NVIDIA Corporation is an American company known as the innovative developer of video chipsets for video games. According to Jen-Hsun Huang, NVIDIA founder and CEO, the company has been engaged in research connected with autonomous vehicles technologies since 2007.
While dozens of corporations take part in a rather aggressive race to win leading positions in self-driving technologies, NVIDIA is receiving a profit on this market right now. The thing is that self-driving technology is an extremely compute-intense process requiring a powerful computer onboard able to process the high inflow of data. As NVIDIA is a leader amongst developers of video chipsets, it was able to easily shift its own focus to develop platforms for autonomous and semiautonomous vehicles.
Since January 2015, at the annual CES in Las Vegas, NVIDIA announced its renewed computer version developed for self-driving cars.
The first generation computers, presented at CES 2015, were named Drive CX and Drive PX.
– Drive CX was presented as a digital cockpit computer ensuring dashboard operation, navigation, informational, and entertainment systems for ordinary cars. Drive CX is equipped with 1 Tegra X1 chip and allows data output on displays with a total resolution of up to16.6 megapixels. It means that it can maintain 2 monitors work with 3840×2160 resolution or 4 monitors with 1920×1080 resolution each.
– Drive PX is a more advanced 1 TFLOP chip developed for (semi) autonomous self-driving technology. It has 2 Tegra X1 chips which allow computing video data captured by 12 cameras with 1920×1080 resolution at 60 frames per second. Using cameras, Auto-Valet car system helps to render 3D environment and find a parking lot for a car on its own, even if there are no road surface markings. It has become possible due to neural network technology utilization. It is able to recognize the surrounding objects very precisely, even if they partially overlap each other.
The second generation computer, represented at CES 2016, was named Drive PX 2 and it was developed in two configurations – Autocruise and Autochauffeur.
Drive PX 2 Autocruise consists of 1 Tegra X2 processing chip and consumes only 10 W. This platform is destined for cars with partial autonomy, including cruise control, highway movement, or movement in traffic jams.
Drive PX 2 Autochauffeur is designed for totally autonomous, driverless cars. It consists of 2 Tegra X2 processing chips based on Pascal-Architecture technology. This supercomputer delivers 24 DL TOPS of compute performance (24 trillion operations per second) and is capable of monitoring 36,000 points in the surrounding environment 8 times per second providing enough power to compute all data coming from LIDAR, cameras, ultrasonic sensors, and a radar.
Cloud-to-Car HD Mapping
NVIDIA also offers carmakers to use its mapping platform which works in real-time to create and update HD maps and allows self-driving cars to navigate in the surroundings maximally well defining their location. This technology works using NVIDIA’s software named DriveWorks provided as a tool for autonomous vehicles developers.
The third generation computer was presented at CES 2017 and named Xavier.
Jen-Hsun Huang, NVIDIA CEO, called this platform as an AI car supercomputer. Xavier consists only of one Tegra Xavier chip and 512-cores Volta GPU. It delivers 30 DL TOPS of compute performance. This supercomputer also includes AI platform capable of self-learning to self-drive while watching the human driver’s actions.
As an example of AI which has self-learnt how to self-drive, the BB8 prototype has been presented which was built based on Lincoln car. This vehicle drove along the route on public roads, including a speedway. BB8 was shown for the first time in September 2016. It showed its capability of driving unsurfaced roads with no markings bypassing obstacles and making sharp turns with blind spots.
Jen-Hsun Huang has also presented the AI Co-Pilot system at CES 2017. Unlike the full-fledged AI self-driving, AI Co-Pilot will be widely used in practice soon. The main goal of AI Co-Pilot is to make driving safer. The hardware part includes 2 rear and front view cameras, 2 side view cameras, 1 camera focused on the driver’s face, sensors, and a microphone. Cameras with sensors inside the car are able to recognize the driver’s identity, read human lips, follow head movements, as well as follow an eye line. According to Jen-Hsun Huang, it is no less important to watch a driver and their condition than to watch the road for safety reasons. The system will warn a driver about any dangerous situation. For instance, if they try to make a turn and a cyclist is approaching in the same direction, or if a pedestrian approaches and a driver looks in another direction. Due to speech recognition and reading lips technologies, AI will execute commands like making a new route or changing music.
NVIDIA is a partner for dozens of other companies that are engaged in driverless cars creation. Among them are Audi, Tesla, Mercedes-Benz, BMW, Volvo, Baidu, Honda, Toyota, Luxgen, and Bosch.
At the beginning of January, at CES 2017, it became known that Audi is the main NVIDIA’s partner in autonomous cars development and that their cooperation has already lasted for about 10 years. The companies said that they would produce their self-driving vehicle before 2020. They showed Audi Q7 based on Xavier AI car supercomputer that learned how to drive watching a human driver only during 4 days.
In October 2016, Tesla announced that from then on, every new model S, X, and 3 would receive Hardware 2.0 which includes AI platform based on Nvidia Drive PX 2. As it is known, these companies have been cooperating for a long time, since model S was manufactured.
At CES 2017, NVIDIA also announced its partnership with Mercedes-Benz lasting since 2014 up to that time. In 2018, the companies plan to produce a partially autonomous car with integrated AI. Most likely, they are talking about Co-Pilot system that will make driving safer.
In 2017, Volvo started a project named “Drive Me”. It includes 100 self-driving Volvo XC90 models equipped with supercomputers Drive PX 2 that shuttle their first passengers on a set route in Gothenburg, Sweden.
At the World Conference in Beijing at the end of 2016, it was announced about the cooperation between NVIDIA and an Internet giant Baidu. Both companies are leaders in Artificial Intelligence development. The goal of their cooperation is to create cloud-to-car autonomous car platform. The platform is a 3D map in the cloud that will renew in real time due to the participants’ driverless cars connection to it.
In mid-March 2017, it was reported about NVIDIA’s partnership with a spare parts supplier Bosch. The companies will work together on an AI self-driving car computer.
In cooperation with other partners, NVIDIA represents advanced information and entertainment, navigation, and other systems based on NVIDIA’s Tegra mobile processor.