Virtual robot "co-pilots" could help distracted drivers spot hazards in semi-autonomous cars, new research suggests.
University of Glasgow researchers found animated agents which use human-like social cues to direct attention towards danger could help drivers remain aware of the road while doing other tasks.
The technology could eventually be used in conditionally automated cars, which can drive themselves in some circumstances but require a human to be ready to take back control.
Researchers tested the approach on 48 volunteers sitting in a mock driver’s seat while watching footage of real journeys.
Participants wore augmented reality glasses and played a simple game designed to divert their attention from the road, simulating activities such as reading or answering emails while a car was driving itself.
Warnings were then displayed before potential hazards appeared.
One took the form of a coloured marker, while another used an animated robot head which turned to look towards the danger.
An initial experiment found distracted participants still performed significantly worse than a control group.
However, when researchers made the warnings more noticeable - including making the robot’s head turn and change colour - participants correctly predicted hazards around 80% of the time, matching the performance of people who had not been distracted.
Read more
- “I was scared”: the student who took a chance on opening her own café
- Design competition launched to rebuild two fire-ravaged Southside sites
- Court hears Graeme Dott told boy to 'stay out of room' while he abused young girl
- Landmark Castlemilk film to be shown 30 years on as community continues to fight
Professor Stephen Brewster, of the university’s School of Computing Science, said the findings suggested simulated body language could provide an effective way of drawing drivers’ attention back to the road.
He said it could offer "a potentially very useful way to empower people to co-pilot a self-driving car".
Researchers also found there could be a limit to how human the virtual assistants should appear.
When the robot was given visible signs of stress, for example - including sweating and trembling - participants became distracted trying to interpret its emotions and were less successful at spotting hazards.
Thomas Goodge, the paper’s first author, said people naturally understand where others are looking from movements such as turning their heads.
He said existing safety systems instead rely largely on lights, flashes and beeps, which may not be immediately understood.
The team now plans to investigate whether sound and conversation could make virtual assistants more effective.
The peer-reviewed research has been published in ACM Transactions on Computer-Human Interaction.
Share