Scalable Autonomy Through Embodied AI - Wayve's AV2.0 Approach to Autonomous Driving

Explore Wayve's revolutionary approach to autonomous driving through embodied AI in this 26-minute conference talk. Discover how Wayve is training models to drive virtually any car, with any sensors, in any city, rather than focusing on a single vehicle or manufacturer. Learn about the company's proprietary AV2.0 model and understand the unique challenges of implementing AI in the imperfect physical world. Examine Wayve's strategy of working with diverse cars and sensors instead of developing their own hardware, and delve into the proprietary foundation model that powers AV2.0. Understand how the company leverages simulated data for learning, including their concept of "bad dreams," and grasp the massive scale and complexity of the data and models required for scalable autonomous driving solutions.

Syllabus

00:00 - Introduction and why Wayve’s approach is different
02:41 - Wayve’s AV2.0 and the difficulty of AI in the imperfect physical world
09:02 - How and why Wayve works with myriad cars and sensors vs. just their own
14:17 - The proprietary foundation model at the core of AV2.0
19:53 - Learning with simulated data a.k.a. “bad dreams”
22:10 - The scale and challenge of all the required data and models