CDA.AI for OpenCDA: AI pathways for cooperative driving automation research

This position paper presents CDA.AI, a comprehensive artificial intelligence framework for Cooperative Driving Automation (CDA) that builds upon established ADAS and ADS technologies by integrating connectivity into a unified system. CDA.AI leverages both semi-end-to-end and end-to-end AI paradigms to enable scalable, interpretable, and adaptive multi-agent traffic systems. By facilitating real-time communication among connected vehicles and smart infrastructure, the framework supports system-level cooperative perception—employing diverse fusion strategies such as early, intermediate, and late fusion—along with coordinated planning and control. Moreover, CDA.AI incorporates closed-loop integration with large-scale vision-language models (VLM) and Vision-Language-Action (VLA) systems, directly enhancing reasoning capabilities and infusing higher-level intelligence into the perception and decision-making processes of CDA. This dual-pathway approach not only ensures high-performance end-to-end learning for complex decision-making but also retains the modular transparency needed for safety-critical applications and seamless human-CDA interaction. Use cases including multi-lane platooning, cooperative merging, and multi-modality sensing demonstrate how the framework improves both individual vehicle behavior and overall traffic system performance. Built upon the open-source OpenCDA ecosystem, this framework provides essential tools, datasets, and benchmark scenarios to accelerate research and facilitate the deployment of cooperative automated systems across diverse environments, while emphasizing a system-level perspective that supports both connected ADAS and ADS.