Variable-drift diffusion models of pedestrian road-crossing decisions

Abstract

Human behavior and interaction in road traffic is highly complex, with many open scientifi?c questions of high applied importance, not least in relation to recent development efforts toward automated vehicles. In parallel, recent decades have seen major advances in cognitive neuroscience models of human decision-making, but these models have mainly been applied to simplified laboratory tasks. Here, we demonstrate how variable-drift extensions of drift diffusion (or evidence accumulation) models of decision-making can be adapted to the mundane yet non-trivial scenario of a pedestrian deciding if and when to cross a road with oncoming vehicle traffic. Our variable-drift diffusion models provide a mechanistic account of pedestrian road-crossing decisions, and how these are impacted by a variety of sensory cues: time and distance gaps in oncoming vehicle traffic, vehicle deceleration implicitly signaling intent to yield, as well as explicit communication of such yielding intentions. We conclude that variable-drift diffusion models not only hold great promise as mechanistic models of complex real-world decisions, but that they can also serve as applied tools for improving road traffic safety and efficiency.

Explaining human interactions on the road by large-scale integration of computational psychological theory

“…Early yield acceptance We use the term early yield acceptance to refer to the tendency of pedestrians to sometimes accept the gap in front of a yielding vehicle before the vehicle has come to a full stop [51][52][53]. We defined a scenario where the modelled pedestrian was initially standing at the kerb (as defined earlier), while the car was initially driving at its free speed at a TTA of 2 s, and applied a constant deceleration to yield to the pedestrian.…”

A Utility Maximization Model of Pedestrian and Driver Interactions

“…One requirement for doing so is to expand further the framework to model agent beliefs about the intentions of other agents, since this is an important cause of safety-critical interaction failures [50], [51]. Such an extension may also improve the model's account of non-critical interactions, because both empirical observation [2], [8], [30], and everyday experience suggests that belief about others' intentions plays an important role in the nuances in road user interactions. For example, a utility function in a vehicle agent can factor in a driver's belief in a pedestrian's crossing intention, and another utility function in the pedestrian agent can also factor in the pedestrian intention and how it is affected by the driver's actions.…”

“…An important feature of the sensorimotor control framework on which we base our model [23] is that it assumes that sensorimotor decisions are made not deterministically as described above, but by accumulation of noisy sensory evidence to a decision threshold, in line with extensive research in decision theory [40]. This assumption has proven crucial for describing full probability distributions of driver behavior in both routine and near-crash situations [23]- [26], and has also been recently adopted in models of pedestrian and driver road-crossing decisions [27]- [30]. We incorporate evidence accumulation into our framework by generalizing the deterministic decision variable above to the stochastically accumulated decision variable.…”

“…We incorporate evidence accumulation into our framework by generalizing the deterministic decision variable above to the stochastically accumulated decision variable. This generalization thus enables the probabilistic human errors that might happen in nearcrash situations [30].…”

“…Boda and colleagues [27] showed the framework could be extended to account for how drivers control longitudinal maneuvers when cyclists crossed the drivers' path. Others researchers have shown the onset latency of the road-crossing decisions in drivers and pedestrians could be modeled by the principle of analyzing sensory cues [28]- [30]. However, these models had only studied one agent in their interactive scenarios.…”