Waymo leverages Genie 3 to create a world model for self-driving cars (arstechnica.com)

Waymo, the self-driving car subsidiary of Google, is leveraging the advanced capabilities of Google DeepMind's Genie 3 to enhance its world model for autonomous vehicle training. With over 200 million miles of real-world driving experience, Waymo is now able to simulate "hyper-realistic" environments, allowing its AI to encounter and learn from rare or non-existent driving scenarios, such as traversing a snow-covered Golden Gate Bridge. This innovative approach addresses a significant challenge in the autonomous driving sector: the limited representation of dangerous, infrequent events in the training datasets. Genie 3 distinguishes itself with its long-horizon memory, enabling the world model to retain contextual details over extended periods, unlike previous iterations that quickly lost track of relevant information. By producing realistic simulations based on straightforward prompts and driving inputs, Waymo aims to train its AI more effectively. Although Genie 3 primarily generates video rather than true 3D environments, its speed and memory capabilities make it a promising tool for developing the complex decision-making required in self-driving technology. This advancement could signal a pivotal shift in how autonomous vehicles are trained, ultimately improving their safety and efficiency on the roads.