The NVIDIA DRIVE™ AGX Developer Kit provides the hardware, software, and sample applications needed for development of production level autonomous vehicles (AV). NVIDIA DRIVE AGX System is suitable for any autonomous vehicle ecosystem developer requiring an autograde AI computing platform that meets safety standards (such as ISO 26262/ASIL-D, ISO/PAS 21448), is compliant with key regulatory requirements (such as NHTSA, Euro NCAP), has an open software framework and a large ecosystem of automotive partners (including auto grade sensor vendors, automotive Tier 1 suppliers) to choose from.



Details

DRIVE AGX Hardware Options

  • DRIVE AGX Xavier™ Developer Kit (SKU 2000): Includes two Xavier Systems-on-a-Chip (SoCs) for developing SAE Level 2/3 autonomous driving applications
  • DRIVE AGX Pegasus™ Developer Kit (SKU 2200): Includes two Xavier SoCs and two NVIDIA Turing™ GPUs for developing robotaxi applications

DRIVE AGX Xavier™ Developer Kit (SKU 2000):
Component Description Details
Two Xavier SoCs 8-core “Carmel” CPUs based on ARM v8 ISA
Deep Learning Accelerators (DLA) 5 TOPS (FP16) | 10 TOPS (INT8)
NVIDIA Volta™-class integrated GPU 20 TOPS (INT8) | 1.3 TFLOPS (FP32)
Programmable Vision Accelerators (PVA) 1.6 TOPS
Image Signal Processor (ISP) 1.5 Gigapixels/s
Video encoder 1.2 GPix/s
Video decoder 1.8 GPix/s
DRIVE AGX System I/O Camera 90 Gb/s over 16x GMSL(R) ports
LIDAR/Radar ~50 Gb/s over Ethernet
Vehicle IO 16 CAN interfaces
Memory Bandwidth Xavier >250 GB/s
Included Accessories Camera 2 Megapixel RCCB Camera with 60-degree FOV
Vehicle harness Connects to vehicle IO ports
Other cables Connects the DRIVE AGX system to a host development computer
DRIVE Software DRIVE OS
Preflashed Linux Ubuntu OS
(Note: QNX option will be available soon)
NVIDIA CUDA® Libraries, including TensorRT® (for access to deep learning hardware accelerator); NVMEDIA (for access to hardware accelerators for image processing and computer vision); compute and NVIDIA CUDA tools.
NVIDIA DriveWorks SDK Provides reference applications, tools, and a comprehensive library of modules that leverage the computing power of the DRIVE AGX Platform.
DRIVE AV Provides modules for perception, mapping, and planning that leverage the DriveWorks SDK.
DRIVE IX Algorithms to visualize the vehicle’s surroundings, AI-based driver monitoring, and in-cabin assistance.

*TOPS measurements are per-chip


DRIVE AGX Pegasus™ Developer Kit (SKU 2200): Same as above with the addition of:
Component Description Details
Two “Turing” Discrete GPUs (DRIVE AGX Pegasus) Turing discrete GPU (TU104 SXM2) 130 TOPS connected to Xavier SoC over NVIDIA®NVLink™ (20 GB/s)
Memory Bandwidth Xavier >250 GB/s
Turing discrete GPU (TU104 SXM2) >750 GB/s
Overall >1 TB/s

Developing with DRIVE AGX


How to get it

The DRIVE AGX Developer Kit is available exclusively for members of the NVIDIA DRIVE Early Access Developer Program. Your company or university must have current legal agreements on file. Please contact your Account Manager (or contact us) to ensure necessary agreements have been signed before requesting to join the program. Users may only join with a corporate or university email address; register for the NVIDIA DRIVE Early Access Developer Program.


How to set up

You will need:


The development environment for DRIVE AGX requires a host development PC (not included with the DRIVE AGX Developer Kit). The image below illustrates the development workflow. All the tools illustrated below are installed on the host PC through the DRIVE Software SDK Manager.



Steps:





How to develop

Development Tasks Getting Started
Integrate DRIVE AGX in a vehicle
Integrate LIDAR or Radar DRIVE AGX
Accelerate image, point cloud and computer vision processing
  • Consider using DRIVE Core, which includes CUDA and PVA-accelerated image, computer vision, and point cloud processing modules.
  • If you would still like to implement your own processing algorithms, get CUDA samples using SDK Manager available on the Download page.
Use TensorRT to accelerate inference of deep neural networks developed for AV perception
  • Consider using DRIVE Networks, which includes a number of neural networks for obstacle, path, and wait perception.
  • If you would still like to build your own neural networks, get TensorRT samples using SDK Manager available on the Download page.
Use DriveWorks SDK to develop perception, mapping, and planning algorithms for autonomous driving
* = Membership in the NVIDIA DRIVE™ Early Access Developer Program is required to access this file.