Original link: https://www.latepost.com/news/dj_detail?id=1783
The green light came on. The car automatically starts to pass through the intersection without any operation from the driver. The car ahead is too slow? Then it will automatically change lanes and pass. Pedestrians and cyclists suddenly appear? Slow down now. At the intersection ahead, you have to observe the situation of the straight vehicles coming from the opposite side, and then find the opportunity to turn left? Basically no problem. Humans just need to be in the driver’s seat, ready to take over when something goes wrong with the system.
This is no longer just a driverless car driving in a limited demonstration area, it is now also appearing in the bustling streets of Guangzhou, Shenzhen, Shanghai, Beijing and other cities. Car companies and their suppliers are intensively promoting the Navigate on Autopilot function called “City NOA” to more cities and more car owners.
The goal is more aggressive than the other, with this year as the time limit—Huawei announced that it will promote the urban NOA to 45 cities, Xiaopeng has 50, and Ideal has brought the Kaesong competition to three digits: 100.
Weilai, which once focused on polishing high-speed NOA, is now mobilizing an autonomous driving algorithm team of thousands of people to work overtime to keep up with its peers. It plans to expand the urban NOA function to dozens of cities in the second half of the year. In order to ensure this goal, “LatePost” learned that NIO even lowered the priority of the Firefly project for low-end and mid-terminal brands. “High-level officials believe that it is difficult to carry out two heavyweight projects at the same time.” A NIO person said.
Not only “new forces” such as Xiaopeng, Ideal, Weilai or Huawei are involved, but auto companies such as BYD, Great Wall and SAIC also plan to launch urban NOA functions on some models this year or next year.
“It’s like buying a building with 20 or 30 floors. It makes a difference whether it has an elevator or not.” Li Xiang, CEO of Ideal Auto, believes, “In the future, in the mid-to-high-end car market, if urban NOA cannot be provided, consumers will buy it or not. difference in buying.”
Xiaopeng’s recent situation is a proof. In the first month since the launch of the G6, 70% of the more than 40,000 orders have chosen the “Max version” with intelligent driving, which exceeded the expectations of most executives of Xiaopeng. Xiaopeng City NGP (Xpeng named the NOA function as NGP) function has been opened in Beijing not long ago, and the mileage penetration rate has reached 98%.
A race for fear of falling behind begins. In order to achieve the set research and development and development goals, the autonomous driving algorithm departments of various companies often work until the early hours of the morning. Due to too much work intensity, the turnover rate of relevant team members of some car companies remains high.
Many autonomous driving practitioners said that if the urban NOA function is to be promoted on a large scale, a large number of technical and engineering problems still need to be solved. A practitioner said: “Most companies are still in the bragging stage, and it is difficult to deliver on schedule.”
But this is also the only way for car companies to stand out from the competition and rush to a larger scale. At least every car company involved believes that: the first to come up with a better urban NOA experience is the key for them to stay at the table and enter the next stage of competition.
For drivers, urban NOA is also a test, and they must adapt to this new function that needs to “fight against their own inertia”: it makes people feel effortless and relaxed, but it is not truly unmanned after all. Once the driver is too relaxed, in the NOA systems may not be able to take over in time if they fail.
When major technological changes are often unstoppable in transforming everyday life, it is even more important to understand how the changes come about.
The technical route is changed, and the racing is pulled to a similar starting line
It is not too late for Chinese car companies to start developing autonomous driving functions. In 2013, Tesla began to develop Autopilot, an assisted automatic driving function. In less than two years, Xiaopeng Motors followed up and regarded “autonomous driving” as the soul of smart electric vehicles. After that, more car companies and suppliers invested in it.
However, the promotion of assisted driving functions by Chinese car companies is very slow. In the past 8 years, the leader in the Chinese market can only use advanced driver assistance (such as NOA) functions in 300,000 kilometers of high-speed roads and expressways in some cities, as well as in a limited number of cities. These areas account for only about 6% of all public roads in China.
And Tesla, which pioneered the self-developed autonomous driving route of auto companies, has tested the full-scenario NOA function in North America in 2021, eight years after its start, and officially launched it in the second year.
Since the beginning of this year, Chinese companies have changed their pace of jogging in the previous 8 years. Companies with advanced R&D progress, such as Xiaopeng, Huawei, and Ideal, have shown a surprising number of urban NOA development goals.
