Leveraging AI Planning in a What-if Analysis Framework for Assessing Traffic Signal Strategies

The widespread availability of data allows traffic authorities to increasingly exploit Artificial Intelligence (AI) techniques for urban traffic management and control. This requires understanding the strategies generated by automated approaches and assessing their suitability to changing or unexpected circumstances. What-if analysis is a well-established method for evaluating the adaptability of strategies to changing conditions and exploring hypothetical scenarios. This paper introduces a framework for conducting what-if analysis using AI Planning to support traffic operators and authorities. AI Planning is well-positioned for supporting what-if analysis, thanks to its capacity for concise knowledge representation, its support for validation and verification of encoded knowledge, and its efficiency in simulating described conditions. We characterise the classes of what-if scenarios addressable by the proposed framework, and we demonstrate its ability to handle a range of cases using real-world data.