NVIDIA shares the beliefs of some experts regarding the evolution of automated vehicle control systems. Cars that in one situation or another require the presence of a driver in the cabin will refuse his participation in control very slowly and gradually. Initially, vehicles designed for unmanned operation will appear on the market even earlier than customer “robotic cars,” but with one important caveat: they can be operated in a strictly limited area.
This week, NVIDIA representatives took part in two industry events for investors: Vice President of Autonomous Cars Rob Csongor spoke at the conference Automotive Technology, and CFO Colette Kress championed NVIDIA's interests at the event . The most interesting thing is that the company’s financial director also had to touch on the topic of “autopilot,” although in the course of answering a question about the profile of his activities.
Financial returns from autopilot will be noticeable from 2022
Both NVIDIA representatives were asked at different times by the presenters of both events about the timing of revenue growth from the supply of automotive components for ADAS systems to significant levels. Now, as you know, NVIDIA's revenue in the automotive segment is determined primarily by components for on-board multimedia systems. At the same time, the company has over two hundred clients who are, to one degree or another, involved in the development of automatic control systems for robots, cars and trucks.
Colette Kress expressed the hope that in a little over a year, production cars with control systems using NVIDIA components will begin to enter the market in significant quantities. Apparently, the Tesla fleet of electric vehicles on the NVIDIA computing platform has not done much well in terms of revenue so far, and we must not forget that the American automaker is moving to its own component base. Among the most important partners in terms of popularizing active driver assistance systems, NVIDIA mentions Volvo Cars. As soon as other customers begin to join the Swedish brand, NVIDIA will be able to significantly increase revenue in the automotive segment. This will happen no earlier than 2022, according to company management estimates.
Platform NVIDIA will help lagging automakers catch up Tesla
If we return to the speech of Rob Csongor, who oversees the creation of autopilot systems at NVIDIA, you can clearly feel the proprietary self-confidence of the company founded by Jen-Hsun Huang. The autopilot segment is now attracting many developers and market players, but when it comes to leadership in the field of specialized processors, NVIDIA gives credit to Intel, which still uses the Mobileye brand. NVIDIA considers itself and this Israeli “daughter” of Intel to be potential leaders in the segment.
Another thing is that the Mobileye ecosystem is quite closed, while NVIDIA cultivates the idea of an open platform. The company does not impose on clients a full set of tools for developing automatic systems. Everyone can get exactly what they need from NVIDIA. For example, BMW uses Mobileye components in its on-board systems, but cloud computing and simulations are carried out on NVIDIA DGX systems. A strict vertical integration is not imposed on anyone.
NVIDIA highly appreciates Tesla's achievements in developing autopilot. Until a certain point, this electric vehicle manufacturer relied on NVIDIA's component base. Then Elon Musk decided to equip the cars with Tesla's own processor, but NVIDIA did not hold a grudge. On the contrary, according to Rob Csongor, all Elon has to do is call NVIDIA, and the company will come to the rescue again. For those companies who are just now ready to join the race to create an autopilot, NVIDIA recommends using its components to close the gap with Tesla.
Architecture with centralization of computing functions NVIDIA does not consider it defective
The host of the Deutsche Bank event asked the vice president of NVIDIA which approach to the layout of automotive systems the company considers more promising. Traditional players in the automotive electronics segment are developing distributed systems with duplication of functions, when individual nodes can function independently if neighboring ones fail.
Rob Csongor's confessions in this context are not unexpected, since he advocates the centralization of computing functions. This is not surprising, because NVIDIA has a rich tradition of developing computer components, where functions are also highly centralized. At the same time, the vice president of NVIDIA does not deny that an autonomously controlled machine will become one of the most complex computers in the world; it will be able to simultaneously work with dozens of neural network models, process huge amounts of information, and at the same time provide a high level of security at high speeds.
When developing components for cars, NVIDIA is forced to focus on a certain “energy budget”. Within this value, it strives to provide the maximum level of performance. According to the company's vice president, NVIDIA has succeeded in maintaining this balance when developing processors of the Tegra family.
Cars with the highest level of autonomy will appear earlier than with the penultimate one
NVIDIA clearly understands the difference between creating transport systems with hybrid control, when control passes from a person to automation and vice versa, and developing fully autonomous vehicles, the control of which is not provided for by humans at all. Recently, loud promises to release fully automatic cars on the market by a certain date have been heard less and less often from automakers, and this is not only due to the awareness of the technical complexity of the problem. As the vice president of NVIDIA explains, much will depend on the availability of legislative permits for the operation of such vehicles. No one has done this before, so it is difficult to predict when all legal obstacles will be removed.
If we talk purely about technology, then cars with level 2+ autonomy will gradually evolve into vehicles of the penultimate (fourth) level according to the SAE J3016 classification, which provides for the transfer of control to the driver only in exceptional cases. Level XNUMX autonomy does not require human intervention at all, but such “robotic cars” may appear on the roads even earlier than cars with level XNUMX systems, since they will be operated in limited areas with a well-studied and predictable situation. In addition, construction equipment can outpace passenger equipment in terms of control automation speed, precisely for the reasons stated above. Various types of cargo shuttles will also be automated earlier than mainline tractors.
Source: 3dnews.ru