Now, in a new research paper published today in Natural Communication, Waymo describes a new computer-based cognitive model that explains how human drivers make split-second decisions to avoid crashes. The company thinks the new model will serve as a benchmark for autonomous driving systems and a way to help move the industry to a greater level of shared safety standards. It’s also this growing peer-reviewed research that Waymo says sets it apart from other autonomous vehicle operators.
Waymo has designed a new model, called ReD for “Reference Driver,” in collaboration with Delft University of Technology in the Netherlands. Much in the same way that the auto industry uses crash test dummies to test a car’s structural integrity and hardware safety, this new model serves as a behavioral dummy to determine how a self-driving car can avoid dangerous situations entirely.
“AV safety testing is multifaceted, and understanding how one handles collisions is an important part of the puzzle,” said Mauricio Peña, Waymo’s chief safety officer. “By establishing this reference model of competent human response, we can help the industry move toward a shared, scientifically based approach to evaluating conflict avoidance behavior.”
RD relies on a neuroscience framework called active inference, inspired by the world’s leading neuroscientists such as professor Karl Friston (who called the RD model a “technological journey” in a statement provided to Waymo). The bottom line is that the human brain is constantly striving to reduce surprise over time.
RED combines several human factors to simulate how a driver handles this stress. People judge longitudinal threats based on “coming,” or how fast an object grows in their field of vision. Waymo’s model replicates this by naturally struggling to judge speed at distances, like a real person. It calculates a “traffic norm” filter that refines its predictions about law-abiding behavior, until it clearly detects a vehicle breaking the traffic norm. It also evaluates surprises like a human driver, triggering a retest of its driving if the surprise reaches a certain threshold that suggests the current plan is failing. The model also accounts for how people use the gas and brake pedals with one foot by introducing a 0.2-second pause when moving between the two.
“By putting our model into practical thinking, we have achieved a complete representation of human collision reactions,” said 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 collision, providing a human-like signal for autonomous driving systems that was previously impossible to do at scale.”
Unlike traditional safety models that only simulate emergencies, Waymo says that ReD is able to “avoid happening” by continuously calculating surprises while minimizing free energy. This allows him to predict accidents in advance and adjust his driving before the situation escalates into a conflict.
Waymo says it is working closely with researchers, regulators, and standards organizations like SAE to find consensus on these reference models. The goal is to move the autonomous vehicle industry toward a shared, scientifically based definition of what constitutes a “careful and competent” human response. To that end, the company is making the RD model open source and publicly available to anyone who wants to test it.
