MENU

Nvidia puts a lot of effort into autonomous driving

Nvidia puts a lot of effort into autonomous driving

Feature articles |
By Christoph Hammerschmidt



The development partnership with the Toyota Group is handled through the Toyota Research institute for Advanced Development (TRI-AD). Together, researchers from Nvidia and TRI-AD plan to develop self-driving cars, train them by means of AI and then validate them. The partnership builds on an ongoing relationship with Toyota to utilize Nvidia’s Drive AGX Xavier AV computer and is based on close development between teams from Nvidia, TRI-AD in Japan and Toyota Research Institute (TRI) in the US. The partnership spans multiple topics and assets, including the advancement of an AI computing infrastructure (of course based on Nvidia GPUs), autonomous driving simulations based on the Nvidia Drive Constellation platform introduced past year, and the further development of in-car real-time computing platforms for automated vehicles.

The agreement includes the development of an architecture that can be scaled across vehicle models and types, accelerating the development and production timeline, and simulating the equivalent of billions of miles of driving in challenging scenarios. In fact, such a computer is currently something like the holy grail of the automotive industry – as part of the digitization of the vehicle, carmakers and suppliers are putting great efforts in replacing the chaotically grown ECU-oriented electronics architecture of the vehicles through a rather centralized approach with only few computing units that run the functionality of former ECUs as software tasks.


The paramount challenge of autonomous driving is establishing consumer confidence by reducing the number of traffic fatalities. “Our vision is to enable self-driving vehicles with the ultimate goal of reducing fatalities to zero, enabling smoother transportation, and providing mobility for all,” said Dr. James Kuffner, CEO of TRI-AD. “Our technology collaboration with Nvidia is important to realizing this vision. We believe large-scale simulation tools for software validation and testing are critical for automated driving systems.”

In this context, AI, and specifically deep learning, has become a vital tool for the production of next-generation automated vehicles, particularly because of the need to recognize and handle the nearly infinite number of scenarios encountered on the road.

Nvidia Drive Constellation
Drive Constallation at work: The Constellation Simulator (left) 
generates the virtual environment signals; the Constellation Vehicle (right)
processes these data as a “virtual vehicle” 

Before a car – and in particular a self-driving car – can hit the road, its functions need to be rigorously tested. Simulation has proven to be a valuable tool for testing and validating AV hardware and software before it is put on the road. As part of the collaboration, TRI-AD and TRI are utilizing the Nvidia Drive Constellation platform for components of their simulation workflow. Drive Constellation is a data center solution, comprising two side-by-side servers. The first server — Constellation Simulator — uses Nvidia GPUs running Drive Sim software to generate the sensor output from a virtual car driving in a realistic virtual world. The second server — Constellation Vehicle — contains the Drive AGX car computer, which processes the simulated sensor data. The driving decisions from Constellation Vehicle are fed back into Constellation Simulator, aiming to realize bit-accurate, timing-accurate hardware-in-the-loop testing. According to Nvidia, safety agencies such as TÜV Süd are already using the simulation platform to formulate their self-driving validation standards.


At the same opportunity, Nvidia also showed extensions to its Drive AV autonomous vehicle software: A new planning and control layer has been added, designed to enable a safe driving experience. A primary component of this software is Safety Force Field (SFFTM), a robust driving policy intended to protect the vehicle, its occupants and other road users. SFF analyzes and predicts the dynamics of the surrounding environment by taking in sensor data and determining a set of actions to protect the vehicle and other road users. According to its developers, the SFF framework ensures these actions will never create, escalate or contribute to an unsafe situation and includes actions necessary to mitigate potential danger.

Backed by robust calculations, SFF makes it possible for vehicles to achieve safety based on mathematical zero-collisions verifications, rather than attempting to model the high complexity of real-world scenarios via limited statistics. Running on the Nvidia Drive platform, frame-by-frame, physics-based SFF computations are performed on vehicle sensor data.

SFF has also undergone validation using real-world data and bit-accurate simulation, including scenarios involving highway and urban driving that would be too dangerous to recreate in the real world.

A unique feature of the software is its ability to take into account both braking and steering constraints. This dual consideration should help eliminate several problematic vehicle behavior anomalies that could arise if they were separated. The policy follows one core principle of collision avoidance as opposed to a large set of rules and expectations.

“By removing human error from the driving equation, we can prevent the vast majority of collisions and minimize the impact of those that do occur,” claims David Nister, vice president of Autonomous Driving Software at Nvidia. “SFF is mathematically designed such that autonomous vehicles equipped with SFF will, like magnets that repel each other, keep themselves out of harm’s way and not contribute to unsafe situations.”  


As an open platform, SFF can be combined with any driving software. A safety-decision making policy in the motion planning stack, SFF monitors and prevents unsafe actions. It separates obstacle avoidance from a long tail of complicated rules of the road. When running on a high-performance compute platform like Nvidia Drive, it adds another layer of diversity and redundancy features to deliver very high levels of safety.

Related articles:

Nvidia makes further inroads to automotive industry

Daimler takes Nvidia inside car central computer

NXP, Kalray defy Nvidia with automotive AI computing platform

Nvidia Drive AGX platform on-board autonomous vehicle fleet

 

If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :    eeNews on Google News

Share:

Linked Articles
10s