This article is transferred from the public number to the door venture,Original address

When Karl Benz invented the automobile and mankind entered the automobile age, science and technology gave a new definition of human "travel". With the continuous development and advancement of technology, mankind has a new vision of intelligent driving. Look forward to and hope.So,How does intelligent driving originate, nurture, develop, and erupt? What kind of inspiration can we get from it? This article will review this history and explore ways to develop emerging strategic technologies and industries.


At present, artificial intelligence is hot, and startups are springing up. Practitioners began to think about how to make technology a ripple effect and promote non-linear and transitional growth of the industry.

Some people compare the relationship between artificial intelligence and industry to "raisins and bread." Although raisins are still raisins, the combination of the two can create new high-value categories.

In recent years, the author has been exploring the industrial opportunities of artificial intelligence, and concluded that in the future 15 years, intelligent driving will be the most value-added industry brought by artificial intelligence, and there is no one.

First, activate, reshape, and create multiple Trillion market.

  • Activating the automotive market, the new experience of smart, safe and human-machine driving will rekindle the need for people to change cars;
  • Reinventing the travel market, driverless + shared cars will solve the biggest problems that are now plaguing consumers and travel service providers – driver costs and “bad guys” risks. If the current network car only solves 2% travel, then the future unmanned taxi can increase this ratio by several times;
  • Created a new consumer economy and productivity market – the passenger economy. Passengers on the road or spending, or working, or entertaining, each car can become a mobile commercial real estate.

Secondly, to solve many social problems that humans have been unable to solve since entering the automobile society-traffic congestion (and exhaust emissions caused by idling), frequent accidents, difficult parking, etc.Unmanned driving is like an "old driver" with hundreds of billions of kilometers of driving experience and millions of years of driving experience. It can fundamentally solve the above problems without fatigue, road anger, drinking and driving, not arbitrarily putting on traffic jams, or worrying about parking. Really satisfy the people's yearning for a better life.

The development of intelligent driving can be divided into 4 stages:

2004 years ago;

2004-2009: the first 6 years-conception;

2010-2015: the second 6 years-growth;

2016-2021: the third 6 years-flowering;

2022-2027: The fourth 6 years-the result.

2004 years ago - the past life of autonomous driving

1921 8, the first unmanned (actually remote) car was born in the United StatesAn electronic engineer from the US Army sat in a car behind and used the radio to control the steering wheel, clutch and brake of the preceding drone.

At the 1939 New York World's Fair, GM's "Future World" exhibition predicted that the 1960 highway will have an electronic track that will work with the car's autopilot system to achieve driverlessness and switch back to the driver until the highway exits. drive.

Since then, GM has not used this prediction as a play, but in the 1956 year, the Firebird II was exhibited. This concept car, which looks like a "rocket", has the first automatic navigation system in history. Two years later, when the Firebird III came out, the BBC broadcast live broadcast of unmanned driving based on road-way collaboration. The pre-embedded cable on the highway and the receiver at the car communicated via electronic pulse signals, demonstrating the future of the highway. Driving style.

In fact, the prototype of truly independent autonomous driving – Shakey, appeared in the 20 century 60 era, was born at the Stanford Research Institute, which was later renamed SRI International to invent Known as the computer mouse and voice assistant Siri, another important contribution of it is the robot.

Shakey was the first robot to have full awareness, planning and control capabilities (this is the general framework for later robots and unmanned vehicles). The father of Shakey is the scientific geek Charles Rosen and the founder of Stanford International Research Institute. The singular media made Shakey’s propaganda beyond his actual ability and even sensationalism. The last episode of the doomsday made the scientists quite embarrassed, and this is the first time that the artificial intelligence community has established a relationship with the media. It was repeated many times.

If Shakey is just a robot moving indoors, then "Stanford Cart" is the first robot to be close to a driverless car. Hans Moravec is hailed as “the strongest supporter of artificial intelligence”. Under his leadership, “Stanford Car” has made great progress. Moravik's team developed a number of new technologies, such as calculating the depth of the scene with a single camera, and later Mobileye used a similar technique. In most cases, the “Stanford Car” needs to be manipulated by remote images. Once it escapes control and goes straight into the busy road, when Moravik sees a real vehicle from the monitor from the “Stanford Car” The whistling and screaming, and was shocked, so the pursuit of "defective robots" became a witty stroke in the history of unmanned vehicles.

Moravik suffered many setbacks in the exploration of machine vision, and later proposed the famous Moravec's Paradox, the high-level intelligence of human beings, such as reasoning, planning and playing chess. achieve. And only a few months old babies can drive low-level intelligence, such as perception and movement, the computer is out of reach. In an era when deep learning is still in the midst of shackles, scientists have yet to find a clue.

In the 80 era of the last century, the KITT self-driving car in the TV series "Knight Rider" was all the rage. Almost at the same time, Japan, Germany and the United States, the automakers, really started the development of self-driving cars. Japan’s Tsukuba Engineering Research Laboratory, the German Munich Defense Forces University and Mercedes joint team, the US Defense Advanced Research Projects Agency (DARPA) and Carnegie Mellon University, respectively, supplemented by “cameras and other sensors Developed prototypes of different self-driving cars and demonstrated compelling capabilities in real-life conditions.

In particular, NavLab of Carnegie Mellon University completed the “No Hands” trip from Pittsburgh to San Diego in 1995, where 98.2% of the mileage was completed by driverless, although the vehicle was not fast, even if it was placed Today, such results are still very remarkable. The unmanned vehicle that later entered the "Robot Hall of Fame" was based on the Pontiac Trans Sport Minivan (small utility vehicle), the main reason is that Minivan can plug in more equipment than the car. Later, Waymo also adopted Fiat Chrysler's Minivan “Pacifica” as the basis for the modification of the unmanned vehicle.

Another innovation at the end of 90 came from VisLab, a visual laboratory at the University of Parma in Italy. They used a stereo vision system consisting of binocular cameras to achieve long-distance 2000 kilometers on the highway, with unmanned drivers accounting for 94%, and speed. It reached 112 km/h.

Almost at the same time, Chinese academic and industrial circles have begun to explore intelligent driving.At Tsinghua University, 1978's research group Qi Guoguang began to study autonomous driving. In 1986, Professor He Kezhong's HTMR team relay, to HTMR-III, really had a prototype that was close to autonomous vehicles.

