The company Waymo has announced a world model based on Google Genie 3 for training its autopilot systems. This system, named Waymo World Model, aims to improve autonomous driving algorithms without risking real cars and people, creating an infinite simulation environment.
The editorial reports, citing its source, that Waymo World Model allows engineers to change weather conditions, time of day, or create new road situations from scratch through text prompts. The system converts ordinary dashcam footage into an interactive 3D environment, allowing for real-time editing.
The main goal of the project is to test scenarios referred to as "rare events." These are situations that occur very rarely in real life but can seriously disrupt the logic of robot systems. Waymo cites unusual situations such as an elephant emerging onto a highway, a tornado chase, or snow falling in a tropical city as examples. The company states that it is not possible to safely test such situations in a live city environment, but it is important for systems to be prepared for these scenarios.
The topic has also become relevant in the context of a recent incident in California where a robotaxi touched a child. Currently, operators assist cars in complex situations, but they do not directly control the vehicle, only provide directional guidance. The company plans to increase the level of real autonomy by minimizing the human factor.
Interest in world models is rapidly growing. Investors and prominent figures in artificial intelligence, including Yann LeCun and Fei-Fei Li, consider this approach to be the next important stage after large language models. The latest pre-release of Project Genie has already inspired dozens of developers to conduct experiments aimed at deepening AI's understanding of physics and space.