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 must be retrofitted?with sensors,?AI?compute?hardware,?data storage,?a vehicle I/O interface,?and?potentially?even?a?drive-by-wire interface.?
- Middleware?must?be implemented to orchestrate?the individual?functional?components.
- Fundamental?software?components?for vehicle?I/O, sensor interfacing,?calibration, and recording?must 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 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 derivative?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?includes?tools for calibration,?sensor and?vehicle interfacing,?recording,?and?other?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?durable.?While 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?as 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 that the 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 of resources?to jump-start?development:?
- The Developing with DRIVE AGX section explains the setup and provides a quick-start guide and an intuitive?setup video.???
- Extensive?documentation?explains technical details and offers samples?that you can use as reference for your own applications.?
- Numerous?posts?and?webinars*?show?how to build applications on DRIVE AGX – from?fundamentals?like?sensor interfacing?all the way up to?Researching and Developing an Autonomous Vehicle Lane-Following System?
*?Accessible?with registration in the NVIDIA DRIVE Developer Program for DRIVE AGX
