Waymo has built a virtual human driver called ReD to benchmark its robotaxis. The new system models how real drivers stay safe, allowing the company to test autonomous cars against a realistic reference. By treating ReD as a “behavioral crash dummy,” Waymo seeks to sharpen collision‑avoidance metrics for safer urban streets and fewer accidents daily.

ReD blends neuroscience and driving data, using the active inference framework that assumes drivers minimize surprise. It simulates how a careful driver updates beliefs as situations shift, manages uncertainty about other vehicles, and selects evasive actions like braking or swerving. The model also factors in traffic norms and a brief pause between gas and brake.

Waymo forged the ReD model with Delft University of Technology and published findings in Nature. The partnership melds Dutch research expertise with Waymo’s real‑world data. By releasing ReD under an academic license, the company invites researchers, safety bodies, and regulators to refine a shared standard for evaluating collision‑avoidance behavior that could reshape safety protocols for autonomous vehicles worldwide.

Industry observers see ReD as a tool to standardize safety tests across autonomous platforms. By comparing robotaxi decisions to a human benchmark, Waymo can pinpoint weaknesses and accelerate deployment. The open‑source approach invites broader scrutiny, potentially lowering barriers for smaller firms to adopt proven human‑like safety logic in their own vehicles and improve public confidence.