NVIDIA’s DriveAV software platform now enables full address-to-address autonomous driving across diverse environments, including dense urban traffic and complex highway interchanges. Built on the Hyperion architecture, the system achieves cost-effective scalability for consumer vehicles by utilizing a sensor suite of cameras, radar, and ultrasonics while specifically omitting expensive LiDAR and HD maps. The technology employs a dual-stack approach, combining a classical perception-based stack with an end-to-end neural network that learns human-like behaviors from vast datasets. To ensure safety at scale, NVIDIA utilizes Omniverse digital twins and Neural Reconstruction (NUREC) to perform over one million daily simulated test replays, allowing the software to navigate "edge case" scenarios like double-parked cars or unprotected left turns. A standout feature is the collaborative driving mode, which allows seamless human intervention and takeover without disengaging the autonomous system, currently deploying in vehicles from Mercedes-Benz, Lucid, and Jaguar Land Rover.
Sign in to continue reading, translating and more.
Continue
