Robot cues helped distracted drivers spot hazards in lab

A robot in a pair of augmented-reality glasses turns its head toward a road hazard. In a University of Glasgow lab, researchers tested whether that simple social cue could bring distracted drivers' attention back to the road in a conditionally automated car—one that drives itself most of the time, but still expects a human to intervene.

The team placed 48 volunteers in a mock driver's seat in front of prerecorded dashcam footage. While the footage played, participants used their gaze to pop virtual gems, imitating the divided attention of someone reading, answering emails or playing a game. The glasses then showed either a colored marker moving beneath the hazard or a robot head turning toward it before the footage cut to black.

Distracted participants performed significantly worse than people who watched without the game. The result changed when the warnings gained extra cues: the marker turned red, and the robot's head sides reddened as it turned. In both tests, participants predicted what would happen next correctly around 80% of the time, matching the undistracted control group. The paper, published in ACM Transactions on Computer-Human Interaction, describes this as the first demonstration of the approach using AR glasses.

The social signal has a boundary. When researchers added sweating and trembling to make the robot appear stressed, participants focused on interpreting its feelings and regularly failed to identify the road hazards. Some also trusted the visual cue too much, waiting for it to change instead of continuing to monitor the road. Professor Stephen Brewster and first author Thomas Goodge say driver trust will need to be calibrated as these systems develop.

So what changes in practice? Car makers could have another way to call back a driver's attention when a semiautonomous vehicle needs help, using a cue people already read instinctively: where another agent is looking. The evidence is still from lab experiments, not a deployed vehicle, and the Glasgow team plans further research into how virtual agents might support people traveling in self-driving cars.