Motorists using automated driving systems overestimate their situation awareness and readiness to respond but are slower to recognise hazards compared with active "hands on" driving, a new study from Macquarie University shows.
The findings, published in the journal Applied Ergonomics, raise significant concerns about driver performance and road safety as increasingly advanced automation systems are integrated into new vehicles, the authors warn.
Researchers from Macquarie's Performance and Expertise Research Centre studied participants who used a sophisticated simulator to "drive" about six kilometres in automated and non-automated driving modes, asking them to rate their situation awareness and measuring their responses to hazards.
Surprisingly, drivers under automated driving conditions self-reported higher levels of situation awareness than those actively driving in non-automated mode.
However, they also demonstrated significantly poorer hazard recognition compared to those participants in non-automated driving conditions, taking a longer driving distance to react to both anticipated hazards (such as another vehicle with its turning indicators on) and surprise hazards (such as a pedestrian emerging suddenly from behind a sign).
"This paradox is an example of what we call the 'out-of-the-loop' phenomenon," says the study's lead author, PhD candidate George Nasser.
"When they're relieved of responsibility for control and as their trust in the vehicle increases, drivers tend to pay less attention to monitoring what the automated systems are doing," says Mr Nasser.
Driver disconnect
"Under these conditions, the driver's situation awareness declines and so does their ability to get back 'in the loop' and take control should they need to in the event of a hazard or malfunction."
The disconnect researchers found between drivers' perceptions and their actual driving performance suggests they develop inaccurate understandings or "mental models" of the automated driving system and what it does.
As an example of a worst-case scenario, Mr Nasser cites a fatal 2016 crash in the US involving a Tesla electric vehicle in which the driver had his hands on the steering wheel for only 25 seconds while the vehicle's 'autopilot' system was active for 37 minutes.
It's not realistic to expect a driver to undergo training when they buy a car with new features, so we think the solution lies in optimised human-centred design.
An incomplete understanding of the system's limitations led to the driver not intervening when the automated system failed to detect and respond to a truck crossing the vehicle's path.
"The driver's mental model – in other words, how fully and accurately the driver understands the automation system and the information it feeds back to them – determines their ability to take control," says Mr Nasser.
"What complicates this area is that our mental models of driving are constantly changing as we encounter more and more automated systems built into new cars."
Human skills remain crucial
The Macquarie team has also undertaken and published a systematic review of prior research exploring the role of mental models in enhancing driver performance in automated vehicles.
"What the review highlights is a need to keep drivers 'on the loop' — monitoring without direct control but remaining vigilant and ready to intervene if necessary," says Mr Nasser.
"Evidence suggests interfaces and visual displays that provide the driver with continuous easy-to-interpret feedback about system status and limitations are the best way to support accurate mental models and optimise take-over performance."
The study and systematic review are elements of a larger research project, funded by an Australian Research Council grant, looking at the effects of automation on skill decay in a range of human-automation settings where safety is critical, including aviation and industrial workplaces such as electricity generation and transmission.
Counterintuitively, the need to respond is more time-critical in driving than in aviation, says study co-author Dr Ben Morrison, who is also investigating the impact of automation on the skills of pilots.
"Pilots may have several minutes to detect a problem or system conflict, diagnose its cause, override systems like autopilot and take manual control if necessary, whereas a driver on the road has to perceive and respond to the unexpected in a matter of seconds," says Dr Morrison.
And while pilots undergo extensive re-training, simulator practice and certification when new flight automation systems are added, drivers aren't required to upgrade or test their skills when they start using a vehicle with advanced automated driving systems.
"It's not realistic to expect a driver to undergo training when they buy a car with new features, so we think the solution lies in optimised human-centred design," says Mr Nasser.
Dr Ben Morrison is Senior Lecturer in the School of Psychological Sciences at Macquarie University and a member of the Performance and Expertise Research Centre.
George Nasser is a PhD candidate in the School of Psychological Sciences at Macquarie University.
