Recent accidents involving Tesla cars may have been a setback for self-driving cars, but Nvidia believes a fast computer under the hood could make autonomous cars and cabs truly viable.
The company’s new Drive PX 2 model is a palm-sized computer for autonomous cars that will marry mapping with artificial intelligence for automated highway and point-to-point driving. The computer’s horsepower will help a car navigate, avoid collisions and make driving decisions.
The Drive PX 2 could be attractive to companies like Uber, which want to deploy autonomous cars as taxis. The computer is also targeted at car makers looking to develop fully or partially autonomous cars, which would typically need human intervention.
The computer is already ticketed for use in a self-driving car from Chinese company Baidu. The car will rely on the computer and cloud-based AI for car control, navigation, self parking and collision avoidance. The car is more a demonstration vehicle to highlight emerging autonomous car technologies.
In late August, Nvidia and Baidu announced they would jointly create a “cloud-to-car autonomous car platform” for car makers. It would use Baidu’s cloud services and mapping technology with Nvidia’s self-driving hardware and programming tools.
Automotive and tech companies are developing fully autonomous vehicles. The U.S. government is keeping a close watch, encouraging the development of autonomous cars and asking companies to keep driver safety in mind.
However, legal questions about self-driving cars remain. Standards are also being established by mapping and automotive companies so cars can exchange critical sensor data — like road conditions and weather — via the cloud. The sensor data, if freely available for autonomous cars to reference, could contribute to safer driving.
The new Nvidia computer is like a small PC, and combines computing power and algorithms to help cars navigate, self park, avoid obstacles and recognize signals, signs and lanes.
It has a Tegra chip that combines one six-core CPU code-named Parker and one Pascal-based GPU. It is a watered down version of the original Drive PX 2 announced at CES, which sits in the trunk of a car and is far more powerful with two CPUs and two GPUs.
Over time, the Drive PX 2 can be trained to provide better location and contextual awareness. Cars will get better at recognizing objects and can make better decisions on a wider variety of situations over time.
The Drive PX 2 trains autonomous cars by analyzing data gathered from cameras, GPS, ultrasonic sensors, radar, lidar and other components.
If the car can’t recognize an object or find its way to a location, it can refer to a larger knowledge base in the cloud. Nvidia said it will be possible to transfer a training model and data gathered by Drive PX 2 to the cloud, which will strengthen the knowledge base.
The autonomous car is central to Nvidia’s wide-ranging deep learning strategy, in which the company’s GPUs are being used for image recognition, natural language processing and scientific simulations. The GPUs are also being used to navigate drones and robots.
Nvidia is off to a good start in the autonomous car market. Volvo is developing an autonomous car based on the original Drive PX 2 introduced at CES, and the chip maker is already working with Ford, Audi and BMW on autonomous vehicle technologies.
Intel is Nvidia’s biggest competitor in the car market. BMW hopes to put an autonomous car on the road by 2021, for which Intel and Mobileye are supplying the hardware and software technologies.
The single-chip version of the Drive PX 2 will ship in the fourth quarter. Car makers, automotive suppliers and researchers are developing autonomous car prototypes based on various Drive PX computers. The Parker CPU has four 64-bit Cortex-A57 cores and two homegrown Denver 2.0 CPU cores, both based on ARM architectures.