Behavioural markers for degraded human performance

Deliverable D1.2 of the H2020 project MEDIATOR
Author(s)
Borowsky, A.; Oron-Gilad, T.; Chasidim, H.; Ahlström, C.; Karlsson J.G.; Bakker, B.; Beggiato, M.; Rauh, N.; Christoph, M.; Cleij, D.; Kint, S. van der; Tinga, A.
Year

Vehicle automation has the potential to improve driving safety and driver comfort. The Mediator system aims to aid the realization of this potential by mediating between the driver and the automation on who is fittest to drive. Make this trade-off in a timely and safe manner, requires both driver and automation fitness to be detected and predicted for the near future. Within the MEDIATOR project, numeral representations of driver and automation fitness, i.e., time to driver fitness and time to driver unfitness and the equivalents for automation fitness, were therefore defined. To take the first steps in estimating these variables in real time, the research described in this deliverable, focuses on detection and prediction of the driver state and finding relationships between driver state and driving performance. These relations were then used for setting behavioural markers that indicate degraded performance, which form the scientific basis for the estimates of time to driver fitness and time to driver unfitness. This research leads to the main  outcome of this deliverable, i.e., the functional requirements for the driver module of the Mediator system.

The driver state detection algorithms for both fatigue and distraction have shown promising results and they will be further improved during the next phase of the project. Specifically, the preliminary prototype of the fatigue detection algorithm showed a Mean Absolute Error (MAE) of about 0.5 points on the Karolinska Sleepiness Scale (KSS), or a binary classification accuracy of about 90%. The accuracy for eyes-off-road classification measured with the F-score was 92.75%. Finally, the realtime comfort detection algorithm, which uses face tracking methods, was found to have potential in detecting users’ satisfaction with the current operations of the automated system.

The evolvement of fatigue and distraction while driving with L2 automation and their effects on driving performance, was investigated in one driving simulator study and one on-road study. These studies revealed that fatigue was induced faster under L2 driving conditions than under manual driving conditions. Furthermore, the results from the driving simulator study revealed that drivers who played a game of Simon under L2 driving conditions reported lower KSS scores than drivers who did not play it. This finding is encouraging as it demonstrates the importance of engaging with a secondary task under L2 driving conditions as means of stalling fatigue development. This also emphasizes the complex trade-off between being distracted versus being fatigued under L2 driving conditions. In terms of driving performance, the driving simulator study showed that under L2 driving conditions, regardless of whether drivers engaged with a secondary task, drivers had fewer number of glances on road hazards, which may indicate poorer situation awareness (SA) as compared to manual driving.

To evaluate the effect of driver distraction during assisted driving, a second driving simulator study investigated driving performance under both manual and L2 driving conditions. The results showed that drivers who were asked to engage with a secondary task during L2 driving conditions had a much smaller probability of identifying a hazard (~0.54) and deactivated the automation due to hazardous situations less often (probability of deactivating the automation =~0.116) than drivers who were not engaged with a secondary task (0.91, 0.22 respectively).

These findings indicate that, when distracted under L2 driving conditions, drivers' SA was poorer than drivers who were not distracted. Although drivers in this study were instructed to engage with a secondary task at specific time windows, the results have shown that drivers do regulate their behaviour and allocate more attention to the road in urban environments than in highway environments under L2 driving conditions.

Finally, long-term effects of drivers' comfort under L3 automated driving conditions were investigated via a literature overview. It was found that offline / long-term prediction of drivers’ comfort based on prior knowledge seems to be a promising approach to detect potentially uncomfortable driving situations in advance, to prevent a decrease in drivers’ comfort and to consider drivers’ preferences for automated vs. manual driving in future driving situations.

Ultimately it can be concluded that, to make a safe and comfortable trade-off between driver and automation fitness, driver states should be monitored and predicted while driving with and without automation functions. Based on the research described in this deliverable, as well as on the knowledge gained during the analysis and experimentation phase of the MEDIATOR project, a first set of functional requirements for the driver module was drafted. These functional requirements provide input to guide the further design and development of the Mediator driver state module.

MEDIATOR has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 814735.

Pages
114
Publisher
European Commission, Brussels

SWOV publication

This is a publication by SWOV, or that SWOV has contributed to.