Let’s admit. Self-driving cars are the next or perhaps already a big thing in the modern automobile market. And as many famous players are already working towards building their own version of autonomous vehicles for the future, Nvidia has unveiled the possible source to power these intelligent vehicles.
Drive PX 2 is the powerful sequel to the Drive CX car computer by Nvidia which was launched last year. And what makes it more powerful you ask ? 12 CPU cores, 8 teraflops processing power, 250 W power consumption, 24 deep learning tera-operations per second, two next-gen Tegra processors, as well as a Pascal-based GPU, and a 16 nm architecture.
If these technical specifications don’t make much sense to you, then consider the facts that Drive PX2 boasts of a power equivalent to 150 MacBook Pros; can process the inputs of 12 video cameras, plus lidar, radar, and ultrasonic sensors; and can process 2,800 images per second.
And all this is packed into a lunchbox-sized device aided by a water cooling system to handle the enormous computing calculations done by it. No wonder that Jen-Hsun Huang, NVIDIA’s CEO is claiming that it’s the first supercomputer made for cars.
Drive PX2 was recently unveiled by Huang at the company’s press conference on the eve of the 2016 International CES, the big tech trade show in Las Vegas this week. It is built keeping in mind the deep learning needs of a self-driving car to identify objects, adapt and respond automatically to the changing surroundings, and train themselves accordingly for all sorts of scenarios.
Our vision is to create the computing platform by which we can achieve these goals. The car is paying attention all of the time. It is a virtual copilot that will keep you out of harm’s way,
Volvo is the first company to adopt the PX2 officially and is going to use it to test its upcoming self-driving cars. In addition to it, Audi, Daimler, BMW, and Ford have also partnered with Nvidia to use PX2 in their cars.
The company also announced its deep learning platform called DIGITS which can be used to share the information collected by the self-driving cars to the cloud network. It is aimed at assisting other autonomous car makers to get started with the testing process of their vehicles.
DIGITS has helped Nvidia to develop a deep neural network which it calls Drivenet. Drivenet is aimed towards solving the common problem of identification and perception of real life objects by the self-driving car systems.It is able to identify five different classes of objects, including pedestrians and motorcyclists on the road.
It is a powerful and complex network of nine separate neural networks termed as “inception layers” embedded within each other. It contains the equivalent of 37 million neurons and has trained 120 million objects so far. Moreover, it gets better with time as it learns more and more about a wide variety of scenarios found during driving.
Drivenet is already being used by Audi to test its self-driving cars. Nvidia showed in a demo how Audi used Drivenet to analyze visual data taken from a snowstorm, and after just one night they were able to detect data that the human eye can’t see.
According to Nvidia, identification and perception of objects and planning basic paths are just the beginning for self-driving cars. In order to effectively function in all driving conditions, they need to get more intelligent and efficient with time, and able to identify various circumstances as well.
And this is something which the company is looking forward to solving by venturing into an arena beyond high-end PC gaming where its GPUs can also power the future self-driving cars.