Recommendations on the classification of "autonomous driving" in passenger cars

The "autonomous driving" classification system for passenger cars is in urgent need of "iteration".

Challenges of existing systems

For a long time before that, the intelligent driving industry generally used SAE's driving automation grading system, sae J3016, as the standard for industry exchange.

The standard has been iterated since 2014, through 2016, 2018, and 2021.

Among them, the September 2016 version is the most popular.

This standard, mainly according to function and responsibility, divides intelligent driving into six levels of L0-L5.

L1 is known as Hands or Feet Free;

L2 is known as Hands & Feet Free;

L3 is called Eyes Free;

L4 is known as Minds Free;

The L5 is called Drivers Free.

This grading system has many successful elements, including the concept of responsible subject and system scope (ODD), which basically laid the status of SAE driving automation grading standards in the industry.

Many subsequent organizations have tried to introduce a variety of grading systems, most of which are based on the September 2016 version of SAE J3016.

However, this version also has some notable problems, such as the L3 level being defined as "human-machine co-driving".

This positioning has "shaken" some well-known enterprises, which have been affected in the formulation of strategies.

One of the most famous cases happened to Google.

When the company formulated the technical route, it abandoned the progressive technical route and adopted the L4 level automatic driving directly, partly because the L3 level "human-machine co-driving" did not have realistic operability.

In April 2021, in order to meet the needs of automated classification of mass production vehicle driving, SAE carried out an "upgrade" of the classification system, in this upgrade, the L3 level automatic driving was repositioned, and the original "human-machine co-driving" was named automatic driving.

This means that under the new system, the L0-L2 level is the assisted driving system, and the L3-L5 level is the automatic driving system.

This solves a problem, that is, the original "human-machine co-driving" problem of L3 is solved, but at the same time it brings a new problem, in the original concept, automatic driving started from L4 level, and now it has become from L3 level.

This concept has not actually been reversed in the short term.

After iteration, the SAE version of the grading system, including some other grading systems on the market, still has a significant problem, that is, it is impossible to effectively classify the capabilities of the intelligent driving system on the market.

This poses a particularly high level of problems.

In order to highlight the capabilities of their own intelligent driving systems, some car companies have begun to create many new grading schemes on their own, such as "L2.5", "L2.9", "L2.9999" and other concepts.

Of course, after Tesla's "lessons from the past", no mass-produced car company dares to call its intelligent driving system an automatic driving system.

In some upcoming practices, people usually say that some cars have the hardware conditions to support L4 level automatic driving, but the driving ability belongs to L2 level.

A more interesting example is that Tesla's FSD function, according to sae's naming convention, is an L2 level automatic driving system in the strict sense, while some models equipped with Mobileye EyeQ4 chips and algorithms can only do some simple ACC and lane keeping functions, also known as L2 level automatic driving.

This not only confuses users, but also seriously hurts the production enthusiasm of enterprises.

This means that at this stage, SAE's driving automation classification system has lost the most basic function of classifying the ability of intelligent driving system.

Many car companies, in order to be able to effectively distinguish from the intelligent driving ability of friendly businessmen, have to carry out a lot of innovation in naming, such as NOA, NGP, NOH, NOP and so on.

The significance of the grading system

For us in this country, smart driving technology is extremely important.

This technology, not only to create an infinitely replicable "robot driver", will greatly improve the user experience and labor productivity, the greater value is that intelligent driving technology will lay the foundation for a new generation of information industry, including:

New input system. These input devices need to let the machine know the entire external world, and it is a three-dimensional world. This will essentially subvert the way the digital world is constructed, that is, from the original two-dimensional flat world to the three-dimensional space world.

A new computing system. In the era of PC and mobile Internet, all information is two-dimensional plane information, mainly scalar, the main processor is the CPU; in the era of intelligent driving, the machine needs to process the information of the three-dimensional world, mainly vector space, the main processor is GPU and NPU.

The development of intelligent driving will not only produce a huge terminal computing chip industry, but also fundamentally change the industrial pattern of cloud computing, from the original processing of scalar data to the processing of vector data. The computing unit is mainly based on the cpu in the early stage, and transitions to GPU and NPU.

A new AI industry. It includes not only industries such as algorithms and models, but also data processing industries. The development of the intelligent driving industry will drive the rise and leadership of China in the artificial intelligence industry.

After the intelligent driving industry takes shape, it will be natural to develop the next generation of Internet industry on the basis of these information industries.

An effective capacity grading system is of great significance to promoting industrial development.

1. Promote technological development.

When the quality of technology and products at various levels cannot be effectively evaluated, it means that the excellent intelligent driving system and the bad intelligent driving system are judged to be at the same ability level, which is unfair, and will also make excellent research and development institutions lose a certain amount of motivation, so that bad institutions can be confused.

2. Help consumers understand technology and products.

When a new technology is implemented and productized, a more effective ability assessment system is needed to help consumers better understand the product.

The problem now is that as long as your ability does not reach L4 level autonomous driving, it is collectively known as L2.

In this level, there are the world's most capable intelligent electric vehicles, equipped with the best computing platform, lidar, and there are also newly started fuel vehicles with almost useless functions.

