NVIDIA Teases Bold Step Forward in Autonomous Driving with Complete DRIVE AV Stack and Alpamayo Models

At CES 2026, NVIDIA announced a significant breakthrough: they created self-driving vehicles capable of navigating extremely difficult roads with little to no human intervention. It’s all centered around their DRIVE platform, which is currently powering real-world automobiles and laying the groundwork for substantial automation.

Mercedes-Benz’s all-new CLA-class car leads the way in this innovation. That is the first car to use NVIDIA’s complete DRIVE AV software stack. The system allows the car to go point-to-point in congested metropolitan streets, from any starting point to any destination. With this, you can basically hand over control to the car; it manages all the tough things like lane changes, turns, and interactions with pedestrians and cyclists in the city, all without the need for expensive LiDAR or pre-mapped roads.

They also staged a live demonstration in San Francisco in which the CLA effortlessly overtook double-parked automobiles, gave way to reversing cars, and maneuvered unprotected left turns while avoiding obstructions, all with only 10 cameras, 5 radars, and 12 ultrasonic sensors, not to mention the two stated above. Production versions of this will be on U.S. roads by late 2026, with the promise of even more enhancements and the ability to simply download and install any new features via the air.

Underneath all of this is a two-part software system that is essential to how everything works. One part uses traditional ways to recognize objects, forecast movements, and create real maps in real time based on what it sees in the cameras and navigation data. The other component is a full end-to-end neural network trained on millions of hours of human driving footage, which provides the car with extremely smooth and natural-looking trajectories. It then selects the two safest solutions, and if the driver is need to intervene, the process will be smooth and efficient. If the driver does not interfere with it, the software remains available and ready to assist without any abrupt changes.

NVIDIA’s Alpamayo model family also includes a slew of new tools for the industry as a whole. Alpamayo 1 is a 10 billion parameter model that can be run on any platform, including Hugging Face; it integrates sensor data and generates driving pathways while also explaining step by step how it arrived at its conclusions. This allows the car to handle odd scenarios such as a traffic light that fails at a busy intersection. It’s also packed with datasets, including over 1700 hours of driving movies, as well as an open simulation framework called AlpaSim, which allows you to test various situations at scale. All of this means that firms like as Jaguar Land Rover, Lucid, and research groups can take this model and fine-tune it using their own data to accelerate progress toward level 4 automation, in which automobiles can conduct their own business in any area that they are instructed to.

As the hardware side improves, NVIDIA welcomes several more partners into the DRIVE Hyperion ecosystem. Sensor manufacturers like as Aeva include their 4D lidar, which can measure the speed of every point in the area it is looking at, resulting in significantly superior long-range detection in all types of weather, as a reference option that will be ready for production in 2028. Suppliers such as Bosch, ZF, and Sony have qualified all of the necessary components so that they can be immediately connected to the platform. Dual Thor processors provide a tremendous amount of power for all real-time decision making across the entire 360-degrees of the vehicle.