University of Glasgow Research Explores Social Cues for Safer Semi-Autonomous Driving

University of Glasgow Research Explores Social Cues for Safer Semi-Autonomous Driving

(IN BRIEF) Researchers at the University of Glasgow have found that animated virtual agents displayed through augmented reality glasses could help improve safety in semi-autonomous cars by using social cues, such as looking and pointing, to direct distracted drivers’ attention toward road hazards. In lab experiments involving 48 volunteers, participants wore AR glasses while playing a mixed-reality game and watching dashcam footage of driving scenarios, with warnings delivered either through a coloured marker or a robot head that turned toward potential danger. The study found that enhanced warnings, including red visual cues, helped distracted participants correctly predict hazards around 80 percent of the time, matching the performance of an undistracted control group. However, when the robot was given visible signs of stress such as sweating and trembling, participants became distracted by trying to interpret its emotions, suggesting that virtual agents may be useful for driver safety but must be carefully designed to avoid adding cognitive load or encouraging drivers to over-rely on automated cues.

(PRESS RELEASE) GLASGOW, 20-Aug-2026 — /EuropaWire/ — University of Glasgow researchers have found that animated virtual agents could help improve the safety of future semi-autonomous vehicles by directing drivers’ attention back to the road when hazards appear.

Researchers from the University’s School of Computing Science have investigated whether the social cues that people naturally interpret in human interaction could be used to alert drivers in conditionally automated cars to potential dangers.

Their peer-reviewed study shows for the first time that augmented reality glasses displaying an animated robot that looks and points toward hazards may be as effective as more conventional visual warnings.

The researchers also found that virtual agents may need to be carefully designed, as adding too much emotional expression or personality could distract drivers rather than help them.

The findings could support car manufacturers as they address one of the major challenges of the transition to autonomous vehicles.

Drivers of conditionally automated cars, which are largely but not fully self-driving, must remain aware of road conditions even when engaged in non-driving tasks, in case they need to take back control in an emergency.

Thomas Goodge, from the University of Glasgow’s School of Computing Science and first author of the paper, said his PhD research examined a central challenge of conditionally automated cars: how to keep drivers aware of road hazards when they are distracted by activities such as reading, answering emails or playing games on a phone.

He said people naturally understand where another person is focusing attention by reading head movements and gaze direction.

For example, when someone looks over another person’s shoulder, it suggests there may be something worth turning to see.

Goodge said previous psychological research has shown that people process social cues more efficiently than conventional visual cues, yet safety-critical systems rarely use such information.

Instead, they commonly rely on lights, flashes and beeps, which can be difficult to interpret immediately.

The research team set up two laboratory experiments to explore whether social cues could be used to support hazard awareness.

A total of 48 volunteers sat in a mock-up driver’s seat in front of a monitor displaying pre-recorded dashcam footage of real-world driving scenarios.

During the experiments, participants wore AR glasses and played a simple mixed-reality game in which they looked at gems to “pop” them with their gaze.

At the same time, dashcam footage played in front of them, simulating the divided attention of a driver occupied by a non-driving task.

The AR glasses presented a warning just before a potential hazard appeared in the footage and the video cut to black.

Participants were then asked to identify the hazard and predict what might happen next.

The same experiments were repeated with a control group that watched the footage without playing the game.

The warnings were presented in two main forms.

One was a coloured bar that moved to sit beneath the hazard.

The other was a robot head designed by the research team, which turned to look toward the danger.

In the first experiment, distracted drivers performed significantly worse than the undistracted control group with both types of warning, often missing hazards in the footage.

In the second experiment, the researchers added additional visual cues.

The coloured marker turned red, while the robot’s head turn was accompanied by the reddening of the sides of its head.

With these added cues, participants correctly predicted the hazards around 80 percent of the time, matching the performance of the undistracted control group.

The team also tested whether the robot’s apparent emotional state affected driver responses.

When signs of stress, such as sweating and trembling, were added to the robot, participants became distracted by trying to interpret its emotional state and more frequently failed to identify the road hazards correctly.

Professor Stephen Brewster, also from the University of Glasgow’s School of Computing Science and a co-author of the paper, leads the ViAjeRo project, funded by the European Research Council, which examines how virtual and augmented reality technologies can improve the experience of travelling in self-driving cars.

He said the findings suggest that simulated body language can help alert drivers to danger and may provide a useful way for people to act as co-pilots in self-driving cars.

Professor Brewster said that as cars move closer to autonomy, manufacturers are also adding features such as AI-enabled speech recognition to make driving feel more social.

He said the research suggests that social cues could support hazard awareness and help drivers and cars work together to improve safety.

However, he added that the study also highlighted an important challenge.

Some participants appeared to place more responsibility than expected on the visual cues, waiting for the marker or robot to turn red rather than paying closer attention to the road.

Professor Brewster said properly calibrating driver trust will be critical to ensure people do not hand over awareness entirely to the vehicle.

The team is now planning to expand the research to examine the potential of virtual agents in greater depth.

Goodge said few safety-critical systems currently make use of social information.

He said the next step is to explore other modalities, including sound and conversation, and to test how drivers respond to a more conversational agent rather than one they simply observe.

Professor Frank Pollick of the University of Glasgow’s School of Psychology also contributed to the research and co-authored the paper.

The paper, titled “The effects of using a Virtual Agent to Signal Danger on Hazard Prediction ability in Conditionally-Automated Driving,” has been published in ACM Transactions on Computer-Human Interaction.

The research was supported by funding from the UKRI Centre for Doctoral Training in Socially Intelligent Artificial Agents and the European Research Council.

Media Contact:

media@glasgow.ac.uk

SOURCE: University of Glasgow