Waymo Develops Virtual Driver to Test Reactions to Road Surprises

Alphabet-owned Waymo has introduced a new cognitive model, ReD (Reference Driver), which simulates human driver behavior in unexpected road scenarios. Published in Nature Communications, the research details how the model helps evaluate the safety of autonomous transport systems by comparing their reactions to human responses.

ReD is grounded in the principle of active inference—a neuroscience concept where the brain seeks to minimize surprises. The model incorporates human perception traits, such as assessing the speed of approaching objects and reacting to traffic rule violations. It also replicates the 0.2-second delay in switching between the gas and brake pedals, a characteristic of real drivers.

According to Waymo’s Chief Safety Officer Mauricio Peña, the ReD model will serve as a critical tool for establishing unified safety standards in the autonomous vehicle industry. The company collaborates with regulators, researchers, and organizations like SAE to make this technology publicly accessible. The model’s code has already been released open-source for testing.

ReD’s development aligns with Waymo’s strategy to create scientifically validated approaches to autonomous system safety. Previously, the company built virtual environments for testing vehicles in extreme conditions, now emphasizing human behavior simulation behind the wheel.