The turning point began in 2021, when Tesla detailed its self-driving “cheats” on AI Day: the Transformer-based BEV (Bird’s Eye View, bird’s-eye view) perception model. Transformer is a model architecture proposed by Google in 2017, which can handle large amounts of data and the relationship between data on a large scale. It is also the basis of the current large language model.
With the support of Transformer, BEV can synthesize road information (lane lines, curbs, etc.) collected by cameras in multiple directions, sensors, and other traffic participants (pedestrians, two-wheelers, cars, etc.) into a 360° A high-degree bird’s-eye view allows the automatic driving system to have a more comprehensive understanding of the surrounding environment of the vehicle-even how the surrounding cars will drive next, so as to plan driving routes and drive more safely. Now, Tesla’s advanced driver assistance system FSD based on this program has about 400,000 users.
Tesla introduced the Transformer-based BEV perception solution on AI Day 2021. Source: Tesla.
The biggest advantage of Tesla’s BEV solution attracting various car companies to follow up is that it does not need to use high-precision maps.
Collecting and using high-precision maps in advance is the mainstream practice when Chinese car companies promote advanced driving assistance solutions. Most of the NOA functions available on China’s high-speed roads now use high-definition maps.
High-precision maps have more information than ordinary navigation maps: in addition to basic information such as road width, length, curves, and slopes, there are road markings (lane lines, etc.), traffic lights, sidewalks, and how roads are connected The accuracy of the information generally reaches the centimeter level. It can reduce the difficulty of algorithm development and provide safety redundancy, allowing the automatic driving system to know whether the road ahead is uphill or downhill, whether there are curves, and whether there are buildings or pedestrians on the side of the road.
The disadvantages of high-precision maps include high collection costs, long time-consuming updates, and low frequency. According to a map dealer, the cost of collecting high-precision maps per kilometer is thousands of yuan, dozens of times that of ordinary navigation maps, and the fastest update can only reach once a quarter.
This is not a big problem on highways with few changes, but it is troublesome on urban roads. Jiang Rui, general manager of the Automobile Business Center of AutoNavi Maps, told LatePost that after collecting some urban roads, AutoNavi Maps found that “an average road of 10,000 kilometers may change by 3,000 kilometers after one year.”
It can be said that high-precision maps are one of the main reasons why it was difficult for advanced assisted driving to quickly expand from highways to urban roads in the past.
However, Tesla’s plan has removed the “sandbag” of high-precision maps for car companies, and the speed of opening a city has been greatly improved.
When the new technology brings a new acceleration, the players participating in the competition are drawn to a similar starting line.
The urban NOA functions launched by Xiaopeng and Huawei in Guangzhou, Shanghai, Shenzhen and other places are all based on high-precision maps. Xiaopeng only spent 250 million yuan to obtain the surveying and mapping qualification of high-precision maps. “It takes a long time to toss a small road in Shanghai,” Yu Chengdong said in April this year. Huawei has invested in Shanghai for one or two years and has not yet collected high-precision maps of the city.
In the second half of 2022, with the verification of Tesla’s technical route, Xiaopeng and Huawei have turned to develop BEV solutions that do not rely on high-precision maps, which requires rewriting some algorithms and training perception models.
And car companies that started to develop autonomous driving late, such as Ideal and BYD, have not had time to invest too much resources in high-precision map solutions. They regard this change in technological route as an opportunity to catch up, see the right time, and expand aggressively.
“One or two years ago, in NIO, Xiaopeng, and Ideal, everyone would rank Ideal (autonomous driving capability) last. Starting this year, this voice will gradually change.” Ideal car intelligent driving co-driver President Lang Xianpeng said this year. In the past half a year, Ideal has cooperated with the team of Tsinghua University to develop the BEV solution, and its intelligent driving team has also expanded to 700 people.
In June of this year, Han Bing, head of BYD’s intelligent driving business, also said at an event: “BEV is an opportunity for BYD to overtake on curves in high-end intelligent driving.” Driving-related software and hardware solutions.
Wu Xinzhou, vice president of Xiaopeng Autonomous Driving, emphasized the experience value of the pioneers: “If the ‘map (high-precision map)’ (plan) has never been done, and the ‘no map’ is sent directly, I think there is still risk in 100 cities. He said that Xiaopeng “saw an infinite number of pitfalls” when developing the urban NOA plan, and he believed that opponents of these pitfalls would have to step on it again.
