The level of automation in vehicles is growing. But until all vehicles are completely automated, there will be a transition period where automated vehicles and human drivers coexist. Because these road users will coexist, it is necessary that automated vehicles understand human drivers and vice versa. This study aims to create a model that predicts human risk perception in different driving scenarios, to provide an understanding of the fundamental features of human threat perception while driving. The model created is a multi-criteria decision-making process that uses KITTI Vision Benchmark data as an input. This model is tested against the data gathered by an online survey, where 1918 participants answered the question: "How high is the risk on a scale from 0-10?" for 100 situations, chosen from the KITTI Vision Benchmark data. The survey response data is then compared to the model. Analysis of the survey data revealed that risk perception of driving situations is non-linear in the extremities of risk, showing that the input image are perceived as normally distributed instead of uniformly distributed. The comparison further shows that a model with features and weights solely based on literature is only slightly capable of predicting the risk of situations with a Pearson correlation coefficient with the survey responses of 0.28, whereas a model with feature weights optimised is moderately capable of predicting the risk of situations with a correlation coefficient of 0.57. However, multivariate regression is better capable of predicting risk with a correlation coefficient of 0.70, and shows that features and weights based on literature were not enough to establish an accurate model. The features that have the most impact on the result are the information about other road users' location and heading, the ego vehicle velocity, and the road type. Further research should focus on determining if speed is the cause of the perceived risk, or if it follows from other unknown predictors. Furthermore, an extension of the questions asked of the participants and the usage of videos instead of images can clarify the discrepancy between the literature-based model and the perceived risk of the participants. ... Read more

The mental model is a very rarely studied topic when it comes to human-machine interaction. Nearly all the studies that are done in this field are on measuring the actual performance of the human while performing the collaborative task with the automation system rather than understanding the complete picture of how the human is understanding and interpreting the automation is his actions due to just chance or does he/she have logical the reasoning for his/her actions.
This research consists of recognising the factors that affect the mental model, designing the experiment that measures the mental model as well as the factors that affect it. This work focuses on ALKS (Automated lane keep assist system) which is a subset of the SAE level 3 automation system for which the rules are laid down by UNECE(United Nation Economic Commission for Europe).
The approach of this research consists of an experimental setup in which the participants are given training and then are allowed to experience the level 3 automation in the driving simulator. The training and the simulation are designed to give a different depth in the functionalities and limitations of the system. The weak training just briefly describes the limitations of the system and how to interpret the user interface based on research done by Strand [35] with the consumer of level 2 automation system, on the other hand, the stronger training explains in detail the functionality and the limitations of the system along with the video explanations of how the automation will react in the different scenarios. Both the training are available in English and Dutch depending language preference of the participant. The participants are divided into 2 groups both of them are given different training but both of them experience similar driving conditions on the driving simulator.
Later the performance of the mental model, as well as the performance of the factors affecting the mental model are measured. Each participant receives the training once and is subjected to 3 trials of the driving simulation to understand the learning effect of the mental model and its related factors. The results of the mental model performance show that there is a significant gap of 5.5% between the weak and the strong mental model group just after the training, this trend then continues until the last trial where a significant gap of 6.5% in the mental model score.
A positive learning curve is also observed starting from the training to the last trial, the learning curve has a positive trend but the data is statistically insignificant to show the difference in the learning rate of the two different groups. There is a significant gap observed between the automation acceptance level of the two groups, although there is no significant rise in the level of acceptance of automation from just after the training to the last trials their difference in the acceptance level is present after the training and all the trials. For the trust level, the difference between the two groups is statistically insignificant, the difference is too small and the number of participants for the experiment is too less. Analysing the pragmatic significance of the trust data, a positive trend can be seen for the strong group whereas the weak group shows a negative trend, even though at the start of the experiment the weak group had a higher trust level. For situational awareness, the weak model group shows the lower situation awareness throughout the lap as compared to the strong mental model group.
... Read more

Measuring cognitive load is essential for understanding driver performance. Under- and overload can result in dangerous situations on the road. Cognitive load can be estimated by monitoring the diameter and movements of the pupils, but during measurements external influences such as changes in light intensity affect pupil diameters. In this paper, we present a novel method for quantifying light intensity with a head-mounted eye-tracker by weighting pixel values around the gaze direction. We demonstrate its effectiveness in cognitive load classification systems that use pupil metrics only. 54 participants in two separate studies have carried out n-back tasks during a simple driving task in a driving simulator. The data is classified by cognitive task (baseline, 1-back, 2-back) with the Random Forest algorithm. The resulting systems are 92.5% accurate with and 85.9% accurate without gaze features available, but are unable to generalise to participants unseen in the training phase of the algorithm. ... Read more

Long before humans will completely trust fully automated vehicles, partial and conditional automation where the human driver is still in the loop will dominate the era of autonomous vehicles. However, more than 90% of traffic accidents are due to human errors, of which approximately half appear to be due to perceptual errors. Especially at busy and complex intersections that have a high density of visual stimuli. This poses a high demand for accurate measurement of the driver's situation awareness, for real-world driver monitoring. Eye tracking seems to be an ideal method to determine what the driver has or has not seen, since people tend to look at what they inquire information from. The main objective of this thesis assignment was therefore to develop a platform that combines the driver's gaze behaviour in combination with the vehicle's road scene perception, to set up a real-world driving experiment to gather such data on the road, and to come up with a proof of concept that gaze behaviour combined with situational knowledge can be predictive of SA. The platform developed consisted of an eye-tracker with four cameras constructed in the available Toyota Prius of the department of Intelligent Vehicles, which is equipped for self-driving. The driver's gaze was layered over the object identification data from the vehicle, to see which objects are looked at or fixated upon and which are not. A real-world driving experiment was then conducted in which participants (N = 14) performed a driving task and a recall task. The driving task consisted of 8 intersection crossings in which mostly left turns were made to manoeuvre the vehicle off a main priority road. After each crossing, the participants performed a recall task in which they had to select images of the object they encountered during the driving task. The results showed that 88.1% of all relevant objects they encountered were seen with central vision, of which 41.8% were recalled. The remainder 11.9% of all relevant objects that were not seen, have only been in peripheral view, of which 18.2% were recalled. These preliminary results indicate that at least 2.2% (18.2% of 11.9%) of relevant objects are perceived by the driver using peripheral vision. The variables seen, first saccade angle and first saccade moment contributed significantly to a prediction model that predicted whether a relevant object would be recalled by the driver. The variables fixation count, total glance duration and saccade count were not significant predictor variables. The conclusion was drawn that the results of this exploratory research confirmed that gaze behaviour combined with situational knowledge can be predictive of driver SA. However, a crucial recommendation for future research is an improved recall task procedure to obtain higher recall rates and therefore more accurate prediction of SA. This outcome could be of great value for future research and development of applications that assist or steer the driver's attention to possible threats or objects the driver is missing or not focusing on. ... Read more