supervisors in AVs. This project aims to overcome these challenges through novel collaboration between the AI system and the human driver. This project will result in algorithms and design principles that help reduce road accidents and are broadly applicable to other related intelligent systems in critical areas such as cybersecurity, national defense, and healthcare. Moreover, this project will support multi-disciplinary training of graduate and undergraduate students across disciplines, the development of course modules that provide students interdisciplinary experience critical to shaping the regional and national workforce, and involvement of underrepresented students in STEM fields at the graduate, undergraduate, and pre-K through 12 levels. This project addresses safety-critical challenges by developing a cognizant human-in-the-loop secure AI mechanism. The project focuses on autonomous driving incorporating three thrusts: (1) Investigate how to maintain human drivers' vigilance through secondary task assignments that incorporate the level of uncertainty in the AI decisions, (2) Develop a fault-tolerant, adversary-aware AI engine that outputs uncertainty levels in its decision as a basis for requesting human inputs. (3) Develop a vigilance-based adaptive task-allocation scheme to calibrate human vigilance online based on a quantitative vigilance model constructed from human-subject data.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.