But it cannot be denied that the change in technology route has narrowed the gap between Chinese car companies in urban NOA due to R&D sooner or later to a certain extent, and compressed the competition of product experience, launch time and the number of cities opened to a tighter time frame. The racing has become more intense.
Collect data, hoard computing power, and train larger models
If car companies want to replicate Tesla’s technical route, the premise is to train a usable BEV model. This is a process of using big data and big computing power to train larger models. According to the information released by Tesla in 2022, the Tesla FSD model using BEV + Transformer has 1 billion parameters, which is about 10 times that of the previous version of the model.
“The more training data, the better the results,” Tesla CEO Elon Musk (Elon Musk) talked about the impact of data on the autopilot system again at the July earnings conference, “Only one million training At two million samples, it (the model) barely works; at two million, it works a little bit; at three million, we marvel, as if we see something. Unbelievable.”
The confrontation of opinions, sales competition and resource competition around data, computing power and models are unfolding among auto companies.
A person from Ideal’s intelligent driving department said that the reason why Ideal dared to say that it will open 100 cities this year is because there are many cars on the road. As of the end of June this year, Ideal has sold nearly 400,000 vehicles and accumulated 600 million kilometers of data. Ideally, total sales this year could reach 360,000 units. When BYD Han Bing explained the opportunity of self-developed autonomous driving, he also specifically mentioned that “BYD is expected to accumulate 600 million kilometers of data this year.” This is equivalent to the sum of the mileage accumulated by Ideal in the 8 years since its establishment.
However, the amount of data should not only be based on sales volume, but also on how many of the cars sold have the ability to collect effective data. The above-mentioned ideal car people emphasized that BYD has sold a lot of cars, but the sensor configuration is different, and the effective data that can be reused may not be so much; and the “front face” design of the ideal L9, L8 and L7 models on sale is almost the same. One of the reasons is to ensure that the sensor position and height are the same, so that multiple models can collect and share a set of data.
This requires car companies to convince users to pay for hardware such as cameras, various lidars and expensive computing chips in advance.
Weilai’s approach is to equip all series with self-driving hardware as standard, even if this will increase the total car price and weaken short-term price competitiveness. Li Bin, CEO of Weilai, once said: “If the computing power is found to be insufficient later, there is no way to make up for it.”
Compared with Tesla, which develops its own chips and does not use lidar, NIO costs more. “The hardware cost for Tesla to realize advanced assisted driving is only 1,500 US dollars, and the cost of those of us who use dual Orin chips is basically more than 4,000 US dollars.” Li Xiang said this year. And Weilai’s solution has four Orin chips.
Car companies such as Xiaopeng, Ideal, and BYD use free software to attract more people to choose and install supporting hardware. Although they have formed autonomous driving software development teams with hundreds or even thousands of people, they do not expect to be like Tesla for the time being. Pull that way to make money directly from software. Huawei, which originally charged for software, has now begun to cut prices: In June this year, two months after the release of the advanced driver assistance system ADS 2.0 that supports urban NOA, Huawei announced a 50% discount on the buyout price, from the original 36,000 yuan to 18,000 yuan .
Just data is not enough. Training the BEV model is still a huge system engineering, and one of the key steps is to label the data. The quality of data annotation directly determines the quality of the model.
“Training a BEV model requires millions of clips of ‘multi-camera’ video, involving about a billion Labeled objects (lane lines, pedestrians, cars, etc.), according to the current manual labeling efficiency, it takes about 2,000 people to label for a year.”
Xiaopeng claims that its labeling system can improve efficiency by 45,000 times on the basis of manual labeling. Even so, Xiaopeng invested more people. “LatePost” learned that Xiaopeng’s data labeling outsourcing team has expanded to 1,000 people, which is 60% more than previous years.
Data labeling companies are getting more business, too. The data service company Beisai Technology received 20% more data labeling orders from car companies in the first half of this year than in previous years, and plans to continue to expand in the future.
The competition among car companies around computing power is very straightforward: buy more GPUs and build larger data centers.
In September last year, when the United States restricted Nvidia from selling high-performance GPUs to China, He Xiaopeng commented that the move would “bring challenges to all autonomous driving cloud training”, but “Xiaopeng has bought back the demand for the next few years in advance. A month before he said this, Xiaopeng and Alibaba Cloud announced the construction of a data center in Ulanqab, using 1,600 Nvidia A100 chips with a computing power of 600 PFLOPS.
