Waymo’s Virtual Driver Studies How Humans React to Road Surprises

Waymo has spent years developing virtual systems to help its self-driving cars make sense of the real world. The company has built realistic 3D environments to prepare for natural disasters and rare edge cases. It also created a digital version of an ultra-alert driver to compete against its own autonomous vehicles in simulated crash scenarios. Now, with a new research paper published today in Nature Communications, Waymo is unveiling a computer-based cognitive model that explains how human drivers make split-second decisions to avoid accidents. The company believes this model can serve as a benchmark for comparing autonomous driving systems, helping push the industry toward shared safety standards. It also adds to Waymo’s growing portfolio of peer-reviewed research, a factor the company says sets it apart from other autonomous vehicle operators.

Developed in partnership with the Delft University of Technology in the Netherlands, the model is called ReD, short for “Reference Driver.” Think of it as a behavioral crash test dummy. Just as the auto industry uses physical dummies to test a car’s structural safety, ReD evaluates how well an autonomous vehicle can avoid dangerous situations entirely. “Evaluating AV safety is multifaceted, and understanding how a human handles conflict is a critical piece of the puzzle,” says Mauricio Peña, Waymo’s chief safety officer. “By establishing this reference model of a competent human response, we can help the industry move toward a shared, scientifically grounded approach for evaluating collision-avoidance behavior.”

ReD is built on a neuroscience framework called active inference, championed by leading neuroscientists like professor Karl Friston, who called the model a “technical tour de force” in a statement provided by Waymo. The core idea is that human brains constantly work to minimize surprise over time. ReD layers several human cognitive traits to simulate how a driver handles stress. For instance, humans judge longitudinal threats based on “looming,” or how quickly an object grows in their field of vision. Waymo’s model replicates this by naturally struggling to gauge speed at far distances, just like a real person. It also includes a traffic norm filter that biases predictions toward rule-following behavior, until a vehicle is observed breaking a rule. And it evaluates surprise the same way a human driver does, triggering a reevaluation once a surprise hits a certain threshold that suggests the current plan is failing. The model even accounts for how humans operate gas and brake pedals with a single foot by introducing a 0.2-second pause when switching between the two.

“By grounding our model in active inference, we’ve achieved a holistic representation of human collision response,” says Arkady Zgonnikov, an assistant professor at Delft University of Technology, in a statement. “This allows us to simulate the internal ‘surprise’ a driver feels during a conflict, providing a more human-like benchmark for autonomous driving systems that was previously impossible to automate at scale.”

Unlike traditional safety models that only simulate emergencies, Waymo says ReD is capable of proactive avoidance. It continuously calculates surprise while minimizing free energy, allowing it to anticipate risks early and adjust driving before a situation escalates into a conflict. Waymo is actively collaborating with researchers, regulators, and standards organizations like the SAE to build consensus around these reference models. The goal is to move the autonomous vehicle industry toward a shared, scientifically grounded definition of what it means to be a “careful and competent” human driver. To support that effort, Waymo is making ReD open source and publicly available for anyone to test.