The result of more than 18,000 hardware engineer years and 10,000 software engineer years in development, the NVIDIA DRIVE AGX platform delivers a true autonomous driving solution for production-level vehicles.
With today’s announcement of the availability of the NVIDIA DRIVE Hyperion sensor suite, which enables manufacturers to validate their technology in the vehicle, the DRIVE platform makes it possible for companies to deploy self-driving cars at scale.
At GTC Europe Wednesday, NVIDIA CEO Jensen Huang outlined the end-to-end hardware and software solution for safe self-driving, demonstrating its capability with an 80-kilometer highway loop driven in an autonomous test vehicle with no human intervention.
“This is not a demo, this is something you can get right now,” Huang said. “The entire computer is production ready and automotive grade.”
The journey to a production-level platform began more than a decade ago, with the vision that future autonomous machines would demand an entirely new type of processor that could handle diverse and redundant compute workloads. From this vision, the Xavier system-on-a-chip (SoC) was born.
The solution has scaled to an entire high-performance compute platform for full autonomous driving that can withstand the rigors of operating inside a vehicle, as well as solutions to test and validate various levels of self-driving technology.
DRIVE AGX by the Numbers
The DRIVE AGX platform is a true engineering feat offering the performance capabilities for production-level autonomous driving. Those tens of thousands of engineering years of effort have resulted in a DRIVE AGX Pegasus compute platform the size of a laptop that delivers the performance of more than 60 laptop computers. DRIVE AGX Pegasus development platform can scale up to 320 TOPS of peak performance.
High compute performance translates to safety when it comes to autonomous driving. This capability enables diverse and redundant processes, meaning the system can continue to operate in the event of malfunction or failure is detected. The industry agrees, with more companies running autonomous vehicle prototypes using DRIVE AGX than any other platform.
“We don’t rely on any one algorithm, sensor or computing block,” Huang said. “As a result our progress has been incredible.”
A key component to this functional diversity is software. DRIVE Software is an extensible, constantly improving suite of self-driving and user experience functionalities powered by AI, accelerated computer vision and image processing. At its core is DRIVE OS, which delivers a safe, secure, real-time hypervisor and other core capabilities for autonomous driving applications.
Layered upon that, sensor processing from radar, camera and lidar sensors enable the car to build a holistic world model, with DriveWorks applications, such as deep neural networks for autonomous vehicle perception and data recording.
In addition to self-driving capabilities, DRIVE Software enables in-vehicle functionalities, such as DRIVE IX for driver monitoring and user experience and DRIVE AR for visualization of the car’s perception of its environment.
Streamlined Evaluation
With hundreds of companies worldwide developing technologies for autonomous driving, solutions must be flexible and customizable. NVIDIA DRIVE AGX developer kits let manufacturers streamline the develop of autonomous vehicle systems.
The recently announced DRIVE AGX Xavier and DRIVE AGX Pegasus Developer Kits — which include the AI car computer and vehicle harness to connect the platform to the car, international power supply, camera sensor and other accessories — let companies test their self-driving applications.
For validation in the car, NVIDIA DRIVE Hyperion Development includes a DRIVE AGX Pegasus development platform along with sensors for autonomous driving (seven cameras, eight radars and optional lidar), sensors for driver monitoring, sensors for localization and other accessories. DRIVE Hyperion allows developers to experience and evaluate NVIDIA’s DRIVE Software suite consisting of DRIVE AV, DRIVE IX and DRIVE AR.
Manufacturers and developers can take advantage of the DRIVE software APIs to integrate their own software and test it in a real vehicle. DRIVE Hyperion is a complete sensor and compute platform setup that can be retrofitted onto a test vehicle.
With these powerful tools for development and validation, manufacturers now have everything at their fingertips to deploy self-driving for safer, more efficient transportation for all.