Less than a year later, in June this year, Li Auto announced at a media event that its autonomous driving cloud computing power had reached 1,200 PFLOPS, twice that of the data center built by Xiaopeng last year. Ideal is also cooperating with Bytedance’s Volcano Engine and is building a new data center in Shanxi.
Xiaopeng, which claims to have “reserved for several years” chips, also continued to expand this year. “LatePost” learned that Xiaopeng has been discussing with Alibaba Cloud to expand the scale of the data center.
Kaesong! Kaesong!
Teams of thousands of people work overtime to reserve resources for longer-term competition. Whether it can be transformed into commercial results and competitiveness depends on how fast and how many people can actually use the city’s NOA function.
Huawei, Xiaopeng, and Ideal have given more aggressive landing promotion plans than one, which requires them to develop one by one like Internet companies do in-store group purchases, takeaways, and taxi services.
The difference is that Internet companies send a large number of promoters to attract merchants and drivers to the platform in these cities, while car companies need to rely on the sensors on the car and the car to collect information in advance that cannot be processed by the BEV model alone, such as being caught by a roundabout or a big bus. The situation of intersections heavily blocked by cars, or special road conditions in some cities.
The tediousness and complexity of this work is no less than that of local pushers trying to persuade merchants to settle on the platform.
Xiaopeng, who was the first to provide urban NOA functions in China, has a lot of experience. “There are some roads in Beijing where the non-motorized lanes are wider than the motorized ones. (Based on the BEV perception model alone) it is very difficult to judge which is the non-motorized lane. This is the most annoying point for autonomous driving engineers.” Xiaopeng Wu Xinzhou said. He also mentioned that there will be arrows on every road in Guangzhou, and there will be arrows in Beijing almost when turning at the intersection: “This is a challenge for pure visual solutions. Sometimes it is too late to see (arrows).”
“The important thing about Kaesong is to establish a rule system for this city,” Wu Xinzhou explained. “It will be a cloud file. When the car arrives in this city, it will read some things in the file and learn how to deal with specific places. drive.”
From last year to this year, due to technical variables and more intense competition, car companies are pursuing faster and faster development speeds, and their development ideas have also changed.
Xiaopeng and Huawei, which were the first to launch the urban NOA function, used to build a fleet or find a map dealer to collect high-precision map information of the city’s public roads in advance, and then test the system functions. The speed of Kaicheng often depends on the collection of high-precision maps. It took Xiaopeng and Huawei a year to open the urban NOA function in only a few cities.
In order to achieve the goal of opening dozens or even hundreds of cities in half a year, auto companies are exploring more efficient ways to open cities. In June of this year, Lang Xianpeng demonstrated how Li Auto does not rely on high-precision maps to obtain information—collecting road feature information at complex intersections in advance, training a model that specifically recognizes traffic lights, and helping the system work better.
Ideal also announced the idea of opening the city with “commuting NOA”: let ideal car owners help collect the above-mentioned road feature information to form a supplement to the BEV perception model, which is equivalent to “crowdfunding to open the city”. The general process is:
- Let car owners who are interested in trying the city NOA in advance set their own commuting route, and after several days of running back and forth, collect the necessary information on this route.
- The ideal engineer judges whether it is enough to open the NOA permission on this road based on the data collected by the car owner. “For more complex roads, it usually takes 1 to 2 weeks to collect data before it can be opened.” A person from Li Auto’s intelligent driving department said.
- As more and more car owners begin to collect their own commuting route information, the available range of ideal city NOA will expand accordingly.
Ideal will collect complex intersection features in advance to make up for the perception defects of the BEV model. The picture is from Ideal Car.
This kind of “open city” may no longer be the “global opening” of a certain city that everyone understood in the past, but refers to the fact that the city’s NOA can run first on the core roads of a city.
“This is the most efficient and marginally beneficial.” A person engaged in the research and development of autonomous driving said, “It is more important to capture 20% of the core data than the 80% of the long tail, which are frequently used by users.”
It is not just a matter of technology for car companies to open up cities for NOA. A product manager in charge of the city NOA function said that there is no large-scale public beta in the industry, and the process is still relatively vague, but what is certain is: “The sales will be specially trained, and the precautions for the test drive will be updated. The team definitely has to step in and evaluate.”
