Font Size: 

Recent advances in sensing, simulation, and computation have significantly expanded the technical capabilities of transportation engineering. More important, however, is a quieter methodological shift: traffic systems are no longer studied only as objects to be modelled and reproduced, but increasingly as dynamic systems that must be observed, inferred, and updated under changing conditions. Traditional laboratory traffic research has relied on deterministic models, manually calibrated simulation environments, and validation workflows built around relatively stable assumptions of observability and control. These assumptions remain useful, but they are increasingly strained by the nonlinear, stochastic, and behaviourally unstable character of contemporary urban traffic. This paper argues that the most important contribution of Artificial Intelligence (AI) to traffic engineering is not only computational, but methodological. It proposes an AI-enhanced laboratory framework in which machine learning, deep learning, and reinforcement learning are embedded directly into the architecture of traffic experimentation, not as auxiliary optimization tools, but as internal mechanisms of inference, calibration, and decision support. The proposed framework integrates multimodal sensing, feature engineering, predictive learning, adaptive control, closed-loop simulation, and reproducible validation within a unified experimental system. To illustrate the methodological implications of this approach, the paper develops a preliminary validation scenario based on a calibrated simulation model of a signalized urban corridor and compares the expected behaviour of the proposed framework with that of a conventional deterministic workflow. The comparison is not intended as a field deployment benchmark, but as a controlled methodological contrast designed to examine differences in predictive robustness, calibration burden, and control adaptability. The results suggest that AI changes the role of the traffic model inside the experiment: from a static representational object to an adaptive component of experimental reasoning. The study contributes a methodological foundation for traffic laboratories designed not only to simulate mobility systems, but also to continuously improve their predictive capabilities through experimentation, while providing a modular architecture that supports the future integration and validation of advanced AI-based traffic management strategies.