Musk: Relying On 5G Technology, There Is No Fundamental Challenge For Automatic Driving

With the landing of 5g, more and more application prospects become clear.

On July 9, Tesla CEO Elon Musk (hereinafter referred to as "Musk") spoke at the World Conference on artificial intelligence, saying that 5g automatic driving will come soon. "I am very confident that 5g automatic driving will be realized in the future, and I think it will be realized soon. Tesla is very close to 5g autonomous driving, and I am confident that we will complete the basic functions of the L5 level this year. "

Speaking at the World Conference on artificial intelligence, musk said 5g autonomous driving will come soon. Picture vision China

In his opinion, relying on 5g technology, there is no underlying fundamental challenge in the current field of automatic driving. However, many details still need to be solved. "We're going to continue to solve all these problems by integrating the long tail. When you solve most of the scene problems, there will be some strange situations from time to time, so there must be a system to solve the training and solve these strange scenes

This means that the commercialization of autonomous driving will also enter a new stage. Since this year, although the epidemic has had an impact on the automobile industry in the short term, it also shows the industry room for vigorous development of "unmanned" economy. According to the standard of the international society of automatic engineers, automatic driving is divided into six levels: l0 to L5. At present, L4 technology has been implemented, and L5 level automatic driving is the real realization of automatic driving in all road conditions.

Previously, unmanned cleaning vehicles, unmanned logistics vehicles, and self driving loading trucks have been launched. In the future, giants will accelerate the research and development of automatic driving in all road conditions. But whether it's Didi, Google or Uber, what needs to be strengthened now is the mileage test. Unmanned driving still needs to face extreme conditions such as bad weather and traffic accidents.

Combination of software and hardware

In recent years, car makers, including alphabet's waymo and Uber, as well as technology companies, have invested billions of dollars in the autonomous driving industry. In China, Baidu and didi travel are also actively arranged. After Changsha announced in April that driverless taxis would operate in the city. Recently, Beijing, Shanghai and Guangzhou have also announced that the driverless online car Hailing has been officially put into operation.

It will take time for the public to fully trust the technology to be ready for autonomous driving. The company's current focus on creating a subset of this real world driving complexity is therefore very automatic. Moreover, he believes that these problems can be solved based on the current hardware version of Tesla vehicles. In the future, Tesla vehicles can realize L5 level automatic driving only through software upgrade.

This also means that more scene data is needed for automatic driving training, because the real three-dimensional road conditions and traffic conditions will be very complex. From the point of view of liberating driving technology and popularizing automatic driving technology as soon as possible. In comparison, the perfect combination of hardware and software is also a big challenge.

In Musk's view, this is just one of Tesla's strengths. As early as 2018, Tesla began to try to develop its own chip. In April 2019, Tesla released the autopilot 3.0 system, which is equipped with Tesla FSD (full self driving computer) autopilot chip.

It is reported that the modified self-developed FSD chip is manufactured by 14nm process, and the operation speed is increased from 200 frames / second to 2000 frames / second. "At present, a dedicated autopilot chip has been developed for point product calculation. If we use traditional TPUs, CPUs, or other similar products that consume hundreds of watts of power, the trunk will be occupied by a huge cooling system, which will be costly, bulky and energy efficient. " The reason why musk talked about self-developed chips at the conference was that there was no system with reasonable cost and low power consumption in the market.

In the automatic driving system with complex software application, energy consumption is very important for vehicle mileage. Musk said the chip has an eight bit accelerator for motor calculation. Taking full advantage of Tesla's fully automated driving system may take at least a year.

In terms of AI algorithm and training system, muska said that AI system is mainly limited by chip heat output and communication rate, so Tesla is developing new communication and cooling systems for developing more efficient computers to process video data more effectively.

At present, Tesla has established an unmanned driving related engineering team in China.

Cautious attitude

Tesla's progress in the field of unmanned driving is also inseparable from the support of 5g technology. Just last week, 3GPP officially announced a freeze on 5g R16. In the first year of development, 3G will be responsible for the development of a series of 5g standards, which will be responsible for the development of a series of 3G and 5g standards.

This has become another important milestone in the history of 5g development, which will undoubtedly accelerate the development of many Internet of things applications including the Internet of vehicles. In the domestic market, driverless driving has also been warming up rapidly in the past two months. Baidu has carried out the operation of driverless taxis in Changsha, autox and Wenyuan Zhixing have opened the test to the public by taking a taxi from Gaode, and Volvo has jointly launched a trial run of self driving taxis.

With the support of the policy, at present, the domestic cities that have opened the permission of manned test are Guangzhou, Changsha, Shanghai, Wuhan, Cangzhou and Beijing. In Shanghai alone, a total of 20 car companies, including SAIC and BMW, have obtained 78 test licenses for its vehicles.

However, in the view of many industry insiders interviewed by the 21st century economic report, safety and prudence are still the most important beliefs, which can not be achieved overnight or blindly accelerated. Dr. Feng Shaoqian believes that the safety of unmanned aerial vehicles (UAVs) in the world will be more secure when they are equipped with standard unmanned aerial vehicles (UAVs). Compared with automatic flight aircraft navigation, unmanned vehicle navigation is a worldwide technical problem.

In comparison, the automobile has higher requirements for positioning accuracy. When the plane is about to land, the accuracy requirement is still meter level. For intelligent driving vehicles, the lane is very narrow, and the distance between the obstacles on the side of the road is very short, then the vehicle positioning accuracy requirements are 10 to 30 cm. In addition, the car driving on the road may be blocked by the surrounding trees, trucks, urban buildings, and even be interfered by other radio signals.

"It is difficult for human beings to judge centimeter accuracy, but the machine needs to judge whether the received navigation information is credible, available and safe. It takes sex. " Feng Shaojun revealed that the perfect spatiotemporal intelligent service will soon cover the high-end mainstream models, making unmanned driving more intelligent, safe and earlier.

?