The first self-driving car in China was the ATB-90 (Autonomous Test Bed-1) in the early 1s. It was jointly researched by Beijing Institute of Technology, Nanjing University of Science and Technology, National University of Defense Technology, Tsinghua University and Zhejiang University. Then ATB -2 The speed is 3 – 4 times higher than that of the first generation, and most of these colleges have become the cradle of unmanned driving talents in China.

Also in the 90 era, Professor Wang Feiyue from the Institute of Automation of the Chinese Academy of Sciences began research on unmanned vehicles in the United States.

Similar to the United States, China has also explored remote-controlled driving earlier. 1980 National Project “Remote-controlled anti-nuclear reconnaissance vehicle”, Harbin Institute of Technology, Shenyang Institute of Automation and National University of Defense Technology participated in the research. In the year before the second phase (2003), the Red Flag CA7460 of the National University of Defense Technology and FAW realized the autopilot demonstration of the expressway with a peak speed of 170 km/h and automatic overtaking.

2004 year ~ 2009 year - the first 6 year: gestation

The big event in 2004 was the Grand Defence Challenge of the Defense Advanced Research Projects Agency (DARPA). At the beginning of the "Second Gulf War", the Ministry of National Defense noticed the casualties of soldiers in the Desert Operation and hoped to use drones to solve the problem.

The DARPA Challenge is a fine tradition in the United States. Congress has earmarked funds to discover transformative, high-return research results through the Challenge, greatly reducing the gap between basic scientific discovery and military applications. The 3 Unmanned Challenge, the 1 Robotics Challenge, and the 2018 Launch Challenge make it famous.

The Challenge required the unmanned vehicle to successfully cross the 240 km desert road. Unsurprisingly, all 2004 teams were smashed in the desert. This made the subsequent 2005 Challenge a glory.

The Red team of Carnegie Mellon University is the favorite to win the championship, and its leader, robotics expert Red Whittaker (Red Whittaker) is determined to win. He believes that driverless driving cannot be achieved just by hard work. "If you haven't done everything, you haven't done a thing." means that you have to know everything before you can be successful. You only know something. Aspect equals zero. This also indirectly indicates the high threshold of unmanned driving.

Among the teams, Stanford's “Stanley” unmanned vehicle is not eye-catching, but the leader Sebastian Thrun is the winner. He is the robot SLAM (synchronous positioning and map creation). A pioneer in technology, previously flirted from Carnegie Mellon University, trying to regain dignity in this game. The traditional top three for driverless cars are Carnegie Mellon University, Stanford University and the Massachusetts Institute of Technology, but among the challengers is a young man from the University of California at Berkeley, Anthony Lewandowski (Anthony). Levandowski), a tall, 2L-meter, maverick young man with a motorcycle called the "Evil Knight", sucked his eye.

The two cars at Carnegie Mellon University were leading the way, and the inexplicable failure in the second half caused the two cars to slow down significantly, only to get the second and third place. "Stanley" Although there were several accidents in the game, it didn't matter. After deleting some insignificant code, it went faster and faster, and finally won the 200 million dollar prize. It was not until 12 years later that Carnegie Mellon’s failure came to the surface. It turned out that a filter between the engine control module and the nozzle broke down, causing the engine to lose its power. “A thousand miles of embankment, collapsed in the ant hole”, there must be a 12-point fear of unmanned driving. Tron later sighed that Stanford could win, purely random.

In this competition, many vehicles used laser radar, high-precision geographic information system and inertial navigation system, which are still the standard configuration of many unmanned vehicles. Of course, the laser radar at that time can be said to be strange. The laser radars made by the Hall brothers are as big as the washbasins. These two brothers are the owners of the audio store and the fans of the "fighting robots". To the research of laser radar, the pioneer of the field of laser radar - Velodyne.

The first time I contacted an unmanned vehicle was in 2005, when Gary Bradski of Intel Research Institute (OpenCV Father) Helped Tron's team to improve their visual ability. He urged Intel's marketing department to sponsor the Stanford team. At that time, Intel had spent 10 million dollars to sponsor Carnegie Mellon University, so Tron gave a friend price. 2 million dollars, Intel fortunately won the bet of this final champion. Interestingly, since the "Stanley" body has been covered with various sponsor's trademarks, Intel's logo can only be attached to the front window, which is a very prominent position, and indicates that this is an unmanned vehicle (because there is no The driver looks through the front window to see the rearview mirror).

In the blink of an eye of 2007, DARPA has been dissatisfied with the unmanned driving in the wilderness and started the “Urban Challenge”. Carnegie Mellon University is making a comeback, this time they are fully prepared to form a team of 40 people, including general Chris Urmson. In addition to the two participating vehicles, there is also a supply vehicle that provides ample replacement of parts. The Winkack of Carnegie Mellon University finally won the crown. It is said that this time Carnegie Mellon University invested so much that it still did not fill the deficit after winning the 200 million dollar prize. In their equipment library, for the first time, a new type of 64 line laser radar appeared. In order to put this equipment into use, Carnegie Mellon engineers wrote a large number of drivers. The Hall Brothers' Velodyne provided this super weapon, from the size of the washbasin to the size of the flower pot, condensing a lot of their efforts. In the following 10 years, 64 line laser radar became the component that most unmanned vehicles in the world must configure.

The two challenges greatly boosted the confidence of the scientific research and also cultivated a large number of talents. It is said that Google's founder, Larry Page, is a geek. He and Tron became close friends because of his interest in robots. For unmanned driving, Page has a new idea. He recruited Tron to Google, first on Google Street View, and in 2009, he secretly set up the unmanned car project "Chauffeur" and gathered a group of famous players in the challenge. Including the aforementioned Ulmson and Lewandowski.

Amnon Shashua is a visual expert who belongs to the Massachusetts Institute of Technology and is a roommate of Tron during his academic vacation at Stanford. As a professor at Hebrew University, he founded Mobileye, the first pioneer to try to commercialize ADAS (Advanced Driver Assistance Systems) technology. Mobileye was founded in 1999 years, and by 2009 year, it has gone through the hard journey of “from 0 to 1” and has installed its products in several models. At the beginning of the establishment, no one thought that it had been ringing the clock until 2014 year. What is even more unexpected is that 2017 was acquired by Intel in the year of 18. In this XNUMX year, it came out of a path that few people took.

