NVIDIA details its robotaxi stack: DRIVE Hyperion 10 and Alpamayo
NVIDIA says every major robotaxi program running at commercial scale today uses its modular stack for training, simulation or in-vehicle computing.
NVIDIA has laid out the platform it offers for robotaxi development, saying every major robotaxi program operating at commercial scale today runs on its modular stack, covering AI training, simulation, in-vehicle computing or a combination of the three.
The company cites projections of a $400 billion global robotaxi market by 2035, with more than 6 million commercial vehicles in operation.
A three-computer approach
The platform splits into a model training computer, a simulation and validation computer, and an in-vehicle computer. Driving models can be trained on NVIDIA DGX systems. The Alpamayo portfolio bundles open reasoning vision language action (VLA) models, simulation frameworks and physical AI datasets; its reasoning models break complex driving situations into smaller steps and select the safest trajectory. NVIDIA also supplies physical AI datasets, reinforcement learning blueprints and recipes for post-training and distillation.
For simulation, Omniverse NuRec reconstructs real driving scenes from sensor data, while Cosmos world foundation models generate physically based variations, turning thousands of real corner cases into millions of combinations of traffic, weather, lighting and sensor conditions. Omniverse and Cosmos run on NVIDIA RTX PRO Servers for closed-loop simulation and validation, and the AlpaSim framework extends the workflow to training and evaluating reasoning-based driving models.
Hardware in the car
DRIVE Hyperion is the reference architecture for level-4-ready robotaxis. Hyperion 10 pairs dual DRIVE AGX Thor systems-on-a-chip, built on the Blackwell platform, with 14 high-definition cameras, nine radars, three lidars and 12 ultrasonics for 360-degree sensor fusion. Redundant compute and sensing support fail-operational driving when a sensor or compute component fails. On the safety side, Halos OS comes with a validation and certification framework spanning independent inspection, system validation, large-scale simulation and continuous testing from cloud to car.
NVIDIA says its robotaxi ecosystem covers Asia, Europe, the Middle East and North America. Full details are on the NVIDIA blog.