In Kaesong, the promotion of city NOA by car companies has become a comprehensive competition involving multiple dimensions including technology, sales, after-sales, service and government communication. The speed and quantity of opening cities will directly affect the usage of users and the scale of data collected by car companies, and the scale of data will further determine the evolution speed of NOA in cities by car companies. Companies that run ahead at this critical point in time may run faster and faster in the future.
Supplier competition has also become more intense
When car companies such as Ideal, Xiaopeng, and Weilai announced plans to develop urban NOA cities, high-precision map suppliers also followed suit: AutoNavi and Tencent will cover 50 cities by the end of this year. Announced that 120 cities have passed the review, and plans to cover 150 cities next year.
Map dealers’ products are no longer high-precision maps that car companies regard as cumbersome, but maps with fewer elements and faster update frequency. AutoNavi calls it HQ Live MAP, and Tencent calls it HD Air lightweight high-precision map. For data, NavInfo simply calls it an advanced driving assistance map.
“The stronger the car’s perception ability, the weaker the role of the map. There is no doubt about it. But no matter how strong the car’s perception is, if the car wants to drive autonomously, it still needs a map.” Jiang Rui, general manager of the Automobile Business Center of AutoNavi, believes, A dynamic map that can make up for the limitations of vehicle perception, provide a global perspective, and can be updated quickly will be the ultimate solution for assisted driving maps.
AutoNavi’s HQ Live MAP, which began to be developed in 2020, pays more attention to road logic information that is difficult for cars to recognize only with sensors. The elements collected on urban roads have been reduced from 67 in traditional high-precision maps to more than 20. Reduced from the original 10 cm to 1 meter. Jiang Rui said that with fewer elements, lower accuracy, and the information collected by a large number of logistics vehicles and other special vehicles, AutoNavi can update the map as quickly as possible every day.
Meng Qingxin, CMO of NavInfo, believes that car companies will eventually compete for “cost-effectiveness” in assisted driving solutions. Using maps with more information elements can reduce hardware costs. “When crossing an intersection, a plan with pictures must be more accurate”, and “How to manage data, how to desensitize, and how to comply with regulations are also the advantages of NavInfo.”
It is still difficult to determine whether there will be high-precision map suppliers in the large-scale promotion of urban NOA. “We have to look at the new products of the graphic dealer before making a decision,” Wu Xinzhou said, “We don’t want to just throw away a pair of crutches and pick up another pair of smaller crutches.”
HD map suppliers don’t want to be left behind by car companies that are committed to urban NOA, and other types of suppliers are trying to find ways to take this express train as much as possible.
In the past year, the chip company Horizon and a number of autonomous driving technology suppliers released an urban NOA solution based on the Journey 5 chip. A self-driving practitioner said that in order to win a city NOA order from a car company, Horizon will even help the car company optimize its Orin chip-based solution.
“LatePost” learned that in order to get a city NOA designation from a well-known company, an autonomous driving supplier did not hesitate to accept the possibility of being replaced after one year. Its customers have stated on many occasions that they will develop by themselves.
DJI Vehicles, as always, emphasizes low prices. In April this year, it released a solution that only uses 9 cameras and 80 TOPS (60% of the computing power of Tesla’s FSD solution) computing power, claiming that it can achieve urban NOA on fixed urban roads. Function.
Momenta, a supplier of autonomous driving, recruited the backbone of digital technology from Zheku, a chip company that was disbanded not long ago, and wanted to make a more competitive solution that integrates software and hardware.
For these solution providers, this race may be even more brutal. Their opponents are not only their peers, but also the car companies themselves who want to master the core capabilities of autonomous driving.
Features You Shouldn’t Race For
In June this year, we experienced Li Auto’s urban NOA function. Starting from Wangjing Kuntai Hotel, we traveled 40 kilometers to Li Auto Shunyi R&D Headquarters, passing 77 intersections, and drove for a total of 90 minutes with an average speed of 26 kilometers per hour. It is a little higher than the speed limit for battery cars in Beijing.
“LatePost” learned that Ideal has been testing this route for 1 to 2 months, but during our experience, the ideal staff took over twice to avoid the starting buses and trucks. At the end of a road with a large number of cones arranged along two lanes, the ideal assisted automatic driving system braked twice, stopped for about 10 seconds, and then drove slowly.
The ideal L9 with the urban NOA function turned on meets the cone brake.
