*Author: Su Qingtao
On April 7, 2021, the Ministry of Industry and Information Technology (MIIT) of China released the “Guidelines for the Management of Intelligent Connected Vehicle Production Enterprises and Product Access” (Trial) (draft for comments). Article 1 states that “this guide is formulated for intelligent connected vehicle production enterprises and their products that have the capabilities of conditional autonomous driving and highly automated driving, which meet the requirements for access.” According to the “Automated Driving Classification” standard released by the Chinese MIIT in March last year, conditional autonomous driving refers to L3.
Article 6 of the draft states that “intelligent connected vehicle products should be able to automatically detect the failure of the driving automation system and whether it continues to meet the designed operating conditions, and take risk mitigation measures to achieve the minimum risk state”. Article 7 states that “in the case where the driver needs to participate in dynamic driving tasks, the driver’s ability to perform corresponding driving tasks should be evaluated”, which are both for L3.
Nowadays, many people in car companies and autonomous driving companies are talking about “L3 autonomous driving”, but perhaps few people have noticed that the so-called L3 is being (or has been) “redefined”.
On March 9, 2020, the MIIT website released the recommended national standard draft for the “Automated Driving Classification”, which specifies the classification of automotive driving automation functions. The deadline for comments was April 9, 2020, and it was scheduled to be implemented on January 1, 2021.
In May 2020, I wrote this article after carefully studying the regulations and policies related to “L3 autonomous driving” in the Chinese, Japanese, and Korean markets. Unfortunately, at that time, this article, which was a completely rational policy review, could not be published due to being “out of line with the mainstream” amidst widespread rumors of “L3 mass production year”.
In March of this year, there were once again misleading statements in the media regarding “L3” mass production by Honda.
Now, looking back, this article written one year ago is still relevant, especially as many of the information inside are still scarce in Chinese media. Therefore, I decided to publish it with some modifications.
Main viewpoints:
1. The “L3 autonomous driving capability” that will soon appear on mass-produced cars in China, Japan, and Korea is not the same as the “L3” under SAE standards. The biggest difference is the responsible party after an accident.
2. Faced with L3, the Japanese government’s attitude is extremely ambivalent – they want domestic manufacturers to seize the opportunity, but don’t want them to bear too much pressure. They encourage consumers to try boldly, but fear they may be too daring.
3. In the “Automated Driving Classification” of the Chinese MIIT, L4 with safety personnel or remote control is considered “L3”.4. Under the SAE standards, only Level 4 vehicles require “risk mitigation strategies”; whereas in the Chinese version of the autonomous driving classification standards, Level 3 already requires “risk mitigation strategies”, showing that the Chinese standards are much stricter than the SAE standards.
5. As early as 2017, Li Xingyu, the former business director of Horizon Robotics, mentioned on multiple occasions that dividing autonomous driving technology according to application scenarios is more valuable than dividing it according to levels. As of 2020, Ideal, NIO, XPeng, and other car companies have adopted similar standards. NOA or NOP is divided according to application scenarios.
6. Overall, the guiding principle for the industrialization and implementation of Level 3 autonomous driving technology in the future can be summarized as “assistance at high speeds, substitution at low speeds”.
7. Car companies selling Level 3 autonomous driving cars to consumers should explain the exact meaning of “Level 3” during the marketing stage to ensure that users can use them properly and reduce related accidents. This is a necessary prerequisite.
Throughout the autonomous driving industry, L3 has been assumed to refer to the L3 defined by the SAE, which means that the system continuously performs driving tasks in a specific scenario, but due to imperfect systems, drivers need to be ready to take over at any time. L3, which meets this definition, has always been controversial.
Superficially, the biggest obstacle to the mass production of L3 is “legal prohibition”; but in reality, after a safety accident, who is responsible for the liability is the biggest obstacle to L3. Therefore, to achieve mass production of L3, it is necessary to redefine “what is L3”.
In February and March 2020, the “Intelligent Vehicle Innovation Development Strategy” jointly issued by the National Development and Reform Commission and other 11 ministries and the “Automobile Driving Automation Classification” released by the Ministry of Industry and Information Technology officially recognized the controversial L3 product form. Before the Chinese government, Japan and South Korea had also introduced some policies and regulations to support mass production of L3 autonomous driving.
After careful review, we found that the L3 mentioned in the legal policies of China, Japan, and South Korea, and the L3 mentioned by Audi, are not the same concept, and the key difference is the subject responsible for accident. Among them, the laws of Japan and South Korea have clearly stipulated that the driver is the primary responsible person in case of accidents caused by Level 3 autonomous driving.
