Fleet News

Autonomous car tests show time lag for drivers to retake control

An early average benchmark of 3.4 seconds has been established for the time it will take a driver to take back control of an autonomous vehicle. The study found that a combination of alerts, rather than just visual, audio or vibration, gave the best results.

Liability around autonomous vehicles is being looked at closely by legislators, with the ability of a driver to take back control a key element in rolling out the technology on public roads.

Nuance Communications, a global technology company that specialises in voice control, and artificial intelligence specialist DFKI ran a simulation to find out how quickly humans are able to resume control after being engaged with other activities like reading, answering work emails or watching videos.

Both Nuance and DFKI work with a variety of car manufacturers on computational linguistics and how humans interact with technology.

The simulation involved 30 drivers in a level 4 automated vehicle for one hour at a time and they knew they would have to take back control of the vehicle at a random point in the test.

While falling asleep at the wheel for 3.4 seconds could prove fatal, a level 4 vehicle is one which is sophisticated enough to not need driver attention for safety, with the driver even able to go to sleep or leave the driver’s seat.

Self-driving is only supported in specific situations. The car must be able to safely stop a trip on its own and park the car in a safe place if a driver does not retake control.

The House of Lords Science and Technology Committee heard in March that Level 4 vehicles are not expected to be widely available on UK roads for at least another five years. 

While participants were engaged in activities during the test, the autonomous system would alert them through vibration (on a smartwatch), with visual and auditory cues to see which of the senses responded the fastest to take the wheel.

The study included a variety of scenarios including inclement weather conditions, system diagnostic warnings, sensor defects, traffic jams and general rules of the road.

A combination of voice, touch and visual cues were preferred by the majority of participants.

Drivers weren’t as responsive when using the same type of alert that matched their current activity. For example, if a visual display was given while someone was reading a book, a combination of alerts was needed.

A key finding from the study concludes that future autonomous vehicle systems will need to have contextual data and information from the car to provide the most appropriate alert.

Independent of the current driver activity, sound was considered as more pleasant and effective than visual cues, leading to faster reactions than simply vibration alerts.

Interestingly, Nils Lenke, Nuance communications senior director of corporate research, told Fleet News that the reaction time was the lowest when the driver was engaged in a audio activity, such as listening to a book or music, something drivers can already do legally at the wheel today.

While having a benchmark on a reaction time is useful, Lenke said it was too early to say definitively what reaction time is reasonable or safe.

He said: “It would have to be decided what is a reasonable handover time for drivers and I think we’re only really going to get to that point with practical testing. Governments and vehicle manufacturers want more information and I think there is still a lot more work to be done.”

He said further research is also needed around cognitive load. This involves monitoring the optimal amount of stimulation humans can handle and still remain concentrated or alert.

Lenke said: “You can overload cognitive capacity, but there’s also the other end of that where if you’re not giving enough stimulation to the passenger to engage with, they can become sleepy or drowsy.

“It could be that in the future the car will look at the road conditions and situation and may either block certain activities or promote certain activities dependent on the status of the driver.

“It may be that if someone is answering work emails, that would aid concentration more than just sitting there and doing nothing at all.”

Click here for fleet management software best practice and procurement insight

Login to comment


  • The Engineer - 28/07/2017 11:02

    3.4 seconds? which probably according to the tinfoil hat brigade at Brake would say a car can travel 8.2 miles and destroy an entire city. Ban cars, ban all of them!!!!

  • AlanD - 28/07/2017 21:35

    I'm surprised it was that quick ! Meanwhile at the other end of the spectrum for conventional vehicles - under permanent control of a driver, we have Brake pushing their B.S. that "thinking time" in the highway code is too short - I don't agree. Also they seem to have conveniently forgotten to mention braking distances were based on 1950s tyres and brakes. ...

  • William Forbes - 31/07/2017 10:23

    "and they knew they would have to take back control of the vehicle at a random point in the test" completely invalidates these test results. Subjects have been "pre-conditioned".

  • Diarmuid Fahy - 01/08/2017 11:35

    This was always going to be the case, and to my mind the sensible thing to do is skip Level 3 and move directly to Level 4 autonomy.

Compare costs of your company cars

Looking to acquire new vehicles? Check how much they'll cost to run with our Car Running Cost calculator.

What is your BIK car tax liability?

The Fleet News car tax calculator lets you work out tax costs for both employer and employee