DARPA's Unmanned Car Challenge motivated its Chinese counterparts. In 2009, under the support of the National Natural Science Foundation of China's “Cognitive Computing of Audiovisual Information”, the first China “Smart Vehicle Future Challenge” contest was held in Xi'an, which opened the curtain of the China Series Challenge.

2010 Year ~ 2015 Year - Second 6 Year: Growth

In 2010, Tron founded Google X as the founder, and here, countless “Moonshot” projects are scrambling.

Projects must meet 3 criteria:

  1. Benefit hundreds of millions of users
  2. Looks a little sci-fi
  3. Can be achieved within a few years with today's technology

There is no doubt that driverlessness meets these conditions.

Google's first unmanned vehicle was retrofitted on a hybrid Prius with 64 line lidar on top to create a high-resolution 3D environment model or high-resolution map. These test cars are disguised as street-view data collection vehicles, often at night to avoid the public's sight, and can also collect high-precision maps on roads where no one has no cars. Even if they are very low-key, they are inevitably caught by the traffic police, James Kuffner (James KuffNerIt was one of the first engineers to be dug up to Google from the Carnegie Mellon team. Now he is the leader of the Toyota unmanned car, and he can recall the scene that was stopped by the traffic police. “The paper can't hold the fire,” the famous journalist John Markov digs from a high school classmate who tests the driver to the shocking information, and exposes it in the New York Times, which makes the “Car City” Detroit deep. Deep shocked. In the end, Neihua reached the first state in the United States to allow unmanned vehicles to go on the road.

Among the core backbone of Google's unmanned vehicles, there is the owner of the "Evil Knight" Lewandowski. The young man who is deeply appreciated by Pecs is a troublemaker who despises the rules and even disapproves of security. He led Google to purchase technology and components from several companies such as 510 SYSTEMS, and later they realized that these companies were operated by Lewandowski himself. Page showed great tolerance to him, not only for the heavy money, but even for buying 510 SYSTEMS.

Google's second-generation unmanned car is the more powerful Lexus, also a hybrid. As mentioned above, the first requirement of the unmanned vehicle's basic model is that it is large and can be equipped with various equipment. The second requirement is electronic control, because the underlying control algorithm of the engine is much more difficult than the motor, and most teams are more willing. Put time on the high-level algorithm.

But the real goal of the world is the birth of Google's third-generation unmanned car "Firefly" in 2014. This koala-like car is completely redesigned for driverless driving, such as removing the wiper. Because there is no need for drivers to see the road in the rain. According to the design, the car does not have a steering wheel, but due to the restrictions of California law, a joystick is installed in the car as a steering wheel. The car later won the Red Dot Design Award.

At the same time, Mobileye won the trust of the car manufacturer, and the visually-oriented ADAS low-cost solution entered the mainstream market. By the time of 2015, the installed capacity had reached nearly 10 million units. Mobileye also secretly started the development of autonomous driving. Mobileye's vision-based approach is unique compared to Google's approach. For example, it uses a visual map, and the map extracted from the visual is very small (only 10 kb data per kilometer, compared to Google is GB level), suitable for real-time uploading, and updated by crowdsourcing. In fact, vision-based positioning is closer to the way humans drive. We evaluate the approximate location based on the signs on the road and make real-time decisions based on changes in the road line (which lane is chosen, whether it is on the ramp, etc.). Then, you only need to extract those signs and lines from the visual, and the crowdsourcing is uploaded to the map, and you can obtain the positioning through visual matching when driving.

In fact, there have been several major events in 2015.

First of all, at the beginning of the year, Mercedes-Benz's unmanned concept car F015 made a stunning appearance at CES, and suddenly presented the unmanned vehicle to the public.

2 At the beginning of the month, the news broke the application of Uber. From Carnegie Mellon University and its affiliated National Robotics Research Center, many 50 scientists and engineers were dug up to build their own unmanned car R&D team. It is said that Uber's founder Travis Kalanick, after taking the Google's unmanned vehicle (Google is Uber's investor), is both excited and scared to think it is a disruptive technology for the industry. However, it will "removing one's own life", so there is a big move ahead.

The most straightforward "the future has come" is the release of Autopilot by 10 Monts Tesla. Although Autopilot is an L2-class assisted driver, many common owners are misled by this name. The three courageous drivers opened the Autopilot mode and completed the crossing of the US and East Coasts, with an average speed of 84 km/h. Of course, there was a danger in this process, and Tesla did not agree, thus laying a hidden danger for the later accident.

2010 Year - This stage of 2015 Year, China is slightly quiet.

In 2010, the aforementioned VisLab four self-driving cars departed from Parma, Italy, and traveled 9 countries and 1.3 kilometers to Shanghai, China. The relationship between VisLab and China has not ended. Later, some unmanned youth troops in China have visited and studied at VisLab. In 2015, a Chinese-era visual chip company, Anba, acquired VisLab.

2011 7 month, the red flag HQ3 driverless car independently developed by the technical team of Professor He Hangen of the National University of Defense Technology, completed the high-speed full-range driverless test from Changsha to Wuhan 286 km for the first time, in which the artificial driving mileage was less than 1%, and compared The previous generation of CA7460 has made significant progress in hardware miniaturization, control accuracy and stability. Based on this, the National University of Defense Technology also won the championship of the "Smart Vehicle Future Challenge" competition. After that, the team of Academician Li Deyi became a frequent visitor to the championship (except for 2013, which was acquired by Beijing Institute of Technology, and the main player will be Dr. Jiang Yan, CTO of the company).

In the second half of 2015, there are three events worth recalling.

