Waymo Develops Virtual Driver Model for Autonomous Vehicle Safety Testing

Driverless vehicle company Waymo has developed a virtual model to simulate human driver reactions in critical situations, aiming to improve the safety of autonomous driving technology. This new approach is designed to capture the entire sequence of events in collision avoidance, from hazard perception to maneuver execution, which current testing methods often do not fully address.

In collaboration with scientists from Delft University of Technology, Waymo created a cognitive model named ReD (Reference Driver). This model is based on a neuroscience framework known as active inference, which simulates how the human brain anticipates and avoids unexpected events. The ReD model functions as a 'behavioral crash-test dummy,' designed to assess an autonomous vehicle's ability to avoid collisions entirely, rather than just evaluating passenger safety during an impact.

The ReD model operates exclusively within a computer simulation. To validate its accuracy, researchers compared the virtual driving behavior of ReD against real-world data from human drivers. The model was tested in simulated collision scenarios, successfully replicating human responses in three distinct situations: a sudden braking of a lead vehicle, an oncoming vehicle unexpectedly changing lanes, and another vehicle failing to yield at an intersection.

The study's authors stated that their findings demonstrate the potential of active inference as a framework for understanding and modeling human behavior in complex driving tasks. The ultimate goal is to utilize ReD as a digital benchmark for testing autonomous vehicles in virtual environments. While the model has shown success in replicating human reactions in these specific scenarios, future work will focus on expanding its capabilities to more complex driving environments and a wider range of real-world traffic situations.