ERI: Adaptive Intelligence for Active Safety Control and Privacy-Preserving Autonomous Driving

这项工程研究启动（ERI）资助将支持有助于与协同传感、建模和学习相关的新知识的研究，以促进主动安全控制和保护隐私的自动驾驶。驾驶是我们日常生活中必不可少的一部分，很容易受到各种干扰，从而影响道路安全。虽然内部和外部传感数据可以收集并用于提高驾驶安全性，但由于其动态性和复杂性，建模、分析和预测驾驶员的行为和注意力是具有挑战性的。数据的收集和处理也引起了人们对司机隐私的担忧，这方面的调查还不够充分，可能导致各种网络安全漏洞。该项目旨在开发一种新的方法来检查和解释驾驶员的行为和注意力，以挖掘基于深度学习的感知、推理、控制和驱动的潜力，同时保护驾驶员的隐私。这项研究可能会发现新的基于人工智能知识的理论，并将这些新理论与深度学习、预测和控制相结合。该项目的成果将有助于实现主动安全，促进保护隐私的自动驾驶，并为智能教育和国防的发展做出贡献。该项目将为本科生和研究生提供研究机会，并为中学生提供外展活动，以培养具有工程和人工智能技能的未来科学、技术、工程和数学（STEM）劳动力。该项目旨在开发一个数据驱动的、基于深度学习的概念框架，该框架具有知识表示和推理规则，用于捕获、分析和建模真实驾驶场景中的驾驶员行为和认知。首先，在考虑隐私保护的同时，通过探索有限的感官数据，研究驱动行为以及行为、注意和行动之间的内在联系。评估人为因素和道路环境因素对驾驶分心的影响。然后，开发一个可解释的神经网络，对驾驶员的行为、驾驶偏好和特征进行建模，并利用知识表示和推理规则来预测驾驶员的注意力和意图。最后，将创建深度学习技术，根据实际驾驶员模型和对周围环境的感知来预测方向盘转向的角度和方向。随着成功的实施，该项目预计将产生一个具有所提出的理论、模型、网络和技术的传感-智能-控制-驱动闭环系统的原型。此外，本项目的研究成果和知识将被整合到计算机科学本科和研究生课程以及教育和推广项目的学习材料中，以提高公众对科学技术的理解，提高他们的参与度。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。

This Engineering Research Initiation (ERI) grant will support research that will contribute to new knowledge related to collaborative sensing, modeling, and learning to promote active safety control and privacy-preserving autonomous driving. Driving is an essential part of our daily lives and is vulnerable to a range of distractions that can compromise safety on the road. While internal and external sensing data can be collected and used to improve driving safety, modeling, analyzing, and predicting driver behaviors and attention are challenging due to their dynamic nature and complexity. The data collection and processing also raise concerns about drivers’ privacy, which is under-investigated in the field and can lead to various cybersecurity vulnerabilities. This project aims to develop a new way to examine and interpret driver behavior and attention to exploit the potential of deep learning-based sensing, reasoning, control, and actuation in a dynamic traffic and driving environment while preserving drivers' privacy. The research can potentially uncover new AI knowledge-based theories and result in a synergistic integration of these new theories with deep learning, prediction, and control. The results from this project will help enable active safety, promote privacy-preserving autonomous driving, and contribute to the development of smart education and national defense. This project will offer research opportunities to undergraduate and graduate students and provide outreach activities to middle school students to foster the growth of the future science, technology, engineering, and mathematics (STEM) workforce with engineering and AI skills. This project seeks to develop a data-driven, deep learning-based conceptual framework with knowledge representation and inference rules for capturing, analyzing, and modeling driver behaviors and cognition in real-world driving scenarios. First, driving behaviors and inner connections among behavior, attention, and action will be investigated by exploring limited sensory data while considering privacy protection. The impacts of human and road environment factors on driving distraction will be assessed. Then, an explainable neural network will be developed to model the driver’s behavior, driving preferences, and distinctive features with knowledge representation and inference rules to predict the driver’s attention and intention. Finally, deep learning techniques will be created to predict the angle and direction of the wheel steering based on actual driver models and perceptions of the surrounding environment. With successful execution, this project is expected to result in a prototype sensing-intelligence-control-actuation closed-loop system with the proposed theories, models, networks, and techniques. In addition, research findings and knowledge made by this project will be integrated into the Computer Science undergraduate and graduate courses and learning materials of educational and outreach programs to improve public understanding of science and technology and increase their engagement.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.