IntelligentHQ
IntelligentHQ
Skip to content

Autonomous Vehicles & Transportation

NVIDIA's automotive stack: DRIVE AGX in-vehicle computers, DriveOS, the Hyperion reference architecture, the Halos safety framework, Alpamayo open driving models, and Cosmos and Omniverse for simulation and validation.

Technology profiles for this category are in research.

Overview

Automated driving needs three kinds of computing: training in the data center, large-scale simulation to test rare situations, and a safety-certified computer in the car. NVIDIA describes its offer as a three-computer approach along those lines.

DRIVE AGX, with Thor and Orin options, is the in-vehicle platform, and DriveOS is its safety-certified operating system. Hyperion is a reference architecture with a validated sensor suite for Level 2++ to Level 4 programs. Halos is NVIDIA's safety framework across vehicle architecture, chips, software and models. Alpamayo is an open family of reasoning vision-language-action models, frameworks and datasets. For validation, Cosmos generates rare driving scenarios, Omniverse NuRec reconstructs recorded drives and AlpaSim is an open-source closed-loop simulator. Training runs on DGX.

The audience is automakers, suppliers, robotaxi and trucking developers, and research groups.

Traffic analytics for city operators, such as counting vehicles at junctions, belongs with vision AI rather than this stack.12

Problems it addresses

  • Rare and dangerous situations1

    Road testing cannot cover every edge case. Cosmos generates long-tail driving scenarios for testing.

  • Replaying real drives1

    Omniverse NuRec reconstructs recorded drives so they can be simulated again with changes.

  • Safety assurance1

    Halos covers vehicle architecture, chips, software and AI models.

  • Real-time computing in the car1

    DRIVE AGX runs the driving software stack in real time.

A typical workflow

  1. Collect and curate data

    Gather fleet data and select the scenarios that matter for training and testing.

  2. Train models1

    Train perception and driving models on DGX, or start from Alpamayo.

  3. Simulate and validate2

    Run closed-loop tests with AlpaSim, NuRec reconstructions and Cosmos scenarios.

  4. Integrate in the vehicle2

    Deploy on DRIVE AGX with DriveOS, using Hyperion as the reference architecture.

  5. Build the safety case1

    Document safety evidence across hardware, software and models, using Halos as the framework.

NVIDIA Solution Architect

Describe your project and get an explainable architecture.

Open the lab

Next steps

  1. Get a DRIVE AGX developer kit for production-equivalent hardware.

  2. Try the open-source AlpaSim simulator with a public driving dataset.

  3. Review Alpamayo model cards and license terms before using them in a product.

  4. Use the NVIDIA Solution Architect to map the training, simulation and in-car parts of your program.

Sources

  1. NVIDIA Autonomous Vehicles (opens in a new tab)NVIDIA · Vendor-reported
  2. NVIDIA DRIVE for developers (opens in a new tab)NVIDIA · Vendor-reported

Fill out the form below to request your copy.

Name(Required)