Research roadmap charts path for AI foundation models in cyber-physical systems engineering

Researchers have published a strategic roadmap for deploying foundation models—particularly large language models and multimodal AI systems—across the software engineering lifecycle of cyber-physical systems (CPSs). The work addresses a significant gap in current research, which has largely concentrated on LLMs while overlooking other types of foundation models and their potential applications.

Why multimodal AI matters for cyber-physical systems#

Cyber-physical systems such as autonomous vehicles, industrial controls, and IoT networks handle diverse data types including images, sensor readings, and audio streams. The researchers argue that beyond text-based LLMs, vision-language models and other multimodal approaches can better serve these complex engineering challenges. Their ability to process and integrate multiple data types makes them particularly suited to CPS environments where visual, audio, and structured sensor data all play critical roles.

Research gaps and opportunities#

Most existing literature has focused narrowly on how LLMs can assist with coding and testing tasks. The roadmap identifies substantial untapped potential by mapping foundation models to standard CPS software engineering phases, making the research accessible to practicing engineers. The authors highlight that vision-language models, audio-based models, and integrated multimodal systems remain largely unexplored in this domain.

Cross-cutting challenges ahead#

The research identifies six key dimensions of challenges that apply across foundation model adoption in CPS work: technical constraints, economic and resource considerations, human factors, and others. Understanding these constraints will be essential as the community experiments with deploying foundation models in production environments. The roadmap provides actionable research opportunities and calls for empirical studies to validate emerging approaches.