  1. During the 8 month, the team of Yutong and Li Deyi teamed up to complete the 33 km unmanned driving of Zhengkai Expressway, which pioneered the world of unmanned buses.
  2. The 11 7 "Smart Vehicle Future Challenge" contest was successfully held in Changshu. The challenge was reported by CCTV News Network. Unmanned driving became the talk of the general public after dinner.
  3. In December, Baidu launched the annual blockbuster of unmanned vehicles. The unmanned vehicle that Baidu and BMW cooperated with the G12 "high-speed-fifth ring-Olympic Forest Park" route went back and forth, attracting countless eyeballs.For some people in this project, this demo is the end, and then they left Baidu to start a new journey.For Baidu, this is the beginning. The autonomous driving department was formally established. Wang Jin is in charge. It is known as "Three years of commercial use, five years of mass production", "If the auto industry does not destroy itself, it will be destroyed by others." ".

These three events made the people realize thatIn the highly sophisticated field of driverless driving, China is not lacking.

2015 year is the eve of the outbreak.

2016 ~ 2021-the third 6 years: flowering

Wu Xiaobo wrote in "The Thirty Years of Surprise":

When this era comes, it is impossible to be. Everything grows arbitrarily, dust and twilight rise, rivers gather into Sichuan, and the nameless hill rises to the peak. The world is incomparably open.. " 

It is more appropriate to use this passage to describe the beginning of the 2016 year.2016 Year is the unmanned "Spring Equinox" season.

The author resigned in 2016 in 2. Many people, including venture capital (VC), are suspicious of business models. In the 3 month, several major events broke out. AlphaGo's five-game victory over Li Shishi ignited the public's enthusiasm for artificial intelligence. GM bought XemonX billion dollars and only several prototypes and 10's Cruise Automation. VC also realized that the era of driverlessness is coming. In China, Beijing’s spring auto show, Chang’an and Bosch and Tsinghua’s several unmanned vehicles “40 kilometers into Beijing”,Unmanned driving also truly enters the Chinese public.

4 month peaks and turns, Intel high-profile claims to bet on the smart driving field, the author as an old Intel employee, but also doing unmanned entrepreneurship, is also very concerned about this. In the 5 month, Intel hired a consulting company to find the author, and I hope to offer advice on Intel's strategy. My suggestion is simple, to acquire Mobileye. After 1, Intel announced the acquisition of Mobileye in 153 billion dollars (not necessarily causal relationship with my suggestion), representing the giants of this PC era officially entered the field.

The spring is chilly, and several accidents make people wonder.

  • In the month of 2, Google’s unmanned car crashed into the bus. This was the first accident to voluntarily acknowledge responsibility, but the slight collision did not cause too much accusation. One of the lessons learned later was that the bus driver could not afford it.
  • In the month of 5, Tesla's first fatal car accident took the headlines. The deceased was a driver, Tesla's enthusiast. At the time the vehicle was running in the high-speed Autopilot mode, the driver was watching the video, completely ignoring the responsibility of keeping an eye on the road. The Autopilot system did not detect that a large truck was crossing the road. The vehicle was drilled from the belly of the truck at a very high speed and the driver died on the spot.

Even though Mobileye vision failed to recognize the lateral side of the white trailer, the forward-looking radar missed the target due to the low installation position. The public began to question: Is this beta version of hardware and software allowed to go on the road? Is the software upgraded to be re-inspected? On the other hand, Autopilot was mistakenly promoted as autonomous driving, but in essence it was still assisted driving.

After 1 months, in the media and industry, Tesla sent a blog to defend himself. Autopilot completed 7 billion miles of autopilot miles in 1.3 months before the accident, while American human drivers averaged 9400 10,000 miles in a fatal car accident, isn't Autopilot safe enough? However, Tesla did not count a similar car accident in China, because in that case, Autopilot's data dropped to a fatal crash every 6500 million miles. Since then, how much accident-free mileage is considered safe, and it has become an unsolved question in front of the industry.

The accident also led to the "breakup" between Tesla and Mobileye. In addition to the responsibility of the accident, there is another important reason. Tesla's ambition to develop computer vision independently touched the core interests of Mobileye. After several rounds of mutual accusation, Tesla first announced Bosch's millimeter-wave radar as the main sensor. By 10, it officially announced that the Autopilot hardware version 2.0 (HW2) uses its own vision system. Maybe it’s Musk’s “first-principles” (people can drive by vision, and drones are the same), maybe Tesla’s promotion, they claim that HW2 has the ability to drive fully, buy a new HW2 Just spend 3000 dollars and you will be able to drive unmanned through software upgrades in the future. The irony is that after 2 years, 2018 years 10 month, Tesla deleted this option from the brochure. In the past two years, Tesla has experienced a lot of talent loss. Many evaluations show that HW2.x has basically reached the level of 1 after the launch of 1.0. But full control of 2.x hardware and software still gives Tesla a lot of data.

One of the big things in 2016's 8 month was Uber's $10 million acquisition of Otto, a truck-driven automaker. When Karannik set up a research and development center in Pittsburgh, he had high expectations for the unmanned entry into commercialization. However, the progress of 6.8 was not satisfactory. The acquisition of Otto was overweight again, but he did not expect it to become Overwhelming his last straw. The founder of Otto was the original Lewandowski who worked for Google and was unruly. Page's tolerance did not make him feel good, and the satisfaction of Google's huge bonuses gradually subsided. The big company's red tape and carefulness made him feel "dragon shoal", so at the beginning of 1, he pulled a bunch of people. Come out and set up a separate portal. Silicon Valley is quite tolerant of “betrayal”, but given the scarcity of the technology of unmanned driving, Google may sign some informal terms of competition with the leavers. Most of the leavers who created the new company deliberately avoided direct competition. . For example, Zhu Jiajun and Dave Ferguson's Nuro do logistics, while Otto is positioned as an unmanned truck. Of course, any term is limited, and the 2016 month of the acquisition happens to be after Lewandowski gets Google’s last compensation.

Google has realized the loss of unmanned talent, and changing organization and incentives is imminent.

2016 12 Month, Waymo was split from the Alphabet matrix as an independent company. This new name became the first master of the unmanned field overnight. Prior to 3 months, John Krafcik became the new leader of the team. He was a car company executive (pre-modern North American CEO) and led an Internet company (similar to used cars). The veterans of the trading website) brought different styles and strategies to Waymo, and they also brought conflicts. In the 1 month before Krafsick’s entry, the second-generation leader Ulmson after Tron also left the company, and Waymo really entered a new era.

If 2016 is "spring equinox" and 2017 is "rainwater".

