Waymo built a virtual driver to study how humans react to surprises on the road
Waymo has built ReD, a virtual human driver model meant to benchmark how autonomous vehicles avoid collisions. The company says it could help the industry compare safety systems with a shared standard.
Intelligence analysis by GPT-5.4 Mini
Waymo says its new Reference Driver model, or ReD, is a behavioral crash test dummy for driving. Built with Delft University of Technology and published in Nature Communications, it tries to model how a competent human reacts to surprises, with the aim of giving autonomous systems a common safety benchmark.
Waymo built a pretend human driver for computers to study. It is like a smart crash-test mannequin for decision-making, helping self-driving cars learn how a careful person might react when something surprising happens on the road.
Analysis
What Waymo built
Waymo says it has created a computer-based cognitive model called ReD, short for “Reference Driver,” to represent how a skilled human driver responds in split-second crash-avoidance situations. The model was developed with Delft University of Technology and described in a paper published in Nature Communications.
How it works
The company frames ReD as a behavioral crash test dummy. Instead of measuring whether a car holds up in a collision, it is designed to measure how well an autonomous driving system can avoid dangerous situations in the first place. Waymo says the model uses a neuroscience framework called active inference, which treats human behavior as a process of reducing surprise over time.
The model combines several human-like traits. It estimates threats through “looming,” or how quickly an object appears to grow in the driver’s view. It includes a “traffic norm” filter that assumes other vehicles are behaving normally until evidence shows otherwise. It also includes a short pause of 0.2 seconds when switching between gas and brake pedals, reflecting the way humans use one foot for both.
Why Waymo says it matters
Waymo says ReD can do more than simulate emergencies after they happen. It can keep calculating surprise and adjust early, which the company describes as “proactive avoidance.” Waymo says the goal is to help the AV industry define what a careful and competent human response looks like in a scientifically grounded way.
The company also says it is working with researchers, regulators, and standards groups such as SAE to build consensus around these reference models. It plans to make ReD open source and publicly available.
Key points
- Waymo introduced ReD, a virtual driver model meant to represent how humans avoid crashes.
- The model was developed with Delft University of Technology and described in Nature Communications.
- Waymo says ReD uses active inference and human-like assumptions about surprise, threat, and traffic norms.
- The company wants ReD to become a shared benchmark for autonomous driving safety.
- Waymo says it will make the model open source and is discussing it with researchers and standards groups.
If ReD becomes widely used, autonomous vehicle makers could compare safety systems using the same yardstick. That could make it easier for researchers and regulators to judge whether a car handles surprise situations as well as a careful human.
The model may be useful only if others agree to use it, which is not guaranteed. If the industry does not align on one standard, ReD could remain another research tool rather than a common safety benchmark.
