In a crash-avoidance situation, a driver has fractions of a second to decide: brake, swerve, or both. It happens faster than conscious thought, shaped by perception, prediction, and physical limitation all at once.
Modelling that process accurately enough to be useful has been surprisingly difficult. Existing approaches capture pieces of it โ reaction time here, steering behaviour there โ but no single model has brought all of it together. Until now.
Researchers at Delft University of Technology, working with Waymo, have developed a unified model that integrates perception, decision-making, and physical execution into one coherent framework.
Given the same information a human driver would have, the model can detect when a situation crosses into dangerous territory, anticipate how it's likely to evolve, and predict the most effective avoidance response โ brake, steer, or a combination of both.
The team tested it against real human behaviour in three hazardous scenarios: a lead vehicle braking suddenly, an oncoming car entering the lane without warning, and a driver failing to yield.
The model produced realistic braking reaction times, made similar brake-versus-steer decisions to human drivers, and โ importantly โ incorporated human limitations, so the resulting behaviour looks and feels recognisably human rather than optimally robotic.
Waymo is already using it. The application is direct: comparing how its autonomous vehicles handle emergency situations against how human drivers handle the same situations. That comparison matters enormously for both safety assessment and regulation.
One of the most contested questions in the autonomous vehicle space is whether self-driving systems are genuinely safer than human drivers โ and answering it rigorously requires a credible, scientifically grounded model of human performance as the baseline.
Waymo's chief safety officer put it plainly: the model can help the sector move toward a shared, measurable approach to assessing collision avoidance.
That's not just useful for one company. It's useful for the entire regulatory and safety ecosystem around autonomous vehicles โ manufacturers, legislators, and the public trying to assess whether these systems are ready.
Building safer roads starts with understanding what human driving actually looks like under pressure. This model does that better than anything before it.
#AutonomousVehicles #Engineering