Most of us have seen self-driving cars in movies (e.g. Minority Report) or first design studies of driverless cars lacking steering wheels or pedals. Automation enthusiasts see an increased safety, economic benefits and higher efficiency as benefits of self-driving cars. The cars that one can find on the road nowadays are oftentimes already equipped with driver assisting technologies such as self-parking technology or lane keeping assistance. Still, it will take some more years for driverless cars to conquer the streets. This is partly due to technical limitations but also considerations on liability need to be done by policymakers. In the meantime, interesting questions on how humans and cars can work together efficiently in a (semi-)automated environment come up.
To classify the terminology in automated driving, the Society of Automotive Engineers (SAE) came up with an important taxonomy that is widely used in academia and industry. The taxonomy includes six automation levels ranging from no automation to full automation. With no automation (Level 0), the engagement of the driver is always needed whereas a fully automated vehicle operates independently under all conditions (Level 5). Currently, much research is done somewhere in the middle ranging from engagement of the driver needed (Level 2) over need to monitor and being ready to take over (Level 3) to a degree of automation that allows fully automated driving under certain conditions (Level 4).
Take-over time, take-over request and time budget
Until we reach the state of full automation, scientists and the automotive industry need to find ways how the human driver and the partially automated car can work together best. Especially when the automated car reaches its technological boundaries, it’s safety-critical that the human driver is able to resume manual control. For that reason, many studies on the take-over time have been performed during the last years.
Recently, scientists from the University of Twente, Delft University of Technology, TNO Traffic & Transport and SWOV Institute for Road Safety Research published a meta-analysis of 129 studies on take-over time to find overall patterns in take-over situations and valid determinants of take-over time.
“the time that drivers take to resume control from automated driving after a critical event in the environment or after having received a TOR [take-over request]”definition of take-over time (TOT), cited from Zhang et al. (2019), page 3
The following illustration explains that take-over time is the sequence between a take-over stimulus (e.g. an auditive signal or request to take over) and the moment, the driver starts the manual driving process. Take-over time is most often measured in seconds. Another important concept is the time budget, which is defined by the time the driver has to take back control until the car’s system limit is reached.
A meta-analysis is a statistical procedure that synthesizes data from multiple studies to find common patterns or identify variation between study outcomes which makes it statistically powerful. With that, the conclusions drawn from meta-analysis are way more generalizable than from single studies.
The 2019 preprinted research synthesized data from 129 studies on take-over time with a total of 4556 participants. The studies they investigated were performed in driving simulators varying in their fidelity. Driving simulation ranges from computer set-ups over driving simulators with motion platforms to real cars on test tracks.
Determinants of take-over time
The researchers found correlations between take-over time and the given time budget. That means if the driver has more time to take over until the system boundary is reached, they will also take more time to intervene. The scientists state that this correlation can be explained by the driver’s motivation to take over quickly which is higher if the situation is urgent. This finding should be considered by car manufacturers and lead to questions like “how can we motivate drivers appropriately?” “how can we help the driver assessing the urgency of a situation?” and more. The time budget could be increased by further developing the sensors of the vehicles, so the system has a larger look-ahead time.
Another finding doesn’t come much surprisingly but may disappoint most of you. The take-over time increases strongly when the driver was performing a task with a handheld device such as a smartphone or tablet before the take-over request started. That means it takes us much more time to resume manual control. Our dreams of watching movies on our tablets, checking e-mails on our smartphones or preparing a presentation on our laptop while sitting in an automated car may come to nothing… The reasons for that increase in response time are the visual distraction of using handheld devices and the need to switch posture and move arms and hands when starting to drive manually again – so listening to a podcast is the better option.
The auditory system also plays an important role when it comes to the modality of take-over requests. The meta-analysis showed that it’s better to use auditory or vibrotactile stimuli to request take-over instead of visual-only. Visual requests to take over can be easily overseen and are not considered as urgent warnings (or how many kilometers did you drive with your check oil indicator light on? 😉). Instead, auditory stimuli are well-known for being more alarming and widely used in other automotive applications (e.g. the annoying sound of you parking assist system). Again, this finding needs to be considered by car manufacturers that aim for a safe and efficient human-vehicle interaction.
The scientists emphasize that also experience has a strong effect on the take-over time. Here one needs to know that simulator studies often use multiple trials with the same participants to generate more data (with less persons = less costs). But during the meta-analysis the scientists found that subjects responded 1 second faster the second time a take-over situation occurs compared to their first try. That means, the average mean take-over time of 2,72 seconds should be interpreted with caution. It’s very likely that it will take longer to get back in the driver’s seat in real life. But the finding has also a good side: it could be an indicator that taking over control can be trained. Again, we can see how big the “human factor” will be in “autonomous cars”.
Source: Zhang, Bo & de Winter, Joost & Varotto, Silvia Francesca & Happee, Riender & Martens, Marieke. (2019). Determinants of take-over time from automated driving: A meta-analysis of 129 studies. 10.13140/RG.2.2.33648.56326.