CAV Pilot Development and Deployment in Midwest Winter

Principal Investigator(s):

Ziran Wang, Assistant Professor of Civil Engineering – Purdue University

Project Abstract:
Our research presents a novel approach for deploying Connected and Automated Vehicles (CAVs) in challenging Midwest winter environments through the integration of Vision Language Models (VLMs). Our system addresses the critical safety and mobility challenges posed by winter conditions through a VLM capable of interpreting complex visual scenes and natural language instructions to generate personalized control strategies. The framework incorporates a Retrieval-Augmented Generation (RAG) memory module that enables continuous adaptation to individual preferences through human feedback. Real-world experiments conducted on professional test tracks demonstrate that our approach reduces takeover rates by up to 76.9% compared to baseline systems, particularly when processing indirect user instructions. The system successfully interprets diverse winter driving scenarios, maintaining high safety standards while adapting control parameters to both environmental conditions and user preferences. Our comprehensive benchmark dataset, focused specifically on winter driving conditions, provides a foundation for future CAV development in extreme weather. This work represents a significant step in autonomous vehicle technology for winter operations and builds a framework for safer, more reliable transportation systems in extreme weather conditions in the Midwest.

Institution(s): Purdue University

Award Year: 2024

Research Focus: Safety, Mobility, Fairness

Project Form(s):