Evaluating active driver intervention strategies in sequential conflict scenarios: a counterfactual reasoning-based...

Multi-round conflicts among different road users are a key contributor to the high incidence of intersection crashes. In sequential conflict scenarios, early round interactions can constrain the maneuvering space available in later rounds and influence road users' attention allocation, thereby producing highly volatile risk throughout the episode. Active interventions that target the process-level evolution of sequential conflicts are essential. However, most existing intervention and evaluation approaches are developed under a single-conflict assumption and do not provide a strategy set tailored to different risk levels in sequential conflicts, nor a consistent evaluation model for assessing intervention effectiveness. This study proposes a counterfactual reasoning-based strategy evaluation model that quantifies the net benefits of active interventions and explains how these benefits manifest through interpretable safety mechanisms. First, we develop a multi-layer indicator system that characterizes within-round risk and cross-round risk evolution. We then construct a mechanism-level outcome space with four interpretable dimensions using exploratory factor analysis. Next, trajectory-level counterfactual reasoning is employed to estimate the benefits of each strategy within this common outcome space. The proposed model is validated through a case study conducted on a six-degree-of-freedom driving simulator, from which risk-level intervention recommendations are derived. The experiment involves 30 drivers and 8 scenarios, and evaluates multiple intervention configurations, including offline, online and their combinations. All proposed strategy configurations yield positive net benefits across scenarios. In particular, the online voice prompts are most effective in reducing peak conflict intensity. The proposed model supports interpretable risk-level strategy selection for sequential conflicts and provides a transferable evaluation template for driver assistance systems.