In China, although the Ministry of Industry and Information Technology’s product definition does not explicitly mention the issue of driving responsibility, several major car companies have redefined their products to explicitly state that drivers are still the responsible party. It is worth noting that L3 under the Chinese standards will be divided into “high-spec L2” and “quasi L4” directions.
The Awkwardness of SAE Standards and Audi A8To reduce obstacles for the 2018 Audi A8, once considered the world’s “first mass-produced Level 3” autonomous vehicle, Germany specifically issued an amendment to Article 8 of the Road Traffic Act in May 2017. This amendment allows the driver to take their hands off the steering wheel if the speed does not exceed 60 km/h under traffic congestion on the highway.
However, Germany’s move has not been followed by other countries. Under UK law, the Audi A8’s autonomous driving capability is limited to Level 2. Eventually, due to a United Nations Economic Commission for Europe regulation, even the Level 3 autonomous driving technology of the Audi A8 could not be put on the road in Germany.
Before its launch in the US in May 2018, the Audi A8’s autonomous driving function was also downgraded, offering only several basic ADAS functions, such as lane-keeping, adaptive cruise control, side collision prevention, and blind-spot monitoring, which are still at Level 1.
These cars still come with zFAS processors and lidar but lack hardware for monitoring the driver’s attention, capacitance touch steering wheels, and brake redundancy. This means that even if US law were to allow it in the future, Audi would not be able to directly upgrade these vehicles from Level 1 to Level 3 through OTA software upgrades.
According to Audi, this is because they are “unable to handle the laws of every state in the United States individually.”
However, this is only the surface reason. The deeper reason is that Level 3 technology is considered “inhuman” and leads to difficulty defining legal responsibility:
“Under the SAE standard, Level 3 autonomous driving systems not only free the driver’s hands but also their eyes for quite a long time. For example, the Audi A8 allows the driver to read books or use their phone while driving at speeds below 60 km/h. However, this system has not yet freed the driver’s brain (attention).”
The Level 3 system is not perfect, and when it encounters extreme situations that it cannot handle, it will issue a takeover reminder to the driver, who must intervene within a very short period (10 seconds for the Audi A8). In actual operation, this requirement faces two challenges:
- The driver’s quality can be quite good, with a strong understanding of the rules. They would choose to follow the “Level 3 Use Norms.” However, the stronger the capabilities of the system, the more the driver trusts the technology, leading to the relaxation of alertness. This could result in not being able to respond in a timely and appropriate manner to the system’s takeover reminders when they encounter danger.
This has already been a profound lesson in the aviation sector.# Automatic Driving in Aviation and Automotive Industries
Decades ago, airplanes already achieved Level 3 automatic driving under the SAE standards. Investigations of several air disasters found that the excessively strong automatic driving capability of airplanes resulted in pilots having nothing to do in most cases, leading to degradation of their driving skills. Therefore, in sudden danger, pilots either can’t take over in time or operate incorrectly after hasty takeover.
In the field of automatic driving cars, Level 4 under testing is equivalent to Level 3 under SAE standards, and the safety driver is equivalent to the driver on Level 3. Therefore, observing the behavior of these safety drivers can help us understand the performance of Level 3 drivers in real-world scenarios.
While testing a Level 4 autonomous driving car from a domestic company, the author found that because the system’s driving capability was good, the engineers temporarily serving as safety drivers behaved carelessly. The author was concerned about whether they could react in time when danger suddenly occurred.
Due to these concerns, Toyota’s attitude towards mass producing Level 3 vehicles is ambiguous. On the one hand, Toyota planned to launch mass produced Level 3 cars during the 2020 Tokyo Olympics in the summer. On the other hand, Gill Pratt, president of Toyota Research Institute (TIR), believes that “making Level 3 is probably as difficult as making Level 4.”
Gill Pratt said, “To ensure safety, Level 3 must leave 10 seconds of takeover time for the driver. The problem is that it’s difficult to predict what will happen in the next 10 seconds in urban/suburban environments. If Level 3 is smart enough to handle this, then it’s better to upgrade to Level 4.”
In order to ensure driver attention, Audi has considered various measures to restrict the driver’s “degree of indulgence”. These measures include whether tablets and smartphones can only be used when connected to the car so that the car can remotely shut down the device when switching. However, the involved parties have not reached a consensus on these details.