There is plenty of rain and everything is recovering. Many companies are struggling. Big companies, whether they are technology giants or OEMs, are really investing resources. At the same time, 2017 is a year in which startups have entered. Another important sign is that unmanned driving is not just a passenger car, but also a variety of commercial vehicles, special vehicles, in addition to manned, logistics has become a larger market.

The International Consumer Electronics Show (CES) in 1 is a weather vane.The hottest topic of the year is autonomous driving.Almost every company in the LVCC North Hall exhibited the concept of autonomous driving, and the North Square was a real-time driving experience. I also launched the concept car “Urban Mobile Space”, which has 360 degree dead angle sensor coverage and L4-class autopilot design without steering wheel throttle brake, especially the unique internal ring sofa layout. The VIP lounge on the road" concept. Many veteran car factory executives, engineers and designers stopped at the front of the car. One of them lamented that they had wanted to do this in the OEM factory. Unfortunately, there is no room for free play.

Alex Roy, guest editor of Time Portal's car portal, said on the podcast: "When I looked at the car, I thought it was the car that Faraday should have built.The car won the Red Dot Design Award in 2017. In the history of Red Dot, there were three unmanned car winners, the aforementioned Google "Firefly", Mercedes F015 and the same year BMW i-inside.

In 4, Intel bought Mobileye for 153 billion. This action is long overdue. In the past 1, Intel was running all the time, but CEO Brian Krzanich had no patience to catch up. The acquisition of Mobileye was the front row ticket. The best choice. 153 billion dollars, according to traditional financial indicators, this price is high, according to the valuation of ADAS company must also be high, but if you look at the imagination of the automatic driving faucet, it does not seem high. Just X months later, Morgan Stanley analysts gave Waymo an 2 billion dollar valuation.

The integration of Intel and Mobileye took nearly a year, involving the balance between the United States and Israel, the old and new forces, but "do not experience the storm, how to see the rainbow", 18 reloaded after the year, Intel has become the most in this arena Important player.

The integration of different cultures requires leadership courage and compromise, as well as trust and authorization.In the past year, General Motors has also been undercurrents. How should Cruise's L1 autonomous driving new force and SuperCruise's Level 4 autonomous driving product team get along?The management led by Mary Barra has used the highest political wisdom. Detroit belongs to Detroit and San Francisco belongs to San Francisco. The Cruise team has obtained a high degree of autonomy. While the personnel is developing rapidly, it is trying to retain Silicon Valley’s Entrepreneurship culture; On the other hand, General Motors also provided automotive engineering capabilities that Silicon Valley did not possess. The two complemented each other's strengths, enabling Cruise to quickly demonstrate its superb level on the busy streets of San Francisco and become the fastest-moving follower after Waymo.For Cruise, San Francisco is the best home stadium to show strength. It claims that compared to the Arizona cities of Waymo and Uber (of course, Waymo is not just in Arizona), the complexity of San Francisco has increased dozens of times.

Ford's "second home" in the US auto market is not to be left behind. Ford has never forgotten the glory of the past (family personnel are still in high positions), it is also the first car company to start contacting Google, but the arrogance of the Internet company makes the negotiations unhappy. In 2016, Ford introduced the 2021 Automated Driving Manifesto – the unmanned commercial operation of 2021. At the beginning of 2017, there was a big deal of investing in Argo AI with 10 billion dollars. Argo AI, the startup that has just been established for several months, has some of Google, Uber and some of the early challengers.

However, for the established giants, the conversion of new and old kinetic energy is struggling. New businesses require long-term and huge investment. Once the old business is in a growth dilemma, the leaders will bear pressure from two aspects.Ford old CEO Mark Fields abdicated sadly, but still handed the scepter to the head of the autonomous driving and mobility department-Jim Hackett.Hackett cools down "2021" after taking office, but this can be understood as "management stakeholder expectations". For the team, "2021" is still a goal worth striving for, the goals of Ford and Argo AI. Lock in Miami first.Compared with San Francisco, the general choice, the traffic situation in Miami is more complicated, it needs to cope with many tourists, more rain, and often "water overflows with golden mountains."

There are still many veteran giants who are also coping with the pain of the transformation of old and new kinetic energy. ZF ZF, a component supplier, experienced a major change in leadership after a series of investments and acquisitions (a successful acquisition of laser radar supplier Ibeo formed a sniper against rival Valeo). In the car circle, “divide” and “combination” become popular. The “points” can be lightly loaded to meet the new four (new energy, sharing, intelligent, networked), enabling faster decision making and easier financing.

The typical case is that Delphi splits out of Amberford and focuses on intelligent networked cars (almost at the same time, the acquisition of the startup NuTonomy by 4.5 billion dollars). Ford also removed Ford Autonomous Vehicles LLC. On the other hand, it is also a good idea to "combine" the enemy as a friend, warm up the group and share the research and development costs.

So, in this arena, everyone stood in line and quickly formed different alliances. such as,

  • Intel/Mobileye, Ambo, BMW, and later joined the mainland, Fiat Chrysler and so on.
  • NVIDIA, Bosch, ZF, Volkswagen/Audi, Volvo, etc. are another circle.
  • Travel service provider Uber has a circle of friends from Daimler, Volvo and Toyota.
  • The "second child" Lyft also has GM, Anbofu, Jaguar Land Rover and other partners.

In terms of the size of the alliance members, Baidu Apollo Ecology can be said to be the biggest circle of friends: 2017 3 month, Lu Qi entered the smart driving division, causing another core talent to leave. It is said that at this time, an engineer of Baidu American Research Institute established open source, Baidu leadership quickly showed great courage. At the Shanghai Auto Show in 4 month, Lu Qi announced the "Apollo" plan to be Android in the automotive industry.

The Apollo mission to the moon means that it is starting from the universe of artificial intelligence. "I hope that in the future, I can free my hands so that everyone can freely look up at the stars when driving." 

A stone provokes a thousand waves, and the whole industry is shaken. At the AI ​​Developers Conference in 7, Li Yanhong took a Suzhou license car that cooperated with Bosch, showed autonomous driving technology on the Fifth Ring, and received a traffic police ticket, but this did not obscure the sensation caused by the announcement of Apollo 1.0. Everyone began to realize that Baidu is serious.

