Can You Just Tell Your Car What to Do? Now You Can

NVIDIA's Alpamayo 1.5 adds language-guided trajectory planning and flexible camera support for Level 4 AV developers

20 Mar 2026

The most persistent problem in autonomous vehicles has never been speed or sensor range. It has been explanation. When an AI-piloted car decides to brake, swerve, or hold its lane, neither the engineer nor the regulator has always known quite why. NVIDIA's latest release tries to change that.

On March 16th, at its annual GTC conference, the company launched Alpamayo 1.5, an upgrade to its open-source AI reasoning platform for self-driving systems. The headline feature is text-guided trajectory planning: engineers can issue natural-language instructions such as "turn left in 200 meters," and the model generates a corresponding path while narrating its logic at each step. The result is AI decision-making that is, for once, legible.

The practical implications are considerable. Regulators in most jurisdictions require some form of explainability before certifying autonomous systems for public roads. By embedding interpretability into the model itself, NVIDIA removes one of the more tiresome obstacles between the laboratory and the street.

The release also adds flexible multi-camera support, letting the model run across different vehicle types without reconfiguring sensor arrays. A new open dataset accompanies the launch: over 1,700 hours of multi-sensor driving footage drawn from 25 countries, together with an autolabeling pipeline that helps teams generate annotated training data at scale.

Traction for the platform has been swift. Since the original Alpamayo debuted at CES in January, more than 100,000 researchers and developers have downloaded it in under two months. The model is not designed to run directly inside vehicles; instead, teams use it as a teacher model, distilling its outputs into leaner, edge-deployable systems suited to production hardware. That architecture matters: it lets smaller automakers and startups borrow sophisticated reasoning infrastructure they could not otherwise afford to build.

The same day brought further news. BYD, Geely, Isuzu, and Nissan are now developing Level 4-capable vehicles on NVIDIA's DRIVE Hyperion platform. Uber, meanwhile, plans to deploy an NVIDIA-powered robotaxi fleet across 28 cities and four continents by 2028.

The autonomous vehicle industry has spent years promising transformation while delivering incremental progress. The machinery of industrialized, software-defined deployment is at last beginning to look real.