In Level 3 mass-produced vehicles, drivers may lack a sense of rules and deliberately take risks while knowing that the automatic driving system is not 100% reliable. This means that no matter how much the automakers emphasize in the user manual that “you must be ready to take over the vehicle at any time, and not free your mind”, the user will still take risks to test the limits of the technology. For example, intentionally not taking over when the system reminds them to do so.
Tesla has clearly stated that their automatic driving level is only Level 2 and the driver is the main responsible party, but there are still many drivers who use his car as “Level 4”, leading to disastrous consequences. This is enough to show that unreliable users are a reality and their number is not small.The more challenging problem with L3 systems is that because their autonomous driving capability is stronger than that of L2, drivers tend to trust them more, thereby increasing the probability of “unreliable” behaviors. An accident caused by an unreliable driver is enough to destroy the confidence that many “uninformed people” have in L3.
Another issue related to the takeover paradox and the unreliable behavior of drivers is the liability dispute after an accident.
According to SAE’s definition of L3, the primary driver responsibility lies with the automated driving system. However, if an accident occurs after the human driver takes over, the human driver becomes the primary responsible party. The problem lies in what happens if an accident occurs between the time the system issues a takeover reminder and before the human driver intervenes.
In theory, if the driver fails to intervene in a timely manner, the responsibility should lie with the driver. But what if the accident occurs after only 8 seconds, even though the driver took over within 10 seconds of the system issuing a warning?
In principle, the automated driving system would be responsible for the latter case. However, in practice, the situation may be more complicated, leading to many confusing disputes.
Germany’s revised Road Traffic Act also states that car manufacturers must install black boxes to collect data during driving to provide the basis for liability determination and distribution in case of accidents. However, this regulation has not dispelled concerns from car manufacturers.
According to an insider from Audi, their in-house legal advisers are generally opposed to L3, because even if the system has a safety rating of 99.9% when delivered to customers, the automaker would still be responsible for accidents that occur during automated driving. Even if it is a problem such as the owner taking over too late, the system may still be held responsible in the end.
This is one of the key reasons why Audi ultimately decided to cancel the L3 project.
Japan and South Korea: Driver Liability for L3 Accidents
Audi’s dilemma with the L3 project actually represents many automakers’ quandaries with L3: to give up L3, they would miss the opportunity to accumulate data and iterate on technology, and may lose the ticket to compete in the L4 race. If they “tough it out,” just one or two fatal accidents may be enough to ruin years of effort.
This sentiment among automakers may also be conveyed to decision-makers in their home countries.
Autonomous driving is an important issue in the technology competition among major powers. Some decision-makers in automotive giants are worried that if their country’s automakers are too conservative about L3 due to too much concern about “who’s responsible,” their country may lose to other countries in the competition of autonomous driving technology. In Japan and South Korea, decision-makers have even more serious concerns about this issue.To dispel doubts of automakers about the mass production of L3 autonomous driving cars, Japan proposed in March 2018 in the revised “Road Traffic Law” that owners (drivers) should bear responsibility for accidents caused by L3 autonomous driving cars; automakers only need to take responsibility in case of system defects.
Of course, automakers need to obtain government approval for the usage conditions of L3 systems, such as weather and road conditions, driving time and speed, etc.
In May 2019, this amendment was officially passed by the parliamentary vote.
However, in another “Road Traffic Law Amendment” passed in June of that year, Japan allowed L3 drivers to eat, use their phones, read books, or use in-car entertainment facilities. The revised law will take effect in May 2020.
And to prevent drivers from playing with their phones for a long time, in September 2019, the Japanese Cabinet approved a rule to punish drivers for “inappropriate use of L3” behavior: if the driver fails to take control of the vehicle in time after the system issues a prompt, they will face a maximum fine of 12,000 yen (110 US dollars).
Taking a comprehensive look at the three paragraphs, it means that the law allows L3 drivers to enjoy themselves while driving, but sorry, once an accident occurs, the consequences are on you; at the same time, it is hoped to limit the excessive indulgence of the driver to reduce accidents.
This reflects Japan’s conflicting attitude towards the impending international competition in autonomous driving – they want their manufacturers to seize this opportunity, but do not want them to bear too much pressure; they encourage consumers to boldly try, but are afraid they will be too daring.
In December 2019, Honda announced that it would release the first L3 autonomous driving commercial vehicle in Japan, Legend, in the summer of 2020.
In South Korea, the “Automobile Accident Compensation Bill” drafted in April 2019 clearly stipulates that the driver is responsible for accidents involving L3 autonomous driving cars. Of course, in actual operation, the owner must purchase insurance for the autonomous driving car, and the insurance company will compensate the victim first after the accident occurs. Then, an inspection is conducted to determine whether there are any defects in the autonomous driving car, to determine if the manufacturer should be responsible for the accident.
