Improving models of pedestrian crossing behavior using neural signatures of decision-making

Abstract

Understanding and modelling pedestrian behavior is important for traffic safety, not least in the context of vehicle automation. There exist competing models for how pedestrians decide if and when to cross a road with oncoming traffic. Distinguishing between these competing models is non-trivial, but recent results in the cognitive neuroscience of decision-making offer a promising method, complementing behavioral data with electroencephalography (EEG): Previous EEG studies have shown that the centro-parietal positive potential (CPP) reflects evidence accumulation during abstract perceptual decision-making tasks, and that it can be used to arbitrate between alternative models of these tasks. However, it is not yet known whether the CPP can be used to support modeling in more complex, embodied contexts, such as human locomotion in road traffic. Here, we address this question by designing an EEG paradigm for pedestrian road-crossing. In a computer-based experiment, participants made road-crossing decisions in a simulated scenario where a car approached them under different time-to-arrival (TTA) conditions. Three perception-based drift diffusion models and one utility-based drift diffusion model were used to model the pedestrian behavior. The behavioral data showed a partial preference for the utility-based model over the perception-based drift diffusion models. The EEG data showed a CPP signal, which helped distinguish between the models in a way that behavioral data alone could not: CPP amplitude was positively correlated with accumulated evidence in the drift–diffusion models, and with stronger correlations for the utility-based model than for the perception-based models. Our results show that the CPP signature can be used to help arbitrate between competing decision-making models also in more embodied tasks, a finding which has applied implications not least in the context of traffic safety engineering and vehicle automation.