In 2020, many automatic game player announced the key node of the automatic driving vehicle. However, Audi announced that it would give up carrying L3 on the A8 model to be launched next year. On the other hand, Ford, Mobileye and other enterprises also announced that they would postpone the plan of self driving taxi. As a result, the situation of this technology is becoming more and more serious. Even though there are some L3 mass-produced cars in the market, there are still many controversies. At present, the only certainty is that the development of automatic driving is more difficult, more expensive and slower than expected.

There are three main reasons for this situation: the blurring of the legal boundaries leads to the slow progress of the research and development of L3 + automatic driving; the technical problems have not been solved; in addition, the technology is facing great pressure from the commercial realization.

The first reason of L3 + autopilot dystocia: policy and standard to block the door

The imperfection or even lack of regulations and standards has always been the main problem hindering the development of automatic driving. This is also the fundamental reason why Audi announced to give up the introduction of L3 level automatic driving technology in the next generation A8 flagship model. It is reported that in the L3 area, the international regulatory bodies even have no agreement on the most basic function approval process. Apart from a few countries such as Germany and Japan, which have opened the use of L3 level automatic driving, Audi’s other target markets have not introduced policies related to supporting L3 grade automatic driving vehicles, including China.

In early March, the Ministry of industry and information technology released the publication of recommended national standards for approval, which indicates that China has its own automatic driving classification standard, but the description of L3 automatic driving scene is still vague. Not to mention the large-scale commercial support for L4 and L5 automatic driving, the supporting regulations for high-level automatic driving are also missing. The classification of driving automation is higher than that of driving automation

Three reasons for landing difficulty of L3 + autopilot

Why does this happen? Is it because the standard has not been issued, or other reasons? In this regard, Wang Zhao, chief expert of China Automotive Technology Research Center Co., Ltd. and deputy director of automotive Standardization Research Institute, said that the main reason is demand. “As the standards are generally based on scientific experience, they need to be based on the actual needs, but from the information we have come into contact with at present, the requirements of the main engine manufacturers, parts suppliers and government agencies for the automatic driving standards are different, and some are even contradictory. It takes time to identify the real problems to be solved, and then formulate the relevant standards.” And in Wang Zhao’s view, because the current automatic driving technology is in the process of rapid development, the demand for standardization is also changing rapidly, which also brings great challenges to the formulation of standards.

Liu Fang, general manager of automotive electronics business in popular area of NXP, agrees with this. In her opinion, the standard is based on the summary of experience, and its introduction depends on the cooperation of technology maturity to a certain extent. This means that only when the products and technologies reach a certain maturity, the laws and regulations will naturally complement to regularize their development. After having the standard, we need to have the means of testing. We need mass production products to do experiments, and then through each step of experiments and customer demand feedback, in turn, we put forward new requirements for hardware and promote technological progress.

However, the good news is that the work related to the formulation of automatic driving standards for L3 and above is being carried out in an orderly manner. “Last year, we have prepared and submitted a draft in combination with different application scenarios of automatic driving, and will speed up the promotion of standard formulation in the future. It is estimated that the relevant standards will be officially issued next year, and the draft for comments may be issued by the end of this year or early next year.” Wang Zhao said.

Reason 2 of L3 + autopilot dystocia: difficult to fill technical defects

The technical problem is a commonplace in the development of automatic driving. From a technical point of view, it is relatively easy to realize bicycle intelligence in simple scenes. But in order to ensure the safe and reliable operation of the autopilot vehicle under complex road conditions, there are still many areas to be improved in terms of technology.

Taking the high-precision map as an example, once the automatic driving at L3 and above level has been reached, the high-precision map has become the core technology of the threshold. It is the basis of path planning. It can provide key information such as location and traffic dynamic information for autopilot, and can also provide sensory redundancy when the sensor is in bad environment or other environment of vehicle.

