Car-following models describe how drivers respond to vehicles ahead, and they underpin both microscopic traffic simulation and the study of collective traffic dynamics. However, building these models is difficult because human driving behavior varies widely across drivers, environments, and situations.
This paper introduces an agentic discovery framework that automatically derives car-following laws from multi-source trajectory data. Rather than relying on hand-crafted equations, the approach uses AI agents to search for formulas that are both accurate and interpretable, potentially offering a scalable way to capture behavioral diversity.
The abstract is the only source for this summary, and it does not provide details on the specific laws discovered or comparison with existing models. Still, the central contribution is clear: a move toward automated, interpretable model discovery in a domain that still depends heavily on manual specification.