Can test the safety and reliability of autonomous vehicles

NVIDIA is introducing a cloud-based system that can test self-driving cars in realistic simulations, providing a safer, more scalable way to get self-driving cars on the road.

In his GTC 2018 keynote, NVIDIA founder and CEO Jensen Huang announced NVIDIA DRIVE Constellation, a computing platform built on two different servers.

The first server runs NVIDIA DRIVE Sim software to simulate the vehicle’s sensors, including cameras, lidar and radar. The second server’s powerful NVIDIA DRIVE Pegasus AI car computer runs the entire autonomous vehicle software stack and processes the simulated data as if it came from the sensors of a real car driving on the road.

“Deploying production self-driving cars requires solutions that can test and validate billions of miles to ensure the safety and reliability our customers demand,” said Rob Csongor, vice president and general manager of Automotive at NVIDIA. “NVIDIA DRIVE Constellation allows us to do this by combining our expertise in visual computing and data center. “Virtual simulations have allowed us to refine our algorithms by testing them over tens of miles, including custom scenarios and rare cornering cases, in a fraction of the time and cost of real-world driving,” he said.

The simulation server is based on NVIDIA GPUs, and each generated simulation sensor data stream is passed to DRIVE Pegasus for processing.

Driving commands from the Pegasus are relayed back to the simulator, completing a digital feedback loop. This “hardware-in-the-loop” cycle occurs about 30 times per second and is used to verify that the algorithms and software running on the Pegasus are operating the simulated vehicle correctly.

DRIVE Sim software generates photorealistic data streams to create a variety of test environments. It can simulate a variety of weather conditions, such as storms and snowstorms, glare or limited visibility at night during different times of the day, and a variety of road surfaces and terrains. It can also simulate dangerous situations to test the ability of autonomous vehicles to respond while avoiding the risk to real drivers.

“Autonomous vehicles need to be developed as systems that can cover everything from training to testing to real driving,” said Luca De Ambroggi, research and analyst director at IHS Markit. “NVIDIA’s end-to-end platform is the right approach. “With Drive Constellation for virtual testing and validation, we will get one step closer to production autonomous vehicles,” he said.