Waymo's New AI Simulator Aims to Teach Cars in a Virtual World

Waymo, the self-driving company under Google's parent Alphabet, has introduced a new artificial intelligence system designed to train its autonomous vehicles. Called the Waymo World Model, the technology creates highly detailed, virtual driving environments to test the cars in ways that would be difficult, dangerous, or nearly impossible on real streets.

The system moves beyond simply replaying past driving data. Instead, it generates entirely new, realistic scenarios—like a child running into the road during a storm—and can produce thousands of subtle variations. This allows engineers to rigorously test the vehicle's software against rare but critical situations without ever leaving a data center.

Waymo's model is trained on a vast library of information gathered from over 50 million miles of actual driving in cities like San Francisco and Phoenix. This foundation enables the simulator to create authentic-looking streets, predict how pedestrians and other drivers might behave, and let the virtual Waymo vehicle interact with the environment in real time.

For the broader industry, this advancement highlights a significant shift. Simulation is becoming central to proving the safety of self-driving technology. Waymo's extensive real-world operations give it a unique data advantage to build such a system, potentially setting a new benchmark for how autonomous vehicles are developed and validated for public roads.

Source: Webpronews