Waymo Reference – Waymo says it has built a new computer model—called the Reference Driver—to compare its robotaxi driving behavior with humans more accurately, including what drivers feel and do leading up to crash scenarios. The company credits the model’s framework and a col
A robotaxi isn’t judged only by what happens in the final seconds before impact. For Waymo. the harder question has always been what should come before—how a careful human would react while the situation is still unfolding. Now the Alphabet-owned company is trying to measure that moment more precisely.
Waymo says it has created a new computer model designed to answer a fundamental benchmark question: how does its autonomous driving software stack up against humans?. The company developed the model of human driving capabilities alongside TU Delft. and published a research paper about the work in Nature Communications on Wednesday.
In a blog post on Wednesday. Waymo said it expects the new model to be more accurate than the previous version it has used over the past several years. The new model was built using a framework called active inference—an approach based on the idea that a driver is constantly imagining possible futures and taking actions to reach the safest. most predictable one.
Waymo’s stated goal isn’t just performance scoring in calm traffic. The company says the model will help it better understand how humans behave in crash scenarios that its robotaxis encounter. “For decades. the automotive industry has used physical and virtual crash dummies to evaluate a car’s safety features. including its hardware and structural integrity. ” Waymo wrote. It said the new model “evolves this concept. ” functioning as “a behavioral benchmark for autonomous driving systems able to realistically represent reasonable expectations on how a careful and competent human driver responds to traffic conflicts.”.
The move lands at a tense point for Waymo as it scales to more cities and faces increasing scrutiny from regulators and the public. In January, a Waymo robotaxi struck a child near a school in Santa Monica, California. Waymo said it relied on its previous computer model. which it used to estimate that an attentive human driver would have made impact at around 14 miles per hour. The robotaxi hit the child at just 6 miles per hour. after decelerating from 17 miles per hour. and Waymo said the child sustained minor injuries.
The crash is still under investigation by the National Highway Traffic Safety Administration and the National Transportation Safety Board.
Waymo’s new model—called the Reference Driver—marks a shift in what it tries to recreate. The company says the biggest difference is that the Reference Driver can reproduce a human driver’s behavior in the run-up to a crash. Previously, Waymo’s models (and other industry models) focused on replicating “last-second, reactive” human maneuvers.
This time, Waymo says the model can simulate the internal “surprise” a driver feels during a conflict. Arkady Zgonnikov. an assistant professor at TU Delft. said in a statement that this provides “a more human-like benchmark for autonomous driving systems that was previously impossible to automate at scale.”.
Waymo also argues that this changes how broadly the model can be used. It says the Reference Driver can be adapted to model a “wide range of road user behaviors beyond collision avoidance.” The company adds that it is better equipped for “large test sets with thousands of scenarios. ” and that the model can represent and evaluate “numerous complex. real-world crashes in a virtual environment. ” finding performance improvements with “unprecedented speed and efficiency.”.
To keep the momentum going, Waymo says it wants other researchers to collaborate on pushing the Reference Driver further. The company said it is making the research code available under an academic, non-commercial license. That license, Waymo said, allows use for research, teaching, personal experimentation, and scientific publication.
Waymo TU Delft Reference Driver active inference robotaxi autonomous driving crash benchmarking Nature Communications Arkady Zgonnikov NHTSA NTSB computer model
So they’re basically making a simulation of a human crashing? cool I guess.
Waymo always says it’s “more accurate,” but I don’t trust any of it. If they can predict what people do, why are there still near misses? Sounds like marketing.
Active inference?? That sounds like AI is just daydreaming about accidents and then pretending it knows what humans would do. Also Nature Communications? Doesn’t that mean like a lot of scientists will disagree? Either way, no one asked for a computer to judge human driving.
I saw the headline and thought they were changing the actual Waymo car to feel what drivers feel before a crash. But it’s just a model. Still, I’m sorry, the “reference driver” will never match real people texting and doing stupid stuff. Like how do you model panic? Half the time humans slam the brakes way too late anyway.