In May, one of the well-deserved star companies on the self-driving track, Xiaoma Zhixing officially obtained the license for the driverless road test of the California DMV, becoming the eighth company in the state to obtain such a license.
At the end of the year, Xiaoma Zhixing ushered in a not so moving news.
From the official website of the California Motor Vehicle Administration (DMV), as of November 19, only about 6 months of pony Zhixing appeared in the list of driverless test licenses, and the figure has disappeared.
Only seven companies currently have driverless test licenses in the California DMV, including Waymo, Cruise and Apollo.
In response to this matter, the new intelligent driving asked Xiaoma Zhixing for verification, and the official response was: Xiaoma Zhixing took the initiative to suspend all unmanned tests, conducted a technical review, and communicated with the regulator in a timely manner in a responsible and cautious manner.
However, another voice appeared on the Internet: according to the California Traffic Management Authority (DMV) statement quoted by the US Consumer News and Business Channel (CNBC), autoevolution, THE TERGE and other media, the original text of the statement should be: "On November 19, the California Traffic Authority notified Xiaoma Zhixing that according to the report of the separate collision accident in Fremont, California, on October 28, its fully unmanned autonomous driving test license was suspended and effective immediately."
As for which is true, Tan Qing said that AI is non-committal, and no further response from Xiaoma Zhixing has been searched on the Internet.
As the head company in the track, Xiaoma Zhixing took a step back at the end of the year, is the driverless real fire or a virtual fire? After a few years of fighting, the problem seems to be back to square one again.
Pony Wise Walk, Not Good 2021
This year is praised by some industry insiders as the first year of autonomous driving, not only large-scale automatic (unmanned) driving road tests began to gradually erupt this year, the main unmanned truck business tucson in the future also successfully listed, known as the "first share of automatic driving" for a while.
Three unpleasant messages, but tied to the pony, the star of the track 2021.
According to foreign media reports, on June 25 this year, the CEO of Xiaoma Zhixing told Reuters that the company is considering listing in the United States to help it achieve its goal of commercializing driverless ride services. In July, its CEO Peng Jun conducted a roadshow to raise $1 billion to $1.5 billion.
At the same time, Pony Chi Hong also announced that it has hired Lawrence Steyn, vice chairman of JPMorgan Chase Investment Bank, as the company's first chief financial officer.
However, the subsequent news has thrown a layer of fog over the listing, and Reuters reported on August 11 that Xiaoma Zhixing has shelved its plan to list in the United States through SPAC at a valuation of $12 billion.
Although the listing plan is shelved, the information of Tianyancha shows that as a star enterprise in the industry, the financing rhythm of Xiaoma Zhixing in the past two years is still quite powerful. Since the angel round in 17 years, there have been sequoia and IDG blessings, and since five years of development, there have been easy financing of hundreds of millions of yuan every year, especially the B round led by Toyota in February 2020, which directly injected $462 million into Xiaoma Zhixing.
Although the "surplus food" of the pony may still be surplus, the research and development of automatic driving is a bottomless pit, and with the listing plan shelved, the other two unpleasant things that have occurred in succession may affect the future financial situation of the pony Zhixing.
First of all, in terms of business structure, Xiaoma Zhixing has drawn manpower from within and officially released the independent brand of truck business "Xiaoma Zhika" in March this year.
After a series of energy and capital injections such as intensive supply chain preparation, license processing, and recruitment of talents in the following months, according to the "Late Post" report, in the second half of this year, Xiaoma Zhixing has encountered important business adjustments and the departure of key technical personnel.
Not only the truck business contracted, the autonomous driving research and development team was merged into the passenger car research and development team, and the technical leader Pan Zhenhao, the head of the domestic automatic driving PnC group, Sun Haowen and other executives also left their jobs or started another business.
In addition, in terms of technology, Xiaoma Zhixing has been licensed by the California DMV unmanned road test to withdraw since May this year, and there have been two traffic accidents, although compared to the "road test black hole" with frequent accidents like Waymo, it seems that it is not worth mentioning, but it is not easy to get the DMV unmanned road test license, and it is undoubtedly difficult for Xiaoma Zhixing to take the initiative to quit, which is undoubtedly difficult not to cause people's various speculations.
