NVIDIA Outlines Robotaxi Platform From AI Training to In-Vehicle Computing

NVIDIA has outlined a modular platform for developing and scaling commercial robotaxis. It covers AI model training, simulation and safety validation, and real-time processing inside the vehicle.

The company says every major robotaxi program currently operating at commercial scale uses its technologies in full or in part. The market is projected in the source to reach $400 billion by 2035, with more than 6 million commercial vehicles in operation.

Three computing components

For model development, NVIDIA offers DGX systems, its Alpamayo portfolio and physical AI datasets. The VLA models are designed to handle complex driving situations by breaking them into steps, evaluating options and selecting a trajectory.

Omniverse NuRec and Cosmos can reconstruct driving scenarios from sensor data and generate variations in traffic, weather, lighting and sensor conditions. Running on RTX PRO servers, the tools support closed-loop simulation and validation, while the AlpaSim framework is intended for evaluating reasoning-based autonomous-driving models.

Hyperion 10 combines sensors

NVIDIA’s level-4-ready Hyperion 10 architecture uses two DRIVE AGX Thor systems-on-chip, 14 cameras, nine radars, three lidars and 12 ultrasonic sensors. Its redundant computing and sensing design is intended to support operation if a component fails.

NVIDIA Halos adds the Halos operating system and a safety-validation framework covering independent inspection, system validation, large-scale simulation and continuous testing from cloud to vehicle.

Companies adopting the platform

Companies using or planning to use NVIDIA technologies include Uber, Lyft, Waymo, Zoox, Pony.ai, WeRide, Wayve, Mercedes-Benz, Nissan, Lucid, Nuro, Stellantis, Hyundai Motor and Kia. Uber plans to scale its Hyperion fleet to 28 cities by 2028 and is working with NVIDIA on a Cosmos-based data factory for rare driving scenarios.

The announced programs span regions and vehicle types. May Mobility is developing its software on DRIVE and has launched a pilot with Lyft in Atlanta; WeRide plans to bring its GXR to Southeast Asian markets through Grab; and TIER IV and Isuzu are deploying level-4 autonomous buses.

Other projects include a Mercedes-Benz S-Class robotaxi ecosystem with NVIDIA and Uber, L4 mobility services from Stellantis, Wayve and Uber, and Tesla’s use of NVIDIA supercomputers to train autonomous-driving neural networks.