“Our logic is still a little more conservative and not particularly aggressive, so we will set a relatively large safety range,” said a person from Lixiang’s autonomous driving department.
After more than a month, we also experienced the assisted driving functions on Xiaopeng P7i and Weilai EC7. The NOA function that Xiaopeng can use on Beijing’s Second Ring Road can be described as “wild”. Changing lanes and overtaking is straightforward, and the driver doesn’t even have time to react.
They emphasize mature capabilities in their promotion, such as lane keeping and automatic cruise control. In our trial, some unimaginable troubles appeared: when the vehicle was about to pass the intersection, the vehicle suddenly turned right and turned into a retrograde, and almost hit a normal vehicle. ; On the side of the bus lane, the system suddenly manipulated the steering wheel to change lanes to the left, forcing the vehicle coming from behind to stop. Once the driver does not react in time, it is an accident.
Even Tesla, which has a longer test time, a longer test mileage, and a theoretically safer system, will emphasize safety based on statistics and probability in its propaganda, while ignoring human inertia: once you get used to relying on the assisted driving system , people tend to let their guard down.
In 2013, Google launched AutoPilot, a semi-autonomous driving system, and gave it to several long-distance commuting employees for testing. These employees are required to keep their eyes on the road and remain vigilant while the system is working, responding in the event of an accident. But they quickly forgot about the road conditions and used computers, slept and put on makeup in the car at a speed of nearly 90 kilometers per hour. The test lasted only a few weeks before stopping and never resumed, and Google then went all out to develop a fully driverless driving system.
The so-called semi-autonomous driving seems to be that the car is driving by itself, but in fact it needs to be taken over by people at any time, and it is difficult for people to maintain focus. The result of being impatient is often that users credulously believe the propaganda of the car manufacturer, which eventually leads to bad results.
According to statistics from the Tesla Deaths website, as of the end of July, 38 people around the world have died because of Tesla’s assisted driving function. When Chinese car companies promoted assisted driving functions, fatal accidents also occurred. In these accidents, it is difficult for car companies that provide assisted automatic driving functions to be held accountable. They will all emphasize in advance that when the assisted driving system is turned on, the driver is responsible for the accident.
Car safety is directly related to people’s lives, and it needs to go through a lot of rigorous safety verifications such as collision, braking performance, and handling stability before delivery. An automotive engineer said that it usually takes a year and a half for safety verification to make changes to a car that involve driving, such as changing the brake system components on a mass-produced car.
As an on-vehicle software system, urban NOA still lacks the constraints of unified standards to what extent it can be considered safe, and most of its iteration progress is controlled by car companies themselves. A person from the ideal intelligent driving department told “LatePost” that he is most worried about extreme scenarios that cannot be safely handled when the automatic driving function is promoted.
“It’s not like having technology can open (city) without limit,” Wu Xinzhou said.
Urban NOA will come, but it may not be as fast as car companies advertise, nor should it be so fast.
Source of the title picture: “Speed and Hurricane”
“TECH TUESDAY” series
In 1957, a man-made object entered space for the first time, orbiting the Earth for three weeks. Human beings can look up and see a small flash across the sky in the night, parallel to the stars in mythology.
Such a feat cut across races and ideologies, sparking joy across the globe. But not in the triumphant joy we might have guessed, touched by human feats. According to the political philosopher Hannah Arendt (Hannah Arendt) observed back then, people’s mood is closer to a long-awaited relief that science has finally caught up with expectations, “Humanity is finally on the way out of the cage of the earth. took the first step.”
People are always rapidly adjusting their expectations of the world based on technological exploration. When a fantasy of a science fiction writer becomes a reality, it is often that technology finally catches up with people’s expectations, or in Arendt’s words, “technology realizes and affirms that people’s dreams are neither crazy nor empty.”
At times like today, a little more dreaming is better.
This is also the expectation of “LatePost” launching the TECH TUESDAY column. We hope to regularly report on new scientific research and technological developments outside of the business world that “Later” focuses on on a daily basis.
These may be about the progress of a cutting-edge research, the observation of a technology application, or a tribute to some outstanding technologies or even an era.
This column will record the various changes in the world from the perspective of science and technology. During this journey, I hope readers can join us in gaining a little understanding of the world.
This article is transferred from: https://www.latepost.com/news/dj_detail?id=1783
This site is only for collection, and the copyright belongs to the original author.