Apollo's opening is worthy of the industry's thumbs up, and it has made great contributions to active ecology, data sharing, and training of talents. But it also caused a lot of problems:

  • The success of Android is after a successful iOS, there is no successful unmanned iOS, is it too early to do Android?
  • Android is the second strategy of Google's voice, because Android is the carrier of mobile to enhance its core search and advertising business, and Apollo's own business model has not yet emerged, and today's core business of Baidu has not been organically combined, as a "burning money" Can the business be supported by shareholders and go on for a long time?
  • The prosperity of Apollo's ecology lies not in the generosity of Baidu, but in whether the members of the ecological community can also fully invest, especially the “old guns” of the automobile industry. Does it agree with the path of “recoding data with data”?
  • Apollo's open source, is a blessing for a startup? Especially those startups of Baidu?

Undoubtedly, for start-ups, Apollo lowered the threshold for making presentations, but at the same time it raised the threshold for bigger and stronger, and must be differentiated from Apollo's technology to survive. The differentiation of many startups is verticalization, scenario and accelerated commercialization, from truck logistics to terminal distribution, from manned, cargo to load functions (such as sanitation cleaning), from mines to ports, from parks to airports and Last 1 kilometers.

In 2017, there are three commercialization events that are worth mentioning.

  • First, the 4 month's ferry between the company and Baiyun Airport between the terminal building and the parking lot is the first open driverless operation in China. Although it is only one week, it is essentially different from the demonstration. If the demonstration is The time is set, the route is specified, and the operation is for the end user and the open environment, full-time work.
  • Secondly, the 6 month's ferry service and the Kaide Group's ferry service in the Raffles underground parking lot in Hangzhou is the first normal operation of several unmanned vehicles in China for several months. The open passenger car environment and narrow lanes The positioning without GPS is a technical highlight.
  • The third event was the Alfaba in Shenzhen at the end of the year. The 4 modified bus showed a good ability on the designed bus route, which was the result of cooperation between the startup and the university. Unexpectedly, there are a lot of "shocking" articles, which is exactly the same as the one that Rosen encountered on Shakey decades ago. In the process of technology evolution, the media and the public’s expectations are managed. It is important.

Of course, the most meaningful event of 2017 year occurred in the United States.

In the middle of 10, Waymo announced that autonomous vehicles without front-end safety drivers have started trial operation.

For a big company that values ​​security very much, it takes a lot of courage and absolute confidence in technology. Of course, to ensure safety, Waymo still has security officers in the back seat for the unexpected. At the beginning of 2018, the California Vehicle Authority further announced that it would allow “there is no need for a safety officer in the car, only a remote security officer.” This huge leap is believed to be related to the confidence that Waymo brings.

The US’s attitude toward unmanned driving, from Obama to the Trump administration, from the Senate to the House of Representatives, from the federal to the state government, has a very high consensus – the United States wants to be a leader.

Two Transportation Ministers Anthony Foxx and Zhao Xiaolan continue to promote the "Automotive Driving Federal Policy", "Autopilot 2.0: Safety Vision" and "Preparing for Future Traffic: Automated Driving 3.0", added to the legal space Exemption, loosened for the industry.

Almost at the same time, Germany also introduced the first law related to autonomous vehicles - the Eighth Amendment to the Road Traffic Act.It allows the automatic driving system to replace human driving under certain conditions, and the world's first automatic driving ethics came into being. These legislative activities cleared the way for the traffic jam Pilot of the world's first L3-class autonomous driving product, the Audi 2018 A8.

China has also been exploring the establishment of unmanned legislation and testing systems.As early as 2016, the road test specification was discussed at the national level. However, the first one was announced in Beijing, 2017 12 month, the Beijing Municipal Traffic Commission, the Beijing Municipal Public Security Traffic Management Bureau, and the Beijing Economic and Information Commission. Formulated and issued the "Guiding Opinions on Accelerating the Road Testing of Auto-driving Vehicles in Beijing (Trial)" and the "Implementation Rules for the Implementation of Road Test Management for Self-driving Vehicles in Beijing (Trial)", just like a flat thunder, let the end of the year The industry is uplifting. Since the first card has landed, the dominoes will not stop. In the first half of 2018, Shanghai, Chongqing, Shenzhen, Guangzhou and other places have launched local road testing policies and guidelines. 4 Month 11 Day, the Ministry of Industry and Information Technology, the Ministry of Public Security and the Ministry of Transport jointly launched the "Intelligent Network Linked Vehicle Road Test Management Specification (Trial)", which was hammered at the national level.

Considering that China's complex road conditions have higher requirements for safety, the domestic road test specifications require test subjects to conduct certain mileage tests in the closed test field. As early as 2016 in 6, the first “National Intelligent Networked Vehicle (Shanghai) Pilot Demonstration Zone” closed test area approved by the Ministry of Industry and Information Technology was opened in Jiading. Subsequently, the national layout of "5+2" was formed. Today, all localities are still constructing or renovating the test site of intelligent networked vehicles. Although there are problems with repeated construction in the short term, in the long run, in the future, there will be great demand for unmanned vehicles whether they are listed or inspected annually. .

After 2017's "rainwater", 2018 may be "shocked" in the years, both commercialized with rumbling springs and cold springs.

The beginning of the 2018 year made some people happy and some people sad. Velodyne cut the price and dropped it by half. For many unmanned companies in 2017, one of the most painful things is that you can't buy a laser radar. It will take several months, when you finally bought it at the end of the year, and when you have a part of the goods, the laser radar on the market has already The price has dropped.

At this time, the Lidar stadium is not the only one out of Velodyne. Valeo's Scala has achieved the first mass production project on the Audi A8. Traditional OEMs and supplier giants have invested in mergers and acquisitions. Only Delphi (Ampford) is betting on the 3 home. . Almost all companies bet on solid-state or semi-solid laser radars. In addition to Quanergy and Innoviz, which were already hot in the previous two years, some startups (such as Luminar, Sagitar and Innovusion) also demonstrated better performance prototypes. Two years ago, Quanergy had some troubles in mass production. Although it still leads the optical phased array technology, solid-state laser based on MEMS micro-mirror, optical two-dimensional galvanometer and Flash technology. The radar has shown faster progress in industrialization, Innoviz has received orders from BMW, and Velodyne's new velarray seems to be coming first.

