Nvidia Corp. has unveiled what it claims is a production-ready autonomous vehicle platform powered by a combination of sensors, its Ampere graphics processing units and its Drive Orin system-on-a-chip, plus a unique mapping technology platform.

Nvidia Drive Hyperion 8 was announced today at Nvidia GTC 2021, and is said to be designed with safety in mind, powered by an array of lidar- and radar-based sensors built by companies such as Continental AG, Sony Corp., Hella KGaA Hueck & Co. and Luminar Technologies Inc.

The company said Drive Hyperion 8 is designed to provide a secure base for autonomous vehicle development, with two Drive Orin SoCs enabling redundancy and fail-over safety as well as ample computing power for level 4 self-driving and intelligent cockpit features. Meanwhile, the Ampere GPUs make it possible for developers to test and validate various new software capabilities, Nvidia said.

One of the key components of Drive Hyperion 8 is the DriveWorks Sensor Abstraction Layer that promises to make sensor setup a breeze with the use of plugins, while the Drive AV software is powered by deep neural networks for perception, mapping, planning and vehicle control.

The platform is modular too, so car manufacturers will be able to use the components they like, integrating them with third-party systems and tools. It’s also scalable, with Nvidia planning to roll out regular upgrades via the Nvidia Drive Atlan and DriveWorks application programming interfaces.

Nvidia claims that by providing a complete sensor setup atop of a centralized compute and AI software system, Drive Hyperion 8 provides everything required to validate and control an intelligent vehicle on the road. The sensor suite encompasses 12 camera, nine radar sensors, 12 ultrasonic sensors and a single front-facing lidar sensor. Those systems are coupled with sensor abstraction tools that enable car makers to customize the platform to meet the needs of their unique self-driving car systems, Nvidia promised.

The system also makes it possible to test and validate individual bits of hardware during the development stage.

“Nvidia has led the modern compute revolution and the industry sees them as doing the same with autonomous driving,” said Luminar founder and Chief Executive Austin Russel. “The common thread between our two companies is that our technologies are becoming the de facto solution for major automakers to enable next-generation safety and autonomy. By taking advantage of our respective strengths, automakers will have access to the most advanced autonomous vehicle development platform.”

AI-driven parking system

Nvidia said Drive Hyperion 8 will effectively allow AI to take over the wheel and perform numerous vehicle-based tasks in the same way humans can. As one example, it highlighted the new parking capability of its Nvidia Drive Concierge AI assistant that Nvidia claims is able to park a car in the most difficult of spaces to enable driverless drop-off and pick-up experiences.

As Nvidia explains, parking a car is one of the trickiest tasks a human driver has to perform. The have to pay attention to road markings to determine if they can in fact park in a specified place, whether it’s on a road or in a parking lot. They also have to consider if a fire hydrant is nearby, or if there are time limits for parking at a particular spot. Add to that, there might be obstacles such as concrete pillars or shopping carts in the way.

Drive Concierge gets around these challenges by using an array of vehicle-based sensors and fish eye cameras, feeding the data they create into artificial intelligence algorithms that perform the actual maneuver.

Nvidia provided a fairly in-depth explanation of how it works. It uses an “Evidence Grid Map” deep neural network that relies on sensory data to generate a real-time dense grid map in the immediate vicinity of the vehicle. After that, the “ParkNet” neural network fuses images from multiple on-board cameras and identifies any possible parking spots nearby. Then, a “Dense Fusion” neural network tries to understand if the space surrounding the vehicle is free of any obstacles.

These all combine to help the vehicle determine which parking space is the best option – before feeding everything to the AI autopilot that safely completes the parking maneuver without colliding into any obstacles or nearby parked cars.

AI driving assistants

Nvidia Drive Concierge is one of two new AI-based driving assistants Nvidia unveiled today at GTC. The parking capability is its standout feature, enabling a vehicle to drop off its driver at the front of a store and then go find a parking spot all by itself.

Besides that, it can also perform tricks ranging from making recommendations to booking reservations, making phone calls and providing alerts, for example if a wallet is left behind in the vehicle after the occupants leave. Drive Concierge will be personalized for each driver, Nvidia said.

As for Drive Chauffeur, it can be thought of as a kind of autopilot capability that takes control of the vehicle when the human driver needs time to rest. Using the high-performance compute architecture and sensor set of Drive Hyperion 8, Nvidia said, Drive Chauffeur can drive from address to address. For those who di want to drive, the system provides active safety features.

DRIVE Mapping

Besides powering autonomous vehicles, Nvidia Drive Hyperion 8 will also help to fuel the company’s new mapping technology with reliable, real-time data.

At GTC 2021, Nvidia founder and Chief Executive Jensen Huang explained that mapping is one of the fundamental pillars of autonomous driving, serving as the memory of self-driving vehicles. Those maps provide a baseline understanding of each vehicles driving environment, but to do that effectively, they need to be extremely detailed, with a level of accuracy within centimeters and also reflect current road conditions such as lane closures or a traffic jam.

To create such accurate maps, Nvidia say it can’t rely on a third-party tool such as Google Maps. It needs to build its own, which explains why it acquired the mapping startup DeepMap Inc. earlier this year.

Since that acquisition, Nvidia said it has worked to combine DeepMap’s technology with its Nvidia Drive technology. The result is Nvidia Drive Mapping, a powerful system that enables vehicles around the globe to work together to create and update a map of the world’s roads in real-time that can be used by fleets of autonomous vehicles.

The system leverages the constant stream of perception data from cameras, radar and lidar sensors fed to it by vehicles running the Drive Hyperion 8 platform. While the cameras provide imagery, the radar data helps to provide a layer of redundancy, Nvidia said, to enable localizing and driving in poor weather or lighting conditions where cameras may be ineffective.

Nvidia explained this data is then fed into an Nvidia Drive AGX AI-based compute platform inside each vehicle. The mapping network uses this real-time data for perception, identifying intersection details, traffic lights, parking spots, road and lane boundaries, essentially replicating everything a human driver would see if they were in control of the vehicle.

Nvidia said Drive Mapping can be thought of as a crowdsourced platform, and that its effectiveness will grow as more vehicles adopt the technology. Drive Mapping leverages Nvidia DGX SuperPOD infrastructure to maintain the maps at a global scale, the company said. The AI systems ingest terabytes of perception data from Drive Hyperion vehicles to create and update maps.

Images: Nvidia