IRES Track 1: International Research Experiences in Learning based Connected and Autonomous Vehicles (CAVs) with Real-World Implementations

IRES 轨道 1：基于学习的联网自动驾驶汽车 (CAV) 的国际研究经验及其实际实施

• 批准号: 2246347
• 负责人: Seta Bogosyan
• 金额: $ 30万
• 依托单位: CUNY City College
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2023
• 资助国家: 美国
• 起止时间: 2023-04-01 至 2026-03-31
• 项目状态: 未结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2246347&HistoricalAwards=false
• 关键词: IRES; Track; International; Research; Experiences

Future societies will depend more and more on artificial intelligence (AI) and networked systems, and, in turn, on autonomous vehicles (AVs) and connected AVs (CAVs) for many services and operations; i.e. transportation, urban logistics, factory automation, smart farming/agriculture and disaster management, to name a few. AVs and CAVs have strong potential to increase performance, safety, and efficiency, and as a result, to contribute to societal well-being and enhance economic growth. On the other hand, autonomous systems still fail to provide generalizable responses to vehicle, sensor, road, and environment related uncertainties, and need human intervention. Reinforcement learning (RL), an emerging branch of machine learning (ML) and control, has a lot to promise for autonomy with its capacity to address unpredictable changes in the system and environment. However, the field still has many research gaps and also suffers from the lack of practical research evaluation. Most AI based AV research is performed in simulations, on simple platforms and for simplified cases that are far from reflecting real-life uncertainties and convincing responses to changing road conditions, especially at high speeds. This 3-year project will support selected undergraduate and graduate students from US universities to tackle the open challenges of fully autonomous vehicles within a cohort experience at Istanbul Technical University (ITU) under the mentorship of subject-matter experts from ITU, from KTH, and a US based autonomous bus company (ADASTEC Corp). Eight US students (4 graduate, 4 undergraduate) will be funded each year for a 10-week on-site, hands-on research experience at ITU, with each student being in charge of his/her own level-appropriate project using ML/RL for one or more of the vehicle autonomy layers; namely, for perception, localization, motion planning, and trajectory tracking and associated practical tests on actual vehicles (with safety drivers). The algorithm tests will be performed around the ITU Campus within real-world scenarios. Each year, a different student cohort will be selected for this unique research and professional development opportunity, thereby contributing to the US leadership in the future of vehicle technologies with a well-prepared workforce. Special recruitment efforts are planned for broadening participation and recruitment of students from underrepresented communities.The adaptive optimality offered by RL provides increased performance and efficiency when compared with classical control approaches that cannot handle unstructured dynamics, often resulting in safe but highly conservative, low-performance solutions. Similarly, because of its adaptability to changes in the environment and problem dynamics, RL offers increased safety when compared with rule-based, heuristic approaches often practiced in industries to address the needs of autonomous vehicles and platforms. The student research projects for each layer of vehicle autonomy will use RL based designs to address uncertainties and disturbances faced in real-life, which have often been ignored in lab based AV research. In our perception/localization projects, novel algorithms will be developed to estimate sensing uncertainties to improve the performance of RL based motion planning and dynamic object tracking. The trajectory tracking algorithms will be based on Zero-Sum Games (ZSG) and online, model-free RL based control algorithms to address disturbances and slip/slide effects, which is also a significant research contribution. At the end of each research cycle, the modularly developed algorithms will be integrated on an actual vehicle and tested individually and in integration, first on the ITU vehicle-in-the-loop (VIL) system, then on indoor RaceCars and finally, at the ITU Living Lab environment for shuttle service scenarios.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.

未来社会将越来越依赖人工智能 (AI) 和网络系统，进而依赖自动驾驶汽车 (AV) 和联网自动驾驶汽车 (CAV) 来提供许多服务和运营；例如，交通、城市物流、工厂自动化、智能农业和灾害管理等。 AV 和 CAV 具有提高性能、安全性和效率的巨大潜力，从而为社会福祉和促进经济增长做出贡献。另一方面，自主系统仍然无法对车辆、传感器、道路和环境相关的不确定性提供通用响应，需要人工干预。强化学习 (RL) 是机器学习 (ML) 和控制的一个新兴分支，因其能够应对系统和环境中不可预测的变化，在自主性方面大有可为。然而，该领域仍然存在许多研究空白，也缺乏实际的研究评估。大多数基于人工智能的自动驾驶研究都是在简单平台和简化案例的模拟中进行的，这些研究远远不能反映现实生活中的不确定性和对不断变化的路况（尤其是在高速行驶时）的令人信服的响应。这个为期 3 年的项目将支持来自美国大学的精选本科生和研究生，在来自 ITU、KTH 和 KTH 的主题专家的指导下，在伊斯坦布尔技术大学 (ITU) 的队列经验中应对全自动驾驶汽车的开放挑战。一家美国自动驾驶巴士公司 (ADASTEC Corp)。每年将资助 8 名美国学生（4 名研究生、4 名本科生）在国际电联进行为期 10 周的现场实践研究体验，每个学生负责使用 ML/针对一个或多个车辆自主层的强化学习；即，用于感知、定位、运动规划和轨迹跟踪以及对实际车辆（有安全驾驶员）的相关实际测试。算法测试将在国际电联园区的真实场景中进行。每年，都会选择不同的学生群体来获得这一独特的研究和专业发展机会，从而通过准备充分的劳动力为美国在未来汽车技术方面的领导地位做出贡献。计划进行特殊的招募工作，以扩大参与度并招募来自代表性不足的社区的学生。与无法处理非结构化动态的经典控制方法相比，强化学习提供的自适应最优性提供了更高的性能和效率，通常会导致安全但高度保守、低性能解决方案。 同样，由于强化学习能够适应环境和问题动态的变化，与行业中经常采用的基于规则的启发式方法来满足自动驾驶车辆和平台的需求相比，强化学习提供了更高的安全性。车辆自主每一层的学生研究项目将使用基于强化学习的设计来解决现实生活中面临的不确定性和干扰，而这些在基于实验室的自动驾驶研究中经常被忽视。在我们的感知/定位项目中，将开发新的算法来估计传感不确定性，以提高基于强化学习的运动规划和动态对象跟踪的性能。轨迹跟踪算法将基于零和博弈（ZSG）和在线无模型强化学习控制算法，以解决干扰和滑移/滑动效应，这也是一项重大的研究贡献。在每个研究周期结束时，模块化开发的算法将集成在实际车辆上，并单独和集成进行测试，首先在国际电联车辆在环（VIL）系统上，然后在室内赛车上，最后在用于航天飞机服务场景的国际电联生活实验室环境。该奖项反映了 NSF 的法定使命，并通过使用基金会的智力优点和更广泛的影响审查标准进行评估，被认为值得支持。