Wayve has released new data detailing how its AI rapidly adapts to new driving environments and vehicle platforms with minimal additional data. The release follows the company’s expansion into the US in late October 2024 and the launch of its German testing center last week.
Wayve’s AV2.0 approach is powered by a foundation model – an advanced AI trained on vast amounts of driving data from Wayve’s fleet and partners. Acting as a universal backbone, it learns transferrable driving behaviors, enabling it to adapt to new environments with minimal additional training. In the UK, it has already demonstrated its ability to adapt to different vehicle platforms and cities. Now, new results show that with just 500 hours-worth of incremental US-specific data collected over eight weeks, Wayve’s AI successfully adapted to US road infrastructure, signage, traffic laws, and driving behaviors, approaching UK-equivalent performance in urban and highway driving.
Key research highlights:
- Fast adaptation to US roads: Wayve’s AI quickly learned US-specific driving behaviors, such as road signs, right turns on red, freeway merging and unprotected left turns, with minimal additional training;
- Better performance through diverse data: Training on datasets from the UK, US and Germany enhanced AI performance across all regions more than training on local data alone;
- Performance strengthens as AI expands to new markets: After multi-country testing, early results in Germany showed a 3x performance boost compared with the AI’s initial US testing, demonstrating the benefits of diverse data exposure;
- Quick adaptation to new vehicle platforms: In testing, Wayve’s AI showed an 8x performance improvement when transitioning to a new automotive platform after training on 100 hours-worth of additional vehicle-specific data.
“Auto makers need breakthrough AI technology that enables safe, seamless deployment of advanced assisted and automated driving across markets and vehicle platforms – without costly, time-intensive reengineering,” said Alex Kendall, co-founder and CEO of Wayve. “These latest results highlight how our AV2.0 approach continues to improve with exposure to new environments, accelerating the path to intelligent driving automation at scale. With each expansion, our embodied AI learns faster, gets smarter and offers a scalable solution for automating millions of vehicles worldwide.”
Interested in AI and Wayve? Check out this exclusive feature first published in the January 2025 issue of ADAS & Autonomous Vehicle International, which examines how companies are exploring different ways of applying generative AI to ADAS/AD to make self-driving cars more flexible, reliable and cost-effective.
