Modular nudging models: Formulation and identification from real-world traffic data sets

The vehicle nudging behaviour suggests that a vehicle in the traffic flow may induce a ‘pushing effect’ to its preceding vehicle. In other words, while the traditional vehicle-following behaviour results in look-ahead interaction, the nudging behaviour may result in look-behind interaction: the combination of the two effects would result in bidirectional inter-vehicle interactions. Unfortunately, all reported numerical examples and traffic simulators indicating that nudging may improve the traffic flow with artificially engineered nudging behaviour. It is still unclear if such behaviour really occurs and is crucial in our roads. To address this question, this work proposes “modular” nudging models, meaning that the model is able to describe both the look-ahead-only scenario (with only vehicle-following behaviour) and the look-ahead-and-behind scenario (with both vehicle-following and nudging behaviour). We apply this modular philosophy to traditional models (optimal velocity model, intelligent driver model) and to a physics-inspired neural network model. By using the NGSIM real-world traffic data sets, the models suggest that the nudging effect plays a smaller and smaller role as the model accuracy improves.