On January 5, 2020, the South Korean Ministry of Land, Infrastructure and Transport announced that it will introduce the world’s first L3 autonomous driving car safety standard. The standard will take effect 6 months after its release, making it possible for automakers to produce and sell “L3 autonomous driving cars with lane-keeping function” starting in July. However, the driver must take control of the vehicle within 15 seconds after the system issues a warning.At the beginning stage, the “L3” system in the Korean market does not support automatic lane change on high speed and lane change still needs to be operated manually. It is not clear when the automatic lane change can be achieved.
In contrast, Tesla has been able to achieve automatic lane change on highways since October 2018. It seems that the “L3” in Korean law is greatly discounted – in fact, “L3” in Korean law refers only to “lane-keeping”.
Similarly, according to the “First European L3 Regulation” passed by the United Nations Economic Commission for Europe’s (WP.29) Automotive Regulations Coordinating Organization in the World Forum on Harmonization of Vehicle Regulations in June 2020, the application scenarios of the so-called “L3” are also limited to automatic lane-keeping functions, and the vehicle speed is limited to within 60 km/h.
It is worth noting that according to the provisions of the United Nations Economic Commission for Europe, the activation conditions of the L3 system are: the vehicle should be on a road that prohibits pedestrians and cyclists; secondly, physical separators are provided on the road to separate traffic flows in opposite directions (single lane). This scenario may be extremely limited.
Next, some German car manufacturers may use words such as “single lane automatic driving” to replace “L3” euphemistically, and the upper speed limit of vehicles will also be adjusted from the original planned 120 km/h to within 60 km/h.
In fact, Audi A8 had previously been evaluated as “lane-keeping + adaptive cruise control function at low speed, which cannot be regarded as strict L3” due to multiple limiting conditions such as “the vehicle speed does not exceed 60 km/h, and there are no traffic signals or pedestrians within the monitoring range of the vehicle sensors” set for the activation of its autonomous driving system.
This is understandable; the evolution and commercialization of autonomous driving technology require a gradual process, which is much safer than directly implementing L3 under SAE standards.
It is worth noting that both Japanese and Korean regulations require the installation of devices to monitor driver attention in cars. Among them, a regulation passed by the Japanese Cabinet in September 2019 stipulated that if the driver uses the L3 function without a monitoring device, a fine of $110 will be imposed.
There is no similar provision in SAE standards. However, Audi A8 also voluntarily added a driver attention monitoring system in product design. It seems that limiting the “unreliability of drivers” through technical means has become an industry consensus.
However, the driver attention monitoring system may cause a severe decline in user experience while improving safety.Normally, human driving relies on “muscle memory”, with many actions being subconscious behaviors that ensure safety without deliberate thought. However, in Level 3 autonomous driving vehicles, the machine is responsible for driving, and drivers cannot rely on muscle memory once their hands are “freed”, so they need to be highly attentive and familiar with road conditions. This can actually be more tiring even than driving themselves.
But if we think of Level 3 as an “upgraded version of Level 2” and have the Level 3 system “assist drivers”, then issues such as “contra-human nature” and “takeover paradox” can be easily solved. Furthermore, the safety and driving experience of the vehicle will be better.
From the progress so far, Chinese automakers are the pioneers in implementing this concept into practice. Additionally, some perceptive suppliers have also integrated this notion into their product strategies.
China: L3 will be separated into “upgraded L2” and “almost L4”
On March 10, 2020, Changan Automobile released its first mass-produced Level 3 autonomous driving vehicle, the UNI-T. The automatic driving function of the car is:
When the vehicle is traveling at a speed of less than 40 kilometers per hour on a highway, the autonomous driving system can simultaneously free the driver’s hands and eyes, and in the event of an accident, the manufacturer will assume responsibility. However, when the vehicle is traveling at a speed of over 40 kilometers per hour, the autonomous driving system can only free the driver’s hands, but not their eyes. In the event of an accident, the driver must take responsibility.
At the time, this regulation was ridiculed by some industry insiders as “Level 2 at high speeds, Level 3 at low speeds.”
In fact, under the current level of technology, downgrading L3 to L2 functionality on highways is a more responsible approach to user safety.
Of course, drivers may not always pay attention to the speed of the vehicle. Therefore, it will be a significant challenge for drivers to quickly recover their attention during the transition period from speeds under 40 kilometers per hour to over 40 kilometers per hour.