An autonomous driving industry analyst told us, "Xiaoma Zhixing attributed this withdrawal to some extent to the principle of 'adhering to the principle of safety first', but this may also cause outside questions about the technology of its company and even the entire industry at this stage." ”
Although the industry aura of Xiaoma Zhixing is still shining, the listing is shelved, the backbone leaves, and the active withdrawal from the license list, this is undoubtedly a bleak fragment that is difficult to smooth out in 2021.
As a result, the 2021 autonomous driving track seems to give us such a contradictory impression, and everything seems to be within reach under the road test explosion, but it is out of reach.
Internal force: Why does Robin Li think that "L5 may not be realized in decades"?
Robin Li, chairman and CEO of Baidu, once admitted in a speech, "L5 is too difficult to achieve for decades." ”
As for why Robin Li made this seemingly pessimistic judgment, Xiao Jianxiong, founder of autoX, a driverless startup, said a fundamental reason, "The premise of all commercialization of Robotaxi is the truly unmanned ability of security." Because this fundamentally hinders the popularization of L4 and above technology in terms of law, ethics and social acceptance.
We can draw a clearer picture of this from the October 28 accident that occurred in California.
According to the public information on the official website of the California DMC, Xiaoma Zhixing made the following report on the accident:
"On October 28, 2021, after turning right from Cushing Pkwy Road into Fremont Avenue, Pony's self-driving car changed lanes to the left in automatic mode. While changing lanes, the vehicle collided with the middle barrier on Fremont Avenue and the traffic signs on the barrier. ”
(Source: Google Map Drafting: Talking about AI)
"The core technology of autonomous driving is based on three steps: perception, decision-making, and execution." Liu Guang (pseudonym), an autonomous driving test engineer, told us, "The accident may occur because the Xiaoma Zhixing road test vehicle has a perception disorder on objects such as lane change lines or isolation belts, resulting in wrong driving decisions, or the inability to make accurate cognition of obstacles in decision-making, thus causing accidents in the final execution." ”
Let's look at a staggering number of DMC self-driving car crash reports, in California alone, in the nearly 11 months that ended November 21 this year, a total of 89 accident reports were submitted, an average of about one accident report every three days.
Of these, there were 43 Roadmo-related traffic accidents, 25 on Cruise, and 4 at Apple. According to Cruise's latest accident report, at 5:22 p.m. and 5:35 p.m. on December 1, two accidents occurred in less than 20 minutes for its test vehicles.
We often say that computers can show the advantage of several dimensions higher than people in speed and accuracy, but judging by the shocking amount of accidents, this seems to be a "joke".
Tan Qing said that AI believes that for L4 and above technical levels, the "safe unmanned ability" required for the current large-scale landing is mainly constrained by two factors: hardware capabilities and algorithm capabilities. We still analyze around the three basic elements of perception, decision-making, and execution.
The first is hardware capabilities, which are mainly reflected at the sensor level.
Sensors for unmanned driving technology, like the driver's eyes and ears, mainly undertake the perception task, such as the current ultrasonic radar, millimeter wave radar, lidar, cameras and infrared radar and other hardware equipment are to undertake this task.
Want to achieve unmanned driving, you need to pull the perception ability full, unfortunately, now there is no sensor can meet this requirement, even if most of the current players put lidar, high-precision cameras and other hardware combination on the car to make up for the short, can not guarantee that the saved car, can be 24 hours a day to achieve full perception ability to achieve full score.
The specific reasons are mainly related to factors such as visual blind spots, climate and weather, etc., and will not be repeated here.
Next is the computing power, which is like the driver's brain for unmanned driving, mainly undertaking decision-making tasks. For human drivers, such as encountering a dog on the road, first use their eyes to perceive, and then let the brain recognize, so as to make decisions and plan the trajectory that needs to be changed to avoid collisions.
Liu Guang told us, "Although the recognition of known objects is getting higher and higher, for example, we teach the obstacles we know one by one to the computer, maybe we can't think of only one percent of the objects, but the randomness of the driving scene is too large, from the perspective of the current mode efficiency, the difficulty of classifying and identifying this one percent is not linear, very complex and inefficient." ”
From this, we revisit what happened to Pony Chi Heng in October, like a "low-level mistake" on the teeth of a tired driver who slammed his car on the road.
