Abstract

Over the past two decades, autonomous vehicles (AVs) have evolved from rule-based prototypes to AI-driven systems with commercial deployments. Yet despite tremendous technical progress, the goals of scalable deployment and guaranteed safety remain elusive. This talk reflects on the historical trajectory of AV development—framed across three paradigms: AV 1.0 (rule-based systems), AV 2.0 (discriminative learning), and the emerging AV 3.0 (generative learning). By examining key technological shifts and persistent challenges, including the curse of dimensionality and the curse of rarity, we will discuss the opportunities with AV 3.0 as a forward-looking paradigm grounded in generative modeling and vision-language systems—offering new pathways for data efficiency, scene understanding, and safety assurance. The presentation will also discuss Mcity’s behavioral safety assessment framework towards large scale AV deployments.

Bio

Henry Liu is the Bruce D. Greenshields Collegiate Professor of Engineering at the University of Michigan, where he serves as Director of the University of Michigan Transportation Research Institute (UMTRI), Director of Mcity, and Director of the USDOT Center for Connected and Automated Transportation. His research bridges traffic systems, vehicle automation, and artificial intelligence, with a focus on safety, mobility, and real-world deployment. Dr. Liu is widely recognized for pioneering work in autonomous vehicle safety evaluation, behavioral modeling, and traffic signal optimization. His research on AV safety validation has been published in Nature and featured as a cover story. He currently serves as Managing Editor of the Journal of Intelligent Transportation Systems and a board member of ITS America. Through his leadership across academia and industry, he is helping to redefine how we design, evaluate, and deploy the next generation of intelligent transportation systems.

Related Info