Waymo has created a new computer model designed to answer a fundamental question about autonomous vehicles: how does its self-driving software stack up against human drivers? The Alphabet-owned robotaxi company, working with researchers at TU Delft, published a research paper about the model in Nature Communications on June 10, 2026. The new model, called the Reference Driver, uses a framework called active inference, which assumes that drivers constantly imagine possible futures and take actions to reach the safest, most predictable outcome.

What Makes This New Model Different From Previous Versions?

The biggest difference between the Reference Driver and Waymo's previous models is that it can reproduce human driving behavior in the moments leading up to a crash, not just the last-second reactions. Earlier models focused on replicating what researchers call "last-second, reactive" human maneuvers, but the new version simulates the internal "surprise" a driver feels during a traffic conflict.

This distinction matters because it allows Waymo to evaluate its robotaxis against a more realistic benchmark of human behavior. The company explained that the Reference Driver "can represent and evaluate numerous complex, real-world crashes in a virtual environment, identifying performance improvements with unprecedented speed and efficiency".

The timing of this release is significant. In January 2026, a Waymo robotaxi struck a child near a school in Santa Monica, California. The company used its previous computer model to argue that an attentive human driver would have made impact at around 14 miles per hour. The Waymo vehicle hit the child at just 6 miles per hour after decelerating from 17 miles per hour, and the company said she sustained minor injuries. That crash is still under investigation by the National Highway Traffic Safety Administration and the National Transportation Safety Board.

How Does This Fit Into Waymo's Broader Strategy?

Waymo is currently delivering around 500,000 driverless paid rides every single week, according to technology consultant Dr. Mario Herger, who studies autonomous driving trends in Silicon Valley. This scale of real-world operations means the company needs increasingly sophisticated tools to understand and improve its safety performance.

The Reference Driver model comes at a critical moment for Waymo. The company is scaling to more cities and facing greater scrutiny from regulators and the public. A more accurate model of human driving behavior is essential for autonomous vehicle companies that need to understand and grade the performance of their robotaxis in crashes.

Steps to Understanding Waymo's New Safety Benchmark

• Active Inference Framework: The model uses active inference theory, which assumes drivers constantly imagine possible futures and choose actions to reach the safest outcome, rather than simply reacting to immediate threats.
• Pre-Crash Behavior Simulation: Unlike previous models that only replicated last-second reactions, the Reference Driver can simulate how humans behave in the moments before a crash occurs, including the feeling of surprise during traffic conflicts.
• Large-Scale Testing Capability: The model can be applied to large test sets with thousands of scenarios, allowing Waymo to identify performance improvements more quickly than was previously possible.
• Behavioral Adaptation: The Reference Driver can be adapted to model a wide range of road user behaviors beyond collision avoidance, making it applicable to diverse driving situations.

Waymo is making the research code for the Reference Driver available under an academic, non-commercial license. This allows other researchers to use it for research, teaching, personal experimentation, and scientific publication, potentially accelerating improvements across the autonomous vehicle industry.

What Does This Mean for the Future of Robotaxis?

The development of more accurate human behavior models reflects a broader shift in how the autonomous vehicle industry thinks about safety. For decades, the automotive industry used physical and virtual crash dummies to evaluate a car's safety features. Waymo's new model "evolves this concept, serving 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".

According to Dr. Herger, the real question facing autonomous vehicle companies is no longer whether the technology works, but how quickly it scales and how rapidly autonomous systems take over a relevant share of driven miles. He expects that by 2035, the majority of miles driven in Europe, China, and North America could be done by autonomous vehicles.

Dr. Herger

"L3 is the slightly better candle; L4 is the light bulb. We are no longer discussing only feasibility, but scaling. In 2035, the majority of driven miles could come from autonomous systems," said Dr. Mario Herger.

Dr. Mario Herger, Author and Consultant, Technology Trend Researcher

The Reference Driver model represents one piece of that scaling puzzle. By providing a more accurate way to evaluate robotaxi performance against human drivers, Waymo is building the tools it needs to demonstrate safety at scale. This is particularly important as the company expands to new cities and as regulators worldwide develop frameworks for autonomous vehicle deployment.