However, Changan emphasized that the L3-level autonomous driving technology in UNI-T is at a “mass-production” level, not just a “mass-production configuration”. Changan UNI-T, which will soon be on the market, will not be equipped with L3-level autonomous driving systems.
In mid-March, GAC New Energy unveiled its first mass-produced Level 3 autonomous driving car, the Aion LX. However, even on highways that are suitable for L3 applications, GAC New Energy still does not allow users to completely free their hands.
At the time, Xu Junhai, director of the GAC New Energy Technology Center, said in an interview with “the First Electric Network”: “It is totally feasible technically, but current traffic regulations do not allow it.”In order to avoid situations where the driver’s attention is not focused and their hands are removed from the steering wheel while using the L3 autonomous driving function, GAC New Energy has installed a camera and pressure sensor on the steering wheel to monitor the situation. If there is any non-compliance, the driver will be alerted through sound and image warnings.
This means that even if the L3 system is activated, the driver is still the main responsible party. According to the SAE classification standards, this still belongs to L2.
However, compared with the mass-produced L2 autonomous vehicles under the SAE standard, the Aion LX has achieved redundancy in both the autonomous driving chip and the wire control brake (for the chip, two EyeQ4 chips are equipped and for the brake, a combination of iBooster and ESP is used).
Therefore, the autonomous driving system of the Aion LX is considered as a “high-end version of L2”.
Similar to the Aion LX, the autonomous driving function of the XPeng P7 is also designed according to the L3 standard. However, before its launch at the end of April 2020, XPeng did not mention the word “L3” much in its press release. It only said that it “can achieve scene expression close to L3 level autonomous driving, automatically select the optimal lane and change lanes according to navigation and road conditions, and automatically pass through highway junctions”.
Previously, Wu Xinzhou, Vice President of Autonomous Driving at XPeng, also clearly stated in an interview with 36Kr: “The driver’s vision should not leave the road for more than 10 seconds. There is a powerful driver monitoring system in the car, which requires drivers to keep their focus on the road, just like using L2.”
Currently, the official designation of the autonomous driving system equipped on the P7 by XPeng is “XPILOT 3.0 Advanced Driving Assistance System”. “Advanced driving assistance” means that the driver is still the main responsible party, which is also L2 under the SAE standard.
However, compared with regular mass-produced L2 vehicles, the XPeng P7 is noted for its “high-end configuration” in both the perception system and computing platform.
The P7’s perception system includes 12 ultrasonic sensors, 13 cameras, 1 in-car camera, and 5 millimeter-wave radars, which greatly exceeds that of Tesla. In addition, the P7 also deploys two sets of computing platforms, XPILOT 3.0 and XPILOT 2.0, which are completely independent and redundant to each other in terms of hardware.
Wu Xinzhou has stated in a media interview that for a car manufacturer, it is not very important to define exactly what L3 is, but rather whether the achieved outcome has value for users.On March 25th, at the forum hosted by Yanjing Automotive, Dr. Liu Dehao, an autonomous driving algorithm expert from XPeng Motors, pointed out in his speech that XPeng’s strategy is to achieve the ability of advanced driver assistance in terms of functionality, even with infinite extension upward, but in terms of safety strategy, we still primarily rely on human driving. XPeng’s goal is to achieve the optimal solution of specific experience brought by specific functions within the safety threshold.
Another new automaker, Ideal Automotive, plans to launch a self-driving feature between L2 and L4 in their mass-produced cars between 2021 and 2022. Normally, the functionality between L2 and L4 is L3, but it’s called NOA internally by Ideal. NOA stands for Navigate on Autopilot, and in Chinese, it is called “Autonomous Navigation Driving Assistant”. Tesla was the first company to propose this concept. In October 2018, when Tesla’s autonomous driving software was upgraded to version 9.0, the Tesla Autopilot had the ability to automatically change lanes on the highway, and this feature was referred to as NOA.
After Tesla’s demonstration of the NOA feature, the industry exclaimed, “Fully autonomous driving is just a step away”. It can be seen that NOA’s technical capability is similar to L3 under the SAE standard.
However, considering that Tesla’s website mentions that the driver needs to pay attention to road conditions at all times in the introduction to “fully autonomous driving” functionality, we can speculate that, under NOA, the driver is still the responsible party.
This responsibility setting is supported by the academic community.
