Affording Realistic Stopping Behavior: a Cardinal Challenge for Driving Simulators

Drivers adopt nearly constant deceleration rates when intending to stop at a distant target. In driving simulators however we often observe different control strategies. Depending on simulator fidelity, deceleration profiles range from saw-tooths to multi-modal profiles to profiles that do resemble those seen in reality. These deviations from reality suggest that drivers have difficulty controlling their vehicle based on the assumption that a constant deceleration rate (approximately constant brake pedal depression) is most efficient at least from a control and attention point of view. The discrepancy between reality and simulation is attributed to the fact that perception of distance, speed, and acceleration, as well as time-to-collision are biased in all current driving simulators (i.e. they are perceived at scaled magnitudes). We introduce a driver model for stopping behavior and demonstrate its capability to reproduce the braking profiles observed in various driving simulators by varying only two model parameters: i) control gain, and ii) the perceived time-to-collision at which braking is initiated. For a given set of simulator dependent perceptual biases in distance, speed, and acceleration estimation, it appears that drivers only need to adapt these two parameters to achieve efficient deceleration, which in some cases takes surprisingly much practice. The reason is attributed to the fact that such a simple adaptation is perceived as counter intuitive when the perceptual biases do not satisfy a particular ratio. If for example, acceleration perception is accurate but speed and distance are severely underestimated then, a simple intuitive adaptation cannot produce efficient braking; in that case, either a heuristic may be needed to decelerate (e.g. brake hard at the last possible moment), tight feedback control may need to be employed (resulting in a non-constant deceleration profile), or a non-intuitive gain adaptation may need to be stumbled on. In exploring the limitations of the model, we derived at the following hypothesis: the optimal scale factor, defined by the ease with which drivers can arrive at an efficient stopping strategy, in a driving simulator is a strong function of the biases in distance, speed, and acceleration perception. The theory underlying this hypothesis is discussed, as are its implications for simulator design and driver distraction research.