In the past 2017 year, data released by the California Motor Vehicle Authority (DMV) showed that based on 35's 10,000-mile test mile base, Waymo implemented a manual intervention every 5596 miles, followed by Cruise, every 1254 Miles for a manual intervention. Compared to every 2016 miles in 5000 years, Waymo only increased 2017% in 10 year, which is slightly disappointing. Looking at Waymo's data, the positive factor is that the number of interventions per intervention in the last few months of 2017 has been greatly improved. Whether 2018 will continue this trend will be announced soon.

When Waymo's engineering director gave a lecture at the Massachusetts Institute of Technology, he said something very profound:

“When you are 90% done, you still have 90% to go (when you think you have completed 90%, you only have 10%)

Waymo has a deep understanding of the arduousness and long-term nature of this road. However, this company is very good at "being a word." Waymo's "Early Rider Project" has taken another big step forward. In the 2018 year, Waymo removed security officers from some of Arizona's vehicles, and early passengers began to truly “exclusively” unmanned vehicles.

2018 3 Month, Waymo signed an agreement with Jaguar Land Rover to customize 2 10,000 unmanned vehicles. Waymo's triumphant enthusiasm brought tremendous pressure on Cruise. At this point, GM made a huge decision to push Cruise to the capital market, using external capital and resources to accelerate development. 5 Month 31 Day, Softbank announced that it will invest 22.5 billion dollars in Cruise. Just one day later, Waymo responded by raising the number of vehicles that purchased Fiat Chrysler to 6.2. In the month of 8, Morgan Stanley pushed Waymo's valuation to 1750 billion, of which the robotic self-driving taxi business was valued at 800 billion and the automated logistics service was valued at 900 billion. On 10's 3 day, Honda further injected $10 billion into Cruise, and pushed Cruise's valuation to 27.5 billion. GM took Cruise at the original 146 billion, and never thought that this part of the valuation has reached the GM's total market value of 10/1 after two and a half years. Lenovo believes that Intel will buy Mobileye for 3 billion dollars, so don't be surprised.

In this context, although the whole year of 2018 is the winter of capital, there is still news of unmanned company financing. Both startups and venture capital (VC) split into two camps.

  • One is the "rocket faction" of Silicon Valley Fan, whose theoretical basis is that since drones are on the moon, they will pick up their sleeves and build rockets. Since the general direction of the future is to travel, the operation of driverless taxis will be carried out in one step. Some people commented that this is "there is no Waymo's life, but it has got Waymo's illness." There is no second company in the world, like Waymo, who buys 8.2 10,000 unmanned vehicles, but there are double-digit companies that bid on Waymo in business model. There is no Waymo's "rich dad". Pay by VC. As everyone knows, even Waymo's 8.2 10,000 cars have limited ability to acquire data, and the 20 and 30 roads are clean and the data is not rich and diverse. This means that Waymo's L4-level commercialization path has scalability issues.
  • The other camp is a pragmatic approach, starting from vertical subdivision, "rural surrounding cities."In terms of racing, if you want to be the first to cross the finish line, you have to finish the race, even from the pit lane.However, in the eyes of the "Rocket School", this is the "Ladder School". If you want to go to the moon, you must build a ladder first. Being pragmatic is pragmatic, but the ceiling in the future is too low.There are a lot of "diss" routes in the industry. Waymo said I want to fly and learn how to jump all day. Some "tall" funds believe that you have evolved into a cactus for survival in the dry season. When the rainy season comes, You can only be a taller, fatter cactus at best, and you can no longer grow into a towering tree.Although these statements are a bit biased, there are indeed some vertical segmentation scenarios that are not compatible with open road L4 autonomous driving, and because of the limited market size, it is impossible to obtain a large amount of data required for algorithm sublimation.

It seems that both paths have problems with data availability.So how much data do you need, or how many miles to prove security?

In terms of driverlessness, Waymo has accumulated the most mileage, and 2018 has accumulated 10 million miles in 1000. Even with L2-class autopilot, we said earlier that Tesla's self-defense in 2016 year, 1.3 billion miles, 2 life accident, the data is not enough. The famous American think tank Rand Company gave a mathematical model. If you want to prove that the driverless driving is 20% better than people, you need 110 billion miles. That means 100 cars, 1 days 24 hours, 1 years 365 days running, running 500 years.

One of Tesla's revelations is to learn to rely on the user's car to get data and verify algorithms. If there are 1000 cars, 1 cars only need to run 1100 miles, 110 billion miles.

Therefore, a more reasonable path is to use rocket technology to make various aircraft.Then use the money and data of the aircraft to upgrade the rocket technology. Specifically, L4-class/L3-class commercialization scenarios (various aircraft) with defined boundaries are used to reduce the dimensions to the L4-class/L4-level commercialization scenarios (various aircraft) with defined boundaries, and these scenarios are deployed on a large scale to obtain cash flow and After a large amount of data, it further broke the limitations of the open road LXNUMX level.驭势科技 adopts such a strategyIn the L3 class of the expressway, the last 3 km micro-circulation L4 class, parking lot autonomous parking and airport unmanned logistics trailers have achieved commercial breakthroughs.In particular, it is worth mentioning that the company and SAIC-GM-Wuling have realized the world's first end-user delivery of autonomous parking, one-click remote parking and one-button parking.Similar technologies have also been used in the time-sharing lease with Shougang Gofun, allowing users to achieve automatic pick-up and return-to-ship, while operators can reduce operating costs through unmanned formation scheduling between stations. The deployment of these scenarios brought a large amount of traffic scene data, thus feeding back the evolution of the open road L4-class autopilot algorithm.

The "grey rhinoceros" of 2018 was an accident. When the entire industry enters the deep water area, the accident has become a high probability risk. There have been many incidents in Waymo, Uber and Tesla, and both have had fatal accidents.

Since Uber acquired Otto, some changes have quietly occurred. Lewandowski's contempt for security has mutated the company's culture, and there is a slogan in the Uber office in San Francisco that is "safe third." Reuters later pointed out that after Uber's test car was changed to Volvo XC 90, the new modified design expanded the sensor's blind spot, while the towering 64 line laser radar changed the car's center of gravity structure, but 2017's one-sided rollover did not cause Too much attention. Since Levandowski left Uber, the attitude of the new CEO to autopilot began to blur.

