A quantitative driver model of pre-crash brake onset and control

Abstract

An existing modelling framework is leveraged to create a driver braking model for use in simulations of critical longitudinal scenarios with a slower or braking lead vehicle. The model applies intermittent brake adjustments to minimize accumulated looming prediction error. It is here applied to the simulation of a set of lead vehicle scenarios. The simulation results in terms of brake initiation timing and brake jerk are demonstrated to capture well the specific types of kinematics-dependencies that have been recently reported from naturalistic near-crashes and crashes.

How Do Drivers Respond to Silent Automation Failures? Driving Simulator Study and Comparison of Computational Driver Braking Models

“…(1) Several models of driver reactions in rear-end scenarios have been developed based on these ideas (Flach et al, 2004;Markkula, 2014;Markkula et al, 2016;Engström et al, 2017;Venkatraman et al, 2016;Svärd et al, 2017). More specifically, these models suggest that drivers react after some fixed looming threshold, or after accumulation (integration) of the looming signal to a threshold, potentially also together with other perceptual cues such as brake lights (Markkula, 2014;Engström et al, 2017;Xue et al, 2018).…”

“…More specifically, these models suggest that drivers react after some fixed looming threshold, or after accumulation (integration) of the looming signal to a threshold, potentially also together with other perceptual cues such as brake lights (Markkula, 2014;Engström et al, 2017;Xue et al, 2018). The accumulation of the looming signal was included in the model by Svärd et al (2017), based on a framework by Markkula (Markkula, 2014;Markkula et al, 2018), but this model also assumed that drivers in emergency rear-end situations react to unexpected looming rather than to looming per se (Engström et al, 2018). The unexpected looming can be understood as the discrepancy between the predicted and actual looming, that is, the looming prediction error.…”

“…The accumulative part of the driver reaction model described by Svärd et al (2017) has the following form:…”

“…The parameter k in Equation 2 can be interpreted as the gain determining the impact of the prediction error on the accumulator while m can be interpreted as the sum of all non-looming evidence for and against the need of braking (Svärd et al, 2017;Markkula, 2014).…”

Models of Human Decision-Making as Tools for Estimating and Optimizing Impacts of Vehicle Automation

“…Also, before the first brake application, decision unit 2 is inactive. The looming-related inputs to the network, all set to zero if the driver's eyes are off the road, were defined as: , modeling the perceptual decision of whether braking is solving the conflict., modeling the decision to increase braking(12). min , modeling a positive looming contribution to the decision to change lane and a negative contribution to the same decision coming into play once , where . The remaining inputs were defined as:  and , both if the driver is directing gaze towards the adjacent lane (whether by mirror checks, shoulder checks, etc, is not defined in the model), zero otherwise. .Manual tuning indicated satisfactory model behavior for s s s .MODEL BEHAVIOR -REPRODUCING FINDINGS FROM LITERATUREReproducing pedestrian crossing behavior shows example simulations with the pedestrian model, all at 50 km/h (31 mph) andwith TTC = 4 s at the moment when the pedestrian steps up to the zebra crossing, at which point the simulated car also initiates a deceleration to stop exactly at the zebra crossing.…”