CogMod: Driver Model for Augmenting Scenario Criticality

A typical approach for scenario-based testing of autonomous vehicles involves participating vehicles that are modeled using static, predetermined, and time-stamped trajectory information. These scenarios not only fail to capture the behavioral variability of human drivers but also limit the usability of the scenario with continuous software updates of the autonomous vehicles, as surrounding vehicles remain static while the AVs' behavior changes due to system updates. We present a human driver behavior model called CogMod, which is based on a hybrid cognitive architecture to represent human cognition during driving. This model surpasses the current state of the art in modeling human driving behavior in several ways. Contrary to most implementable models, where control is directly based on observed variables, CogMod actions rely on a rich internal representation. This internal representation is the result of a novel perception mechanism that enables the CogMod driver to have a selective update of the surrounding environment. The model is capable of simulating human drivers' perceptive and cognitive limitations, thus capturing human driving variability. To evaluate our model, we compare CogMod-generated microscopic distributions with real-world driving data obtained from the HighD dataset. Our result shows that our model can augment existing real-world driving scenarios, thereby increasing their complexity and criticality for testing autonomous vehicles.