There are also some companies that bravely shout out L3 level intelligent driving, which makes many "friendly businessmen" feel depressed.

For consumers, nothing but confusion is confusion.

3. More conducive to market communication.

An efficient ability grading system, for all R & D enterprises, the more important value is to spread more conveniently, greatly improve the efficiency of communication, but also will enhance the enthusiasm of communication.

Effective market communication will promote the implementation of technologies and products at a faster speed.

4. Conducive to industrial exchanges.

Some of the standards given by SAE are not mandatory standards, but more like consensus-based recommendations.

SAE's grading recommendation standard has become a common standard for communication and exchange among industry people for a long time, and has a wide influence.

However, due to some problems in its ability grading, it has become increasingly unable to effectively promote industrial exchanges and effectively support industry management.

What is a good grading system

Measuring whether a driving automation classification system can meet the needs can be discussed from at least the following two dimensions:

1. Whether it is possible to effectively distinguish responsibilities.

The responsibility distinction here needs to be solved in the responsibility distinction between the two scenarios.

One scenario is that intelligent driving system research and development enterprises need to have a better responsibility distinction when promoting their own products and technologies.

At least, the original self-driving system is not very suitable at the moment.

Similar to Tesla's Autopilot, it is indeed a product naming practice that can easily cause ambiguity.

Another scenario is when an accident occurs, where there is a need to effectively distinguish between responsibility, whether it is the responsibility of the driver or the responsibility of the provider of the intelligent driving system.

In this regard, SAE's responsibility division system is successful, that is, assisted driving, human drivers are the driving body, but also the responsible subject of traffic accidents; automatic driving, intelligent driving system is the "driving subject", but also the responsible subject of traffic accidents.

2. Whether it can effectively distinguish the ability of intelligent driving system.

This function is more fundamental to a driving automation system.

For any intelligent driving system R&D unit, they urgently need the ability of their own R&D system to be recognized by the outside world; for any user, they also need to understand the capabilities of all purchased intelligent driving systems.

This is the biggest challenge facing SAE's automated grading system.

Of course, the SAE system still has great reference value, that is, it proposes the attribution of responsibility as an important criterion for distinguishing the capabilities of intelligent driving systems.

SAE also proposed another important criterion for dividing the capabilities of intelligent driving systems, that is, ODD, that is, coverage, unfortunately did not refine and standardize this piece of work.

SaE's biggest regret is that it only divides the state of intelligent driving into two states: assisted driving and automatic driving, but in fact, between automatic driving and assisted driving, there is also a transitional form of navigation assisted driving, and automatic driving is essentially navigational automatic driving, that is, automatic driving that needs to be entered into points A and B.

This point, Wei Xiaolite and great wall of the milli found this problem in practice, and has been reflected in marketing practice.

Therefore, a better driving automation grading system should be a combination of SAE standards and the practice of existing mass production car companies.

Refer to the recommendations

Based on this, the "Jianyue Car Review" proposes a reference framework for the intelligent driving classification system of passenger cars to throw bricks and stones, so that more industry elites can participate in these tasks, improve the grading system, and make it possible to keep up with the needs of industrial development.

This reference standard, in the naming convention, will be assisted driving, navigation assisted driving and automatic driving are all unified as intelligent driving, unified concept is conducive to communication; at the same time, defined as "intelligent driving" can also avoid legal disputes, while having a better marketing effect.

In terms of intelligent driving types, as mentioned earlier, they are divided into assisted driving, navigation assisted driving, and autonomous driving.

These three types of intelligent driving correspond to primary intelligent driving, intermediate intelligent driving and advanced intelligent driving respectively.

Assisted driving is divided into high-speed assisted driving and urban assisted driving;

Navigation assisted driving is divided into high-speed navigation assisted driving, urban navigation assisted driving and full-scene navigation assisted driving;

Automatic driving is divided into high-speed automatic driving, urban automatic driving and all-scenario automatic driving.

In terms of ability level, in order to be able to distinguish it from sae's system, it is divided into C1 to C8 by starting with the uppercase letter C.

In terms of differentiation, the reference standard of the "Jianyue Car Review" edition subdivides the L2 level of the SAE version into C1-C5, which can greatly improve the differentiation of the intelligent driving system.

At the same time, we will standardize the coverage scenarios, including three "typical scenarios" of high-speed, urban area and full scene, to distinguish the level of the intelligent driving system.

The speed domain factor still needs to be taken into account, and theoretically all levels of functionality should be available within the speed limit.

In order not to make the grading system too complicated, the grading reference standard of the "Jianyue Car Review" version does not incorporate usage indicators such as the number of takeovers, the proportion of driving time or the mileage coverage rate of the unified scene into the system, which also leaves room for optimization for peers to continue to iterate.

For example, the same is the urban navigation assistance driving system, that is, the ability level is C4, and some enterprises' intelligent driving systems have a very low takeover rate when they are used, and some have a high takeover rate, which is actually a need for further classification.

However, we cannot do everything.

End

This is just a reference standard, throwing bricks and stones.

In view of the strategic significance of the intelligent driving industry, it is hoped that the broader number of people of insight in the industrial chain can strive to create an efficient intelligent driving classification system.