Waymo creates new model to compare robotaxis with humans

Alphabet-owned Waymo has developed a new computer model that allows for a more accurate comparison of its autonomous driving software with human capabilities. The research on this model, created in collaboration with TU Delft, was published in the journal Nature Communications. The company expects this new system to be significantly more accurate than previous versions used in recent years. This is reported by Techcrunch.com reports.

The new model is built on a framework called "active inference." According to this theory, a driver constantly imagines potential future scenarios and acts to achieve the safest, most predictable outcome. With this model, Waymo aims to better understand how humans behave in collision scenarios that robotaxis might encounter.

Waymo representatives state that for decades, the automotive industry has used physical and virtual mannequins to evaluate safety features. The new model advances this concept, serving as a behavioral benchmark that realistically reflects how a cautious and skilled human driver reacts to conflict situations on the road.

The main difference between this new model, called Reference Driver, and its predecessor is its ability to reconstruct human behavior in the pre-collision process. While previous models focused only on reactive maneuvers in the final seconds, the new system can also simulate the driver's internal "surprise" in unexpected situations.

Waymo is currently expanding its services to more cities and is under strict scrutiny by regulatory bodies. The new model will be used not only to prevent collisions but also to analyze a wide range of road user behaviors across thousands of different scenarios.