DEVELOPER BLOG

Autonomous Vehicles |

Fast-Track ADAS and AV Research with NVIDIA DRIVE AGX

Driver assistance technology is an incredibly active research domain – from supervised assistance functions all the way to fully autonomous driving. The best way to showcase the capabilities of novel AV approaches is to demonstrate them in a real car, but there are significant challenges to this type of deployment.  

Getting to the point where a new approach for multi-agent prediction, camera-based localization or night-time obstacle detection runs in a car requires efforts on multiple levels: 

  • A vehicle needs to be retrofitted with sensors, AI compute hardware, data storage, vehicle IO interface and potentially even a drive-by-wire interface. 
  • Middleware needs to be implemented to orchestrate the individual functional components. 
  • Fundamental software components for vehicle IO, sensor interfacing, calibration, recording need to be implemented.  

The NVIDIA DRIVE AGX autonomous vehicle compute platform is designed to substantially simplify these efforts, enabling researchers to focus on what’s most important. 

Research in Motion 

Prof. Daniel Watzenig and the Autonomous Racing Graz team (a collaboration between Graz University of Technology and Virtual Vehicle Research) are pushing autonomous driving to the limit with driverless racing. Powered by NVIDIA DRIVE AGX, the team’s vehicle managed to secure the third place in the 2020 Roborace Season Alpha and was the highest-finishing academic team.  

“Autonomous racing comes with very high requirements against software and hardware but also against weight and space. As researchers, we need to focus on trying out new approaches and iterate quickly – the NVIDIA DRIVE AGX platform is a perfect fit to these needs and has proven to be a key factor of our team’s success.” Watzenig said. 

The team has published their research in papers such as Smooth Reference Line Generation for a Racetrack with Gates based on Defined Borders and Increased Accuracy For Fast Moving LiDARS: Correction of Distorted Point Clouds with more publications coming. They’ll also race in the new Indy Autonomous Challenge, competing against 30 other university teams.  

Why DRIVE AGX? 

The NVIDIA DRIVE AGX Developer Kit provides the hardware, software and sample applications needed for the development of autonomous vehicles. The platform is built on production auto-grade silicon, features an open software framework, and has a large ecosystem of supported auto grade sensors to choose from. The developer kit comes with unrivaled compute performance in a compact form factor, reaching up to 320 TOPs (INT8).  

An ADAS and AV development platform must offer a software environment that supports established research tools. The DRIVE AGX Developer Kit supports a tailored Linux-derivate that provides a familiar environment to researchers. The possibility of running numerous popular Linux libraries makes it easy to migrate existing code. 

The comprehensive NVIDIA DRIVE Software gives researchers a head start with a rich software suite that provides low-level hardware interfacing and middleware out-of-the-box.  
DRIVE OS provides Hypervisor, CUDA, deep learning inference with TensorRT and camera interfacing. DriveWorks comprises tools for calibration, sensor and vehicle interfacing and recording and many additional tools and APIs. Finally, samples showcase typical AV modules that can be used as a reference.  

Another important research requirement is that the development platform should be compact and durableWhile it is possible to install desktop computers and additional hardware in cars to develop ADAS and AV functions, these setups typically require a large amount of space and add additional points of failure, especially since these components will likely not be auto-grade. 

Getting Started with DRIVE AGX 

The DRIVE AGX Developer Kit is available through the NVIDIA DRIVE Developer Program for DRIVE AGX. Please contact your NVIDIA representative (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. 

The DRIVE AGX Developer Kit comes with plenty resources to jump-start development: 

 
* only accessible post NVIDIA DRIVE Developer Program for DRIVE AGX registration.