As early as March 2019, in a paper entitled “Can Advanced Autonomous Driving Break Through the Bottleneck of L3 Industry?” published in the “Intelligent Connected Vehicles” magazine, Professor Bian Mingyuan and Professor Li Keqiang from the School of Vehicle and Mobility of Tsinghua University mentioned two forms of L3 level autonomous driving technology in the actual landing process:
Currently, most of the industrialized solutions of L3 are embodied in the integration and integration of different levels of driving assistance (L1/L2) functions, which are equivalent to the technical route of upgrading ADAS system (achieving L2+ level autonomous driving); The future development direction of L3 industrialization needs to approach L4 to achieve a smooth transition, and its technological development difficulty will inevitably further increase.
The former route is mainly applicable to the high-speed highway area of closed and semi-closed areas, structured and semi-structured roads (60~150 Km/h). In these scenarios, L3’s primary value is to improve traffic safety (extended L2.5). The latter route is mainly applicable to areas suitable for low-speed driving (0~60 Km/h), such as urban roads, parks, ports, and mining areas. In these scenarios, L3’s primary value is to replace the driver (L4-).
Refer to the figure below for details:
In general, the guiding principle for the industrialization of L3 autonomous driving technology in the future can be summarized as “assisting humans at high speeds and replacing them at low speeds.”
In March 2020, the Chinese Ministry of Industry and Information Technology released the “Classification of Vehicle Driving Automation,” which although did not clearly define who is responsible for driving at the L3 level, clearly stipulated that L3 production vehicles must be equipped with a system that can monitor the driver’s ability to take over at all times.
If the monitoring system determines that the driver’s ability to take over is “about to not meet requirements,” then even if the autonomous driving system has not yet failed, it has a responsibility to request the driver to take over. This means that regardless of who is responsible for the accident, drivers of domestic L3 production vehicles cannot read books or play on their phones in the car as some people had previously imagined.
Considering that many “production L3” vehicles at this stage only have redundancy in certain key components/systems rather than “full system redundancy,” it is a more pragmatic approach to position “L3” as “assisting humans at high speeds” from the perspective of protecting consumer safety and ensuring the long-term healthy development of the industry.
As we mentioned earlier, requiring drivers to “maintain a high level of concentration at all times” while the system performs the driving task for the vast majority of the journey will lead to a poor user experience.
However, if the driver uses L3 as a “high-end version of L2,” compared to ordinary L2, both safety and user experience will improve.
Another point worth noting is that in the Chinese “Classification of Vehicle Driving Automation,” the explanation for the “dynamic driving task takeover user” of L3 mentions that the takeover person can be either inside or outside the car.
Under the SAE standard, it is difficult for us to understand what it means for a “person outside the car to take over” an L3 autonomous driving vehicle in an emergency.
My understanding is that under the Chinese standard, besides being an “upgraded version of L2 under the SAE standard,” L3 also has a form that includes a safety driver or a remote control operator for testing/operational trials, which can also be called “quasi-L4.”
Under the Chinese standard, L3 vehicles taken over by “person outside the car” in emergency situations are equivalent to L4 vehicles without a safety driver taken over by a remote control operator under the SAE standard. This also coincides with the concept of “assisting humans at high speeds and replacing them at low speeds” in the landing of L3 autonomous driving technology summarized by Bian Mingyuan and Professor Li Keqiang.The detail that confirms the above speculation is that under the SAE standard, only Level 4 vehicles require “risk mitigation strategies”, which means that the system needs to be able to take measures to reduce risks such as parking within the lane if the automated driving system or the user is unable to perform driving tasks.
However, under the Chinese version of the automatic driving classification standard, Level 3 vehicles require the addition of “risk mitigation strategies.” It can be seen that Chinese standards are much stricter than SAE standards.
For “almost Level 4” vehicles, it is reasonable to equip them with “risk mitigation strategies.” For those “high-end Level 2” vehicles, adding “risk reduction strategies” undoubtedly improves the safety level (enjoying the treatment of Level 4 in some moments).
As one of the automakers involved in drafting the “Automotive Driving Automation Grading” by the Ministry of Industry and Information Technology, Changan specially mentioned when introducing its Level 3 automated driving function: “When the vehicle speed exceeds 40 kilometers per hour, once the system detects that the driver has left the road for a long time, it will emit sound and instrument icon warnings. If the user fails to take over after being reminded, the system will execute risk mitigation strategies to slow down and stop the car.”
At that time, Changan’s explanation was mocked as “L2 on the highway.” But few people noticed that due to the addition of “risk mitigation strategies,” this “L2” has a certain ability that only Level 4 vehicles have under the SAE standard in some extreme situations.
