In-Vehicle Supercomputer Holds Promise for Self-Driving Cars

Nvidia Corp. says its Drive PX 2 supercomputer’s advanced artificial intelligence and graphics processing capabilities will help pave the way for fully autonomous vehicles.

The Santa Clara, Calif.-based company, which pioneered the use of separate graphic processing units for computers in 2007, detailed the second-generation PX 2 this week at the Consumer Electronics Show in Las Vegas.

The new computer, which is about the size of a lunchbox, features two Tegra GPUs with a combined processing power of 8 teraflops. Nvidia says this is equivalent to 150 MacBook Pros operating together and is about 10 times as powerful as the first-generation Drive PX introduced a year ago. The system, which draws about 250 watts, uses liquid cooling rather than fans to help bolster performance in hot weather.

The PX 2 can simultaneously process information from a dozen high-resolution video cameras, GPS, lidar, radar, ultrasonic sensors and other inputs. This allows self-driving vehicles to quickly and more precisely identify and analyze an array of complex traffic variables, according to the supplier.

The system can run 24 trillion “deep learning” artificial intelligence operations per second. Such operations are said to be particularly well suited for processing unexpected challenges, such as road debris and erratic maneuvers by other vehicles. It also is superior for operating in poor weather and other conditions that could limit the effectiveness of one or more individual sensors, according to the supplier.

The initial focus is on improving object detection and recognition. This eventually will include the ability to distinguish between everyday cars and trucks and emergency vehicles and to enable vehicles to operate autonomously on snow-covered roads with unclear markings.

Nvidia predicts supercomputer-controlled vehicles eventually could achieve “superhuman” levels of situational awareness. The task will be aided by special training algorithms that allow the computer’s software to learn from data gathered during previous drives and through shared information uploaded to the cloud by other vehicles.

Nvidia says more than 50 companies—including BMW, Daimler and Ford—currently are using its first-generation supercomputers in developmental programs. The company will begin shipping the PX 2 to select partners in the second quarter, with a full rollout due late this year.

Volvo already has announced it will use the supercomputer in a fleet of 100 XC90 crossovers vehicles in tests of automated features that will begin in Sweden next year.