Trust_banner4.png

Trust in Self-Driving Cars

It is imminent that self-driving cars will become an everyday aspect of our lives in a near future. The technology for it has been ready for quite some time; yet, the challenge has lied in its adoption and how it can be shaped to earn people’s trust. In this research project, we set out to investigate how we could objectively measure trust in this automation.

 
Tesla_Autonomous Vehicle.jpg
 
 
 

| Aim

Driving involves a great deal of monitoring different areas (e.g. the road ahead, side and rear mirrors, traffic lights, pedestrians, etc). Therefore, a driver’s eye gazing behavior is a potential objective metric. The question is whether there is a relation between trust and eye gazing behavior on specific areas when piloting a self-driving car? In other words, if a driver has a higher level of trust in an autonomous system, will she monitor less the areas typically used when driving herself?

 
 
 
In the  top , eye-gazing behavior concentrated in the road ahead suggests higher level of trust; while the  bottom  image suggest lower level of trust as driver monitors more often the dashboard, side mirror, and GPS.  Source:  Gold, Körber, Hohenberger, Lechner, & Bengler (2015)

In the top, eye-gazing behavior concentrated in the road ahead suggests higher level of trust; while the bottom image suggest lower level of trust as driver monitors more often the dashboard, side mirror, and GPS.

Source: Gold, Körber, Hohenberger, Lechner, & Bengler (2015)

 
 
 

| Research Design

To test whether eye gazing behavior correlates with trust, we first needed to create a condition that would cause participants to trust or distrust a car simulation during automated driving. For this, we employed the similarity-attraction theory in social psychology. Just as people tend to trust others that are similar to them in appearance, values or attitudes, we hypothesized that having similar or dissimilar driving goals (safety, comfort, fuel efficiency, and speed) between a person and a system would influence the level of trust.

Sixty participants experienced one simulation that shared and another that did not share their same driving goals. After, they self-reported in a questionnaire their level of trust towards both systems. During the simulations themselves, eye gazing behavior was measured using eye-trackers, in which we collected the frequency and duration of eye fixations in predefined areas of interest (rear and side mirrors, dashboard, GPS, and road ahead) versus the rest of the areas displayed on the screens. We postulated that drivers would show a different ratio between these two sets of areas based on their level of trust.

 
 
 
 
 
 

| Results

The manipulation of shared and unshared goals was purely psychological. There was no difference in the actual driving behavior of the autonomous system between the two conditions. This was due to the simulation software, which had limited customization. Thus, to avoid participants from noticing, each condition was experienced in either a highway or an urban setting.  

Most participants indeed believed to have perceived a difference in the driving behavior. However, it was not enough to obtain the differences in self-reported trust we posited. Instead, we found a significant difference in trust between the highway and urban settings, in which participants reported higher trust for the autonomous system in the highway.

 
 
 

 
 
 

| Lessons Learned

What I love about research is that continuous wonder of trying a theory, figuring out how to test it, and if things don’t go as expected (which happens often), understanding what can be done better next time. For these experiments, it might be that participants did not experience realistic risks to their safety because it was a simulation, so their levels of trust weren’t influenced as expected. If the driving behavior of the system could be actually manipulated, perhaps the shared-goals condition would be stronger. Also, we observed that participants were mostly looking at the center of the screen, which made it unclear whether they were paying attention to the system’s driving or it was simply a natural, comfortable position for the eyes.

The challenge that remains is how we can distinguish between the monitoring of a system due to trust concerns vs. paying attention for being curious or to become situationally aware. These two concepts are entangled in terms of eye gazing behavior, but arise through different underlying motives.