Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Now, in a A new research paper published today in Nature Communications, Waymo describes a new computer-based cognitive model that explains how human drivers make split-second decisions to avoid collisions. The company believes the new model will serve as a benchmark for comparing autonomous driving systems as a way to help move the industry toward a greater degree of common safety standards. It’s also Waymo’s newest A growing body of peer-reviewed research It says it sets it apart from other AV operators.
Waymo designed the new model, which is referred to as ReD (Reference Driver), in collaboration with Delft University of Technology in the Netherlands. Similar to how the auto industry uses crash test dummies to evaluate a car’s structural integrity and hardware integrity, this new model serves as a behavioral dummy to determine how well a self-driving car can completely avoid dangerous situations.
“Assessing spacecraft safety is multifaceted, and understanding how humans handle conflict is an important piece of the puzzle,” says Mauricio Peña, Waymo’s chief safety officer. “By establishing this reference model of competent human response, we can help the industry move toward a common, science-based approach to evaluating collision avoidance behavior.”
ReD is based on a neuroscience framework called active inference, which is supported by world-leading neuroscientists like Professor Karl Friston (who described the ReD model as a “technical tour de force” in a statement provided by Waymo). The basic principle is that human brains constantly strive to reduce surprise over time.
The ReD system synthesizes several human cognitive traits to simulate how a driver copes with this stress. Humans judge longitudinal threats based on “looming danger,” or how quickly an object expands into their field of view. Waymo’s model replicates this by naturally struggling to judge speeds over long distances, just like a real person. It represents a “traffic standards” filter that biases its predictions towards rule-abiding behavior, until it clearly notices a vehicle violating a traffic standard. It evaluates surprises just like a human driver, causing it to re-evaluate its driving once the surprise reaches a certain threshold that indicates the failure of the current plan. The model also explains how humans operate the gas and brake pedals with one foot by introducing a 0.2-second pause when switching between the two.
“By basing our model on active inference, we achieved a comprehensive representation of the human collision response,” Arkady Zhgunnikov, an assistant professor at Delft University of Technology, says in a statement. “This allows us to simulate the internal ‘surprise’ a driver feels during conflict, providing a more human-like standard for autonomous driving systems that was previously impossible to automate at scale.”
Unlike traditional safety models that only simulate emergency situations, Waymo says ReD is capable of “proactive avoidance” by continuously calculating surprise while minimizing free energy. This allows her to anticipate risks early and adjust her motivations before the situation escalates into conflict.
Waymo says it is actively collaborating with researchers, regulators, and standards organizations like SAE to reach consensus on these reference models. The goal is to move the self-driving car industry toward a common, science-based definition of what constitutes an “accurate and competent” human response. To that end, the company is making the ReD model open source and publicly available to anyone who wants to test it.