Like Bian Mingyuan and Professor Li Keqiang, Vice President of Changan Automobile Intelligentization Research Institute, Li Yusheng, also holds the view that “at the Level 3 stage, intelligent driving products will achieve structured road low-speed substitution for humans and high-speed assistance for humans.”
In the international market, companies such as Mercedes-Benz and Toyota have simply defined features previously designed to L3 standards as “L2.”
Major suppliers have redefined L3 as “ADAS.” Legal responsibility is borne by the driver, which largely dispels car manufacturers’ concerns when launching L3 functions and is conducive to the continuous expansion of the application scenarios and capabilities of automatic driving. This will drive the application of automatic driving chips, LiDAR, and high-precision maps on mass-produced cars at a faster pace.
As if predetermined, suppliers of automatic driving components, such as Mobileye, Velodyne, and Bosch, are also adjusting their definition of L3 towards “high-end L2.”
In Mobileye’s previous business planning, the L3 for consumers came after Robotaxi. This meant that Mobileye, which is deeply bound to traditional car companies, was very conservative in its thinking and had a very cautious attitude towards controversial L3.
Actually, as early as the second half of 2017, Mobileye proposed the concept of L2+. Although there was no specific definition, it is clear that the biggest upgrade in function compared to L2 is the active lane change (ALC). This requires vehicles to have both longitudinal and lateral automatic control functions (ACC/AEB/LKA) and the ability to increase local or global path planning.
Distinguishing from L3, L2+ still requires the driver’s full attention throughout the journey, which avoids the two major problems that currently constrain the landing of L3: responsibility division during the handover of control between human and vehicle and the response time during the transition process. The former needs to be followed up in legal regulations, and the latter, if too short, will reduce user experience.
In a conference call after purchasing Moovit, Mobileye CEO Shashua said, “Level 3 is a very problematic niche. We do not believe in Level 3.” In Mobileye’s cooperation with some car companies, the car companies mention L3 in their PR, while Mobileye’s official website says L2+.
It can be seen that Mobileye is extremely cautious about the controversial L3, and they will not assume responsibility for automatic driving accidents caused by the driver’s failure to take over in time.
In the summer of 2016, after the first automatic driving accident at Tesla, Mobileye “blocked” Tesla on the grounds that Tesla had made inappropriate and misleading claims about the product functionality involving Mobileye chips. In the future, when so-called “L3” vehicles have accidents, Mobileye may still “sanction” customers to defend its values.
Before 2018, when mentioning LiDAR, there was a widely agreed statement that it was “essential for L3 and above automated driving vehicles”, and it was rare to hear that L2 also needed to be equipped with LiDAR. But in early 2019, Velodyne announced the launch of Velarray LiDAR for L2, the first time LiDAR was “associated with” L2.
What’s even more interesting is that in October 2019, after Velodyne reached a strategic cooperation agreement with Mobis, a Tier 1 company under Hyundai, the press release issued by Hyundai and Mobis mentioned “collaboration for mass production of L3”, while Velodyne’s official website only mentioned “collaboration for ADAS projects” without any mention of so-called “L3”.According to the SAE standard, the essence of L2 is “system-assisted driving”, so the autonomous driving system can be considered as ADAS (Advanced Driver Assistance System), while the nature of L3 is “human-assisted driving”, which should not be regarded as ADAS. Moreover, the view that “L3 is the watershed of autonomous driving” has long been an industry consensus.
Velodyne calls the “L3” mentioned by Modern and Mobis “ADAS” precisely because they have realized that the “L3” in Korean law is not the same concept as L3 under the SAE standard. In the Korean standard “L3”, the driver is still responsible for the driving, that is, the role played by the so-called “L3 level autonomous driving system” is still just ADAS.
Velodyne also linked the laser radar to L2 at the beginning of 2019 for the same reason.
In December 2018, Velodyne received strategic investment from the Japanese company Nikon, and the two sides decided to cooperate in the production of laser radar in Nikon’s Japanese factory. Japanese car companies are also its main target customers. According to the Japanese law revised in March of that year, the L3 under the Japanese standard is actually the “high-end version” of L2 under the SAE standard. Therefore, Veldoynye’s plans to develop L2 market are reasonable.
At CES 2020, Velodyne launched the Velabit laser radar targeting L2. Meanwhile, Anand Gopalan, the new CEO of Velodyne, mentioned in an interview with the media that two-thirds of the company’s automotive business contracts are related to ADAS, while one-third are related to autonomous driving cars.
This further confirms the concept difference that “L3 in the eyes of Japanese and Korean car companies is only ADAS in the eyes of Velodyne”.
