Open-world mobility describes dynamic real-world environments in which vehicles, robots, infrastructure, and people interact under partial observability, distribution shift, long-tail events, and changing operational conditions. These settings require AI systems that can jointly perform perception, semantic reasoning, prediction, and closed-loop action under real-time constraints. In this talk, I present our recent work on physical AI for autonomous driving and open-world mobility, focusing on model architectures and system designs that connect foundation models to deployable autonomy. I begin with Driving with Regulation, which treats traffic laws, norms, and safety guidance as structured inputs to decision-making through retrieval-augmented regulation understanding. I then present AutoVLA, a vision-language-action model for end-to-end autonomous driving that unifies scene understanding, semantic reasoning, and trajectory generation within a single autoregressive framework. Next, I discuss a central systems problem for deployable VLA-based autonomy: the temporal mismatch between slower high-level inference and fast, safety-critical control. Our work explicitly models delayed semantic updates during action generation and enables latency-aware integration of reasoning and control in dynamic environments. Finally, I discuss our work on multi-agent perception and prediction, including V2XPnP, to show how these ideas extend beyond single-agent autonomy toward cooperative intelligence. Overall, these efforts illustrate our pathway toward scalable, trustworthy, and deployable physical AI systems that bridge foundation models, structured reasoning, and real-world control for open-world mobility.
Dr. Jiaqi Ma is a professor at the UCLA Samueli School of Engineering with joint appointments in Civil & Environmental Engineering and Computer Science, and he serves as Director of the FHWA/UCLA Center of Excellence on New Mobility and Automated Vehicles. Dr. Ma’s research focuses on Physical AI for autonomy, building foundation-model-driven systems that tightly integrate multimodal perception, world modeling, semantic reasoning, and closed-loop control for autonomous driving and mobile robotics. He has led and managed a large portfolio of projects funded by the U.S. and state Departments of Transportation, National Science Foundation, IARPA and ARPA-I, as well as industry partners including NVIDIA, Motional, and Amazon. Dr. Ma is Editor-in-Chief of the IEEE Open Journal of Intelligent Transportation Systems. He also serves as Chair of the Transportation Research Board (TRB) Standing Committee on Connected and Automated Vehicle Systems and is a member of the Board of Governors of the IEEE Intelligent Transportation Systems Society.