Abstract
A number of driver models were fitted to a large data set of human truck driving, from a simulated near-crash, low-friction scenario, yielding two main insights: steering to avoid a collision was best described as an open-loop manoeuvre of predetermined duration, but with situation-adapted amplitude, and subsequent vehicle stabilisation could to a large extent be accounted for by a simple yaw rate nulling control law. These two phenomena, which could be hypothesised to generalise to passenger car driving, were found to determine the ability of four driver models adopted from the literature to fit the human data. Based on the obtained results, it is argued that the concept of internal vehicle models may be less valuable when modelling driver behaviour in non-routine situations such as near-crashes, where behaviour may be better described as direct responses to salient perceptual cues. Some methodological issues in comparing and validating driver models are also discussed.
Notes
1. Various approaches were explored for fitting the linear model also to recordings with more severe yaw instability, but were not found to improve the fit of the resulting driver model.
2. Following Land and Horwood [Citation40], Salvucci and Gray [Citation37] specified preview in terms of angles down from the horizon, which in practice amounts to the same as using a preview distance. Here, it was also attempted to make the preview speed-dependent, as in the MacAdam and Sharp et al. models, e.g. , but if anything this reduced the model's ability of fitting the human data.
3. Alternatively, one could have rerun the studied scenario in closed-loop simulation from initial conditions, fitting parameters to achieve a match between resulting driver steering histories or vehicle trajectories. Such an approach was not adopted here, both due to it being several orders of magnitude more computation-intensive, and since the inherent instability of the low-friction scenario would presumably have rendered fitting very difficult; a small error in driver model or initial conditions can lead to large deviations in scenario outcome.
4. Except for one single subject driver where the number of available instances was odd, for which one more instance was allocated to the training set.
5. A p<0.05 criterion for statistical significance is adopted here.
6. For studies of a yaw angle nulling model with a desired path, see [Citation53] or [Citation54].
7. On a circular road, far point rotation nulling corresponds to nulling of yaw rate error.