Also at CES 2020, Carnavicom, a Korean company, mentioned that “in order to deploy ADAS technology, we invested 4 billion won in 2019 to build an SMT production line, hoping to increase the annual production capacity of laser radar sensors to over 1.8 million”.
The ADAS here refers to the same concept as the ADAS mentioned by Velodyne, which is the so-called “L3” in the laws and automakers of Japan and Korea.
In the Chinese market, Bosch is the supplier that responded most significantly to the change in the definition of “L3”.
In October 2019, Chen Liming, the president of Bosch’s chassis control system in China, said in an interview with 36Kr that “the progress of L3 has clearly deviated from expectations”.In January 2020, Jiang Jingfang, Senior Vice President of Bosch Chassis Control System China, mentioned in a speech that L3 and higher-level autonomous driving still needed to be discussed in the field of electric vehicles.
However, at the online press conference on May 14, Bosch China President Chen Yudong emphasized that for L3 and lower-level autonomous driving, mature technologies would be pushed to the market as soon as possible. “In the field of L2.5, L3 autonomous driving, including hands-free function on highways, Bosch Shanghai and Suzhou can currently undertake mass production projects.”
Valeo, who has always believed that “L3 is a fake problem”, has also noticed that “L3 is being redefined.”
In January 2020, Christophe Marnat, Executive Vice President of Valeo Electronics and Advanced Driving Assistance Systems, mentioned in an interview with the media that “ADAS functions are improving, and the entire industry is developing towards L2+ or L2++ or L2+++” and “Automakers are stopping to accept responsibility related to Level 3” at the same time.
With giants like Bosch and Valeo re-focusing on the L3 market under the new definition, more and more car companies will join the “L3 mass production” camp in the next few years. However, consumers who purchase L3-level autonomous driving vehicles must understand that this L3 is no longer what they thought it was.
Dividing according to application scenarios has more value than dividing by “level”.
For car companies selling L3-level autonomous driving vehicles to consumers, it is necessary to explain the meaning of your L3 clearly during the market promotion stage. This is a necessary prerequisite to ensure that users can use it in a standardized manner and reduce related accidents.
In fact, as early as the second half of 2017, Li Xingyu, the former Director of Horizon Business, mentioned on multiple occasions that dividing automatic driving according to application scenarios is more valuable than dividing by level. Tesla’s NOA (Navigation on Autopilot) proposed in October 2018 is defined by scenario.
In August 2020, Ni Kai, the founder of WeRide, mentioned in a speech that they don’t emphasize that autonomous driving technology must be extended to a certain level. “From our perspective, only emphasizing L4, rather than emphasizing actual commercial applications, does not conform to WeRide’s mass production route.”
Ni Kai said that WeRide doesn’t pay much attention to the “manual takeover rate” indicator, but stands from the perspective of end users, “and pays more attention to whether it can adapt to different road conditions, and can degrade or upgrade functions when switching high-precision maps.”In 2020, there was an interesting phenomenon in the Chinese market. Many models’ autonomous driving capabilities were originally built to meet the L3 standard, and were marketed as such during the earlier pre-launch phase. However, as the launch date approached, references to the “L3” concept were downplayed, and instead replaced with scene-based definitions such as NOA, NOP, NGP. Baidu’s ANP, introduced at the end of the year, is similarly defined by scene.
This transition was a wise move, as scenes are more important to users than levels.
However, in reality, some car companies are still struggling psychologically, even though they no longer refer to L3, but instead “navigation-assisted driving”. On one hand, they hope to avoid liability issues by emphasizing that their systems are “just assisting”, but on the other hand, they are not satisfied with the idea of their systems only serving as “assistant” roles.
For instance, certain car manufacturers’ key executives mention the concept of “manual takeover rate” when referring to their navigation-assisted driving systems, which defaults to the idea that the system is the “driving behavior subject” (with humans assisting the system). However, this conflicts with the positioning of “assisted driving”. This vague positioning may result in a future automatic driving accident, which would cause the company or even the entire industry to face public relations crises.
Currently, it is especially important for auto manufacturers to manage user expectations through clear product definitions before customers place orders. This will help avoid situations where customers “discover this is not what I wanted” after purchase, resulting in disappointment or even turning away from the brand.
After realizing these issues, many car manufacturers, not only those offering L3, but even those offering L2, will emphasize the usage scenarios and user experience of their autonomous driving functions, while trying to avoid explicit references to autonomous driving levels.
This article is a translation by ChatGPT of a Chinese report from 42HOW. If you have any questions about it, please email bd@42how.com.