According to Business Insider, the team is worried that the project will be cancelled, the rapid progress must be used to please the leaders, and the leadership must meet the requirements of smoothness, thus ignoring many security designs. These factors have accumulated, which eventually led to the fatal accident caused by the world's first driverless car on 3's 18 day (the few times Tesla's unmanned driving), a Uber drone driving at night. At the time, he killed a pedestrian who pushed the bicycle illegally across the road. Of course, there is the responsibility of the pedestrian himself, and the Uber safety driver's major responsibility (similar to Tesla's first fatal accident, but also watching the video), but Uber's own problems are not evasive, such as technically Ride disables the original car's automatic emergency braking system, misses the last second of security, and manages to reduce 2 personally to 1 individuals. After the accident, Uber suspended all tests, re-examined the safety design and management, and did not return to the road until the end of the year. The lessons were painful. Another news under the clouds was that Uber closed the autopilot truck division, which made the acquisition of Otto even more unpredictable.

What is indispensable for being unmanned is the awe of safety and respect for the basic laws of the industry.There have also been several accidents in Waymo, including a car accident caused by a safety officer falling asleep. The Information has successively exposed that Cruise and Waymo unmanned vehicles are still not ideal in real road conditions. An October article by Reuters also pointed out that Cruise still has a long way to go to reach L10 mass production.In this field, lightly talking about L4 mass production and neglecting safety and leap forward development will inevitably pay a price.

Waymo adopted a strategy of “going back to the second”. On 10's 30 Day, the California Vehicle Administration issued Waymo a completely unmanned test license to legally test unmanned vehicles without security officers on public roads in California. However, at the end of 11, after careful consideration, Waymo put the safety officer back into the driver's seat. At the same time, Waymo appointed Deborah Hersman, former chairman of the National Transportation Safety Board, as the chief security officer.

At the beginning of 12, on the outskirts of Phoenix, Arizona, a driverless taxi payment service called Waymo One officially began operations. Prior to this, Krafsik had talked about the difficulty as usual. He believed that the ubiquitous and omnipotent L5-class autonomous driving still had to wait for decades, hoping that the media and the public would lower their expectations.

Going back to this sentence "When you are 90% done, you still have 90% to go (when you think 90% is completed, it only takes 10%)", if today's technology and cost requirements can't solve the last 10 quickly % of the question, is it possible to solve it through the overall idea of ​​the human-car-environment? This is the concept of road coordination.

In 2018, China is nurturing a new infrastructure. Both Ali and Baidu have proposed the concept of “car-way coordination”. Based on the over-the-horizon sensing capability and high-reliability low-latency links brought by LTE-V2X and 5G, Part of the perception and decision-making ability is placed on the roadside, using the idea of ​​the edge cloud to solve the environmental and infrastructure problems.

Car-side computing is another important element in the value chain, especially chips. The main chip used in high-level driverless is either from NVIDIA or from Intel/Mobileye. Since the emergence of ZTE’s ban, the United States has also issued a list of business controls. For the Chinese industry, it is necessary to plan ahead. Domestic chip makers such as Huawei, Cambrian, and Horizon are accelerating the development of AI-accelerated chips for driverless use. As the unmanned algorithm is gradually fixed, the dedicated acceleration chip will play a more important role, and Tesla has adopted this strategy.

Another factor that cannot be ignored in the value chain is data. Unmanned vehicles need to be more intelligent and need data.The EU has always been ahead of the data legislation. The General Data Protection Regulations (GDPR) is said to be the most stringent in history, making countless Internet companies squandered. Data is a resource and a hot potato. This is true for Internet companies, but it is very tolerant to the car factory. In the last EU vote on the registration of autonomous vehicles, the 7A clause determined that “the data generated by autonomous vehicles is automatically generated, and its nature is not creative, so it does not apply to copyright protection or database rights”. This means that without the owner's consent, the depot can collect data generated by autonomous vehicles (remote sensing information, including GPS trajectory information) and sell it to third parties. This can be said to be the best assist for the car factory.

In China, the "Network Security Law of the People's Republic of China" requires important data not to leave the country. Foreign and joint venture car manufacturers cannot send data to foreign countries for research and development. Therefore, it is imperative to establish a research and development team and a research and development supply chain in China. This is a great opportunity for domestic technology companies and new supply chain players.

The 2018 year is coming to an end and there are 2021 years from 3. The industry's goal is to be able to achieve L4-class unmanned scaled applications in certain areas, such as a city area. From the current perspective, we can maintain cautious optimism.

From 2022 to 2027-the fourth 6 years: results

If the 3 6 year goal is successfully achieved, the 4 6 year will be the time of the open road 4 level unmanned driving.

The change brought by driverlessness is far more than the automobile industry. It will completely change travel and logistics and change the movement of atoms in this world.

By the fifth six years (5-6), most of the vehicles on the road will be driverless shared cars, and the number of cars will be reduced by more than half, but the utilization rate of cars will be greatly improved, and traffic jams will be a thing of the past. The sky returned to blue, parking spaces were converted into parks, activity spaces and residences, and car accidents were almost zero.

Traffic flow, information flow, and energy flow are all three, and all traffic related to people or things will be redefined. The insurance industry needs Nirvana rebirth, and the service industry will find new explosion points. - The above-mentioned driverless taxi is the third space in addition to home and office. It is a mobile commercial real estate, a mobile cinema, a mobile office space, and a mobile coffee shop.

Intelligent driving is an innovative product of artificial intelligence combined with traditional cars, and is the future of the automotive industry. As a transformative technology, smart driving is both technological innovation and social innovation. Whether it is law, regulations and policies, or moral and ethical debates, we must have the courage and patience to care for and guide their healthy development.

I am eager to see that the road is not blocked, the sky is blue, and the day of free travel arrives soon.

This article is from the 2018 12 monthly magazine of Artificial Intelligence magazine sponsored by CCID Research. The author is Wu Gansha.Zhang Yuxin. Original title: "The Evolution of Intelligent Driving: Dreams Come into Reality"