Waymo has created a new computer model designed to more accurately answer a fundamental question: How does self-driving software stack up against humans?
Alphabet-owned robotaxi, which developed the computer model of human driving abilities in conjunction with TU Delft, published a research paper about it in Nature Communications on Wednesday.
Waymo said it expects the new model to be “more advanced” than the previous version it has used in recent years. The new model was built using a framework called active inference – the theory that a driver is constantly imagining possible futures and taking steps to get to the safest, most predictable one.
Waymo said the new model will help it better understand how people behave in conflict scenarios that its robotaxis faces.
“For decades, the auto industry has used physical and virtual crash dummies to evaluate a car’s safety features, including its material and structural integrity,” Waymo wrote in a blog post Wednesday. The new model, Waymo said, “evolves this concept, serving as a behavioral benchmark for self-driving systems capable of realistically representing reasonable expectations of how an attentive and capable driver responds to traffic collisions.”
A more accurate model of human driving behavior is table stakes for autonomous vehicle companies that need to understand and rate the performance of their robotic behavior in accidents. And it comes at a critical juncture for Waymo, which is scaling into more cities and facing greater scrutiny from regulators and the public.
In January, when a Waymo robotaxi hit a child near a school in Santa Monica, California, the company relied on its previous computer model to claim that a careful driver would have hit it at about 14 miles per hour. The Waymo robotaxi hit the child at just 6 mph after slowing from 17 mph, and the company said he suffered minor injuries. (The crash is still under investigation by the National Highway Traffic Safety Administration and the National Transportation Safety Board.)
The biggest difference between this new model — which Waymo calls the Reference Driver — and its predecessor is that it’s able to replicate the behavior of a human driver before an accident. Previously, Waymo’s models (and others in the industry) focused on replicating “last-second, reactive” human maneuvers, according to the company.
The Reference Driver, meanwhile, can “simulate the internal ‘surprise’ a driver feels during a crash, providing a more humanoid reference point for autonomous driving systems that were previously impossible to automate at scale,” said Arkady Zgonnikov, assistant professor at TU Delft.
Waymo says this new driver model can be adapted to model a “wide range of road user behaviors beyond collision avoidance” and is better equipped to apply to “large test sets with thousands of scenarios.”
“The model can represent and evaluate numerous complex, real-world conflicts in a virtual environment, identifying performance improvements with unprecedented speed and efficiency,” the company wrote.
Waymo wants others to work together to further promote the reference driver program. The company announced Wednesday that it is making the research code for the model available under an academic, noncommercial license that allows its use for research, teaching, personal experimentation and scientific publication.
When you purchase through links in our articles, we may earn a small commission. This does not affect our editorial independence.
