Waymo created a driver to learn how people react to surprises on the road

Now, in a A new research paper published today in Nature Communications, Waymo describes a new computer-based information system that explains how human drivers make distributed decisions to avoid accidents. The company envisions the new model as a benchmark for self-driving systems as a way to help move the industry toward greater standards of safety sharing. It’s also the latest in Waymo’s lineup the number of peer-reviewed studies which it says sets it apart from other independent car operators.

Waymo developed a new prototype, called ReD for “Reference Driver,” in collaboration with Delft University of Technology in the Netherlands. Just as the car industry uses crash test dummies to assess the vehicle’s condition and hardware safety, this new model serves as a dummy to determine how well an autonomous vehicle can avoid all accidents.

“Evaluating AV security is multifaceted, and understanding how a person reacts to conflicts is very difficult,” said Mauricio Peña, chief security officer at Waymo. “By establishing this model of how people can respond, we can help companies move toward a shared, scientifically based approach to collision avoidance.”

ReD relies on a neuroscience system called active inference, led by world-leading experts such as professor Karl Friston (who called the ReD model a “technical tour de force” in a statement provided by Waymo). The bottom line is that the human brain is always trying to reduce surprise over time.

ReD puts together a number of personal data to compare a driver’s reaction to this stress. People judge long exposures based on “coming,” or how large an object is in their field of vision. Waymo’s model also explains this by struggling to judge distance, like a real person. It calculates a “traffic filter” that biases its predictions towards law-abiding traffic, until it correctly detects a vehicle in violation of traffic laws. And it looks for surprises like a human driver, triggering a self-driving review once it reaches a critical point that indicates the current system is failing. This model also describes how people use the gas and brake pedals with one foot by triggering a 0.2 second pause while switching between the two.

“By basing our model on the concept of sustainability, we have achieved a complete response to human impact,” said Arkady Zgonnikov, an assistant professor at Delft University of Technology, in a statement. “This allows us to capture the internal ‘surprise’ that a driver feels during a collision, providing a human-like signal to self-driving systems that was previously impossible to use on a large scale.”

Unlike safety devices that are limited to emergency situations, Waymo says ReD can “prevent quickly” by continuously calculating and reducing free energy. This allows him to quickly anticipate what is going to happen and adjust his driving before things start to escalate.

Waymo says it is actively working with researchers, regulators, and standards organizations such as SAE to establish agreements regarding these models. The goal is to move the independent industry to define a shared, scientifically based definition of what constitutes a “smart and creative” human response. To that end, the company is making the ReD version open source and publicly available to anyone who wants to try it.