It is not difficult to find that if it is difficult for humans to avoid accidents such as ghost probes, which is the hard-core bottleneck in the true sense of bicycle intelligence, then a large number of "low-level mistakes" in the current road test announced by the California DMV undoubtedly mean that in terms of internal force, bicycle intelligent technology still has considerable room for progress. After making up this part of the space, the accident rate can be greatly reduced compared to this year.
External force: Mushroom car federation began to overtake in curves?
In May 2021, as one of the earliest players to enter the autonomous driving track, Baidu Robin Li clarified for the first time three business models of Apollo: one is to provide Apollo autonomous driving technology solutions for host manufacturers, the other is Baidu car manufacturing, end-to-end integration of Baidu's automatic driving innovation, and the third is to share unmanned vehicles.
At the same time, another team of autonomous driving, Car-Road Synergy, has also begun to "shout" to the traditional bicycle intelligent players, and the most representative of this year is undoubtedly the Mushroom Car Union, which received strategic financing from Tencent and JD.com in September.
Seeing the current high incidence of bicycle intelligent accidents, large-scale profitability can not see the dawn, Apollo gradually no longer put eggs in a basket, the voice of vehicle-road coordination began to be high, does this mean that bicycle intelligence is about to usher in a rout?
Tan Qing said that AI believes that whether it is the diversified development of Apollo or the beginning of vehicle-road collaboration, we still cannot deny the value of bicycle intelligence, mainly because the problems faced by the current autonomous driving track are not short-term, and the relationship between bicycle intelligence and vehicle-road collaboration is more reflected in cooperation rather than confrontation.
Specifically, in fact, as Xiaoma Zhixing CTO Lou Tiancheng said, "Bicycle intelligence is the ability of every soldier in the army is very strong, and vehicle-road coordination is more of an army command system, and the two promote each other and are not contradictory." ”
Taking this as a starting point to further look at the relationship between the two paths, we cannot deny that in order to achieve the real sense of unmanned driving, bicycle intelligence will eventually face very big problems in perception and decision-making, for these problems, vehicle-road collaboration is an auxiliary solution with great value.
The important thing is that the vehicle-road collaboration is also the same, "road" is a good solution to break through the final perception bottleneck of bicycle intelligence, and the intelligence of single "car" is the basis for ensuring that the concept of vehicle-road collaboration can be established, which is very prominently reflected in the decision-making level of bicycle intelligence.
Specific explanation We use rock climbing as an example, for the sport of rock climbing, the mainstream gameplay usually requires two types of equipment, namely protective equipment such as main ropes and seat belts, and auxiliary equipment such as rock nails and hanging pieces.
The bicycle intelligence of autonomous driving is undoubtedly a protective device, that is, all information processing needs to be based on the decision-making layer as the brain to carry out cognition. Vehicle-road collaboration is more like auxiliary equipment, allowing the "Sky Eye" to break through the limitations of the car, thereby providing the brain with a richer and more comprehensive dynamic real-time information interaction.
The technologies of both are based on the main line of perception, decision-making, and execution, and are therefore interdependent, and only when combined can they exert greater potential energy.
Just like in the road test accident record, Waymo, Cruise and other hard-hit areas, as well as the pony Zhixing directly dried the car to the road teeth, as the top echelon of automatic driving players, many of the current unmanned road test accidents, can not completely throw the pot to the bottleneck of visual limitations, which means that protective equipment has not yet been equipped, bicycle intelligence is far from ushering in the end.
Therefore, Tan Qing said that AI believes that under the premise that there is still a lot of room for progress in the intelligent internal force of bicycles, the external force of developing vehicle-road coordination is not the primary problem at present.
For the car-road collaboration players, entering the world of automatic driving at this time node does not mean that it will be large-scale profitable faster than bicycle intelligence, there is no mature technical reserve of bicycle intelligence, and players who take this route are facing a cold winter at the beginning.
From an objective point of view, the basic disk of bicycle intelligent technology will be the core competitiveness of Xiaoma Zhixing in the future, but the vehicle-road cooperation players can survive the cold winter, and when the real technical bottleneck of bicycle intelligent technology arrives, the technical reserve may provide it with a first-mover advantage to promote the landing of the entire L4 and above technology.
Perhaps cooperation is the optimal solution for whether L4 and above technology can be landed in the future.
Ps: Tan Qing said AI, turn left new energy vehicles, turn right smart driving, there is depth, there is temperature, the author Zheng driving, reprint please retain copyright information.