​Nvidia redefines autonomous vehicle testing with VR simulation system

The Drive Constellation simulation environment allows for sensor data to be processed as if it were coming in from sensors on a real car cruising the streets.
Written by Asha Barbaschow, Contributor

Nvidia demonstrates Drive Constellation at Nvidia GTC 2018

Image: Supplied

Nvidia has on Tuesday announced Drive Constellation, a cloud-based system for testing autonomous vehicles using photorealistic simulation via virtual reality, aiming to speed up the delivery of autonomous cars in a safer and more scalable way.

According to Nvidia senior director of Automotive Danny Shapiro, every two minutes, four to five people die in vehicle-related accidents -- totalling 3,000 people per day globally.

"This is a big problem, so we're really focused on bringing the hardware and software to solve the challenge," he said.

Drive Constellation is a computing platform based on two different servers. The first runs the Nvidia Drive Sim software to simulate an autonomous vehicle's sensors, such as cameras, lidar, and radar; while the second boasts Nvidia Drive Pegasus, which is an artificial intelligence car computer that runs the autonomous vehicle software stack and processes the simulated data as if it was being fed in from sensors on a real car.

The Drive Sim software also generates the photoreal data streams to create a range of different testing environments, including natural occurrences such as storms, snow, high glare, low light, and different types of road surfaces and terrain.

"Essentially, we're running the complete hardware/software solution that would normally be in the vehicle, but we've moved it to the datacentre," Shapiro explained.

"The output from the simulator goes into the Drive Pegasus, performs its deep learning operations, it senses the environment ... and then instead of actuating the steering wheel on a real vehicle with the accelerator/brake, it sends those commands back to the simulator."

Driving commands from Drive Pegasus are fed back to the simulator 30 times per second.

This allows for the simulation of dangerous situations to drive the vehicle for billions of kilometres and test the autonomous car's ability to react, without putting an individual in harm's way.

"Self-driving is hard, we recognise that and I think it becomes more and more apparent every day as companies are out there really trying to solve this challenge," Shapiro told press at a briefing event.

"We're working very hard on this ... we've got teams developing both the hardware and the software. It requires a massive amount of compute, so Nvidia is really the ideal company to bring the hardware and the deep learning software to solve this challenge."

Nvidia CEO and founder Jensen Huang demonstrated the platform during his keynote on Tuesday, and announced the availability of Drive Constellation by the third quarter of this year.

"It's incredible what advances are going to be able to be made by our customers, being able to really accelerate the development and refine it," Shapiro added.

"RAND Corporation has put out a study that said we need to have billions of miles of driving to test these cars are safe for humans, but the reality is to do that would take more than our lifetime.

"Instead, by doing it in simulation, being able to have GPUs generating sensor data and feeding that back and testing it, our customers will be able to quickly refine their algorithms and accelerate their development."

Over the last year, Nvidia's automotive business has grown, with Shapiro highlighting that the company's ecosystem comprises virtually anyone developing something to do with cars, trucks, and delivery vehicles.

"The size of that ecosystem is over 370 different companies -- these are the car makers, the tier one suppliers ... but also we've got over 200 startups that are developing on our platform," he said. "We've seen such incredible growth in this ecosystem."

Volkswagon, Toyota, and Volvo are just a few manufacturing giants that have moved onto Nvidia's platform in the past year.

Nvidia suspends on-road self-driving testing

The importance of safety with autonomous vehicles was highlighted last week; Uber suspended its driverless vehicle testing in Tempe, Pittsburgh, San Francisco, and Toronto, when a female pedestrian died after being struck by an Uber car operating in autonomous mode in Tempe, Arizona.

Tempe police reported that the car was in autonomous mode with a human safety driver at the wheel, which is required by law, when it struck the pedestrian who walked into the street with her bike.

An Nvidia spokesperson on Tuesday told ZDNet it too has paused on-road trials in the wake of the tragedy.

"The accident was tragic. It's a reminder of how difficult SDC technology is and that it needs to be approached with extreme caution and the best safety technologies. This tragedy is exactly why we've committed ourselves to perfecting this life-saving technology," the spokesperson said.

"Ultimately AVs will be far safer than human drivers, so this important work needs to continue.

"We are temporarily suspending the testing of our self-driving cars on public roads to learn from the Uber incident. Our global fleet of manually driven data collection vehicles continue to operate."

Huang said Nvidia is dedicating itself to this problem, to make self-driving vehicles the safest option in the future.

Disclaimer: Asha Barbaschow travelled to GTC as a guest of Nvidia


Editorial standards