However, there are still multiple limitations in the large-scale commercial application of high-precision maps. First of all, the cost of high-precision map acquisition is high, the update is slow, and there are problems in data map format exchange, data interoperability standards, so it is difficult to achieve real-time update. Secondly, according to the national regulations, at this stage, high-precision maps can not cover all scenes, especially the areas related to national security, there is no way to carry out civil or only show part of the elements. Therefore, at least at this stage, for high-precision map, its use conditions and application scenarios are more stringent.

At the level of analysis and decision-making, L3 + autopilot also puts forward higher requirements for data processing ability. It is generally believed in the industry that the computing power required to achieve L2 level automatic driving is about 10tops, L3 needs 30 – 60tops, L4 needs more than 250-500tops, and L5 needs at least 1000tops.

Under the requirement of computing power, many autopilot chips and computing platforms with high-level autopilot are limited in computing power. However, many domestic enterprises have seized this development opportunity, actively carried out technological research, launched positive PK with NVIDIA, Qualcomm, Mobileye and other foreign giants, and won the admission ticket of high-level automatic driving market.

For example, Huawei has launched the MDC intelligent driving computing platform for the field of automatic driving, which is based on Huawei’s self-developed AI chip, vehicle control OS and other basic technologies. Focusing on this product, in the past year, Huawei has successively worked with more than 50 main engine plants, Tier1, application algorithms, sensors, actuators and other customers and ecological partners in the industry to realize landing intelligent driving solutions in L2 + passenger cars, intelligent heavy trucks, port logistics, unmanned mine cards, unmanned distribution and other fields.

On the eve of Beijing International Auto Show 2020, Huawei launched a new generation of MDC intelligent driving computing platform. The new platform adopts a unified hardware π architecture, which can provide powerful computing power of 48-160tops. Based on the intelligent driving operating system AOS, Vos and mdccore, it covers different levels of intelligent driving applications from L2 + to L4.

Then, at the 2020 Beijing auto show, heizhima intelligent technology also launched a similar product, fad automatic driving computing platform. The maximum computing power of the platform can reach 140tops, and it is expected to be upgraded to 280tops in the future. The platform has high flexibility and scalability, and can meet the computing requirements of L2 to L4 automatic driving from bottom to top. According to reports, the black sesame fad computing platform has received hundreds of orders. With the joint efforts of relevant enterprises, although there are still many technical constraints in high-level automatic driving, compared with previous years, it has improved significantly.

Third, the business model needs to be broken

Laws and regulations can be issued, and technical problems can be solved one by one, but if no one pays for the product, there is still no way to implement it. This needs to start from the cost, cost control is a very important factor whether the product can achieve large-scale mass production.

Especially when it comes to L3 + autopilot, we have to mention the problem of multiple redundancy. In some scenes, L3+ automatic driving needs to transfer part or even part of the control to the system. This means that we must have multiple redundancy, including perceptual redundancy, communication redundancy, power redundancy, steering and braking redundancy, to ensure that when one of the systems fails, the self driving vehicle can still run safely, improve safety and reliability, and this will undoubtedly enter. Further increase the cost burden.

Take hardware as an example. According to the report on application scenarios and commercialization path (2020) of automatic driving released by China electric vehicle 100 people’s Congress, the cost of automatic driving hardware is about US $50000 / vehicle at the present stage, and it is expected that this part of the cost will be reduced to US $5000 by 2025. At that time, the lidar and computing platform, as the core components of automatic driving hardware with the highest cost, will become more expensive after large-scale use Both are expected to decline. For example, lidar, velodyne’s 64 line lidar was sold at US $80000 per piece. In 2017, Google successfully reduced the price of related products to US $7500 per piece after its self-developed lidar. In the future, the cost is expected to be further reduced to several hundred US dollars after the shift from mechanical to mass production of solid-state lidar.

In addition to the hardware cost, the R & D investment of enterprises is also a huge cost that can not be ignored. Take waymo, the leader of Robo taxis, as an example. The company needs to spend $1 billion a year on R & D, fleet spending and operating costs. But for now, the technology won’t bring much benefit to waymo, at least for the next few years.

As Robo taxis still need to spend high renovation and maintenance costs at this stage, McKinsey predicts that in the future, as the cost per kilometer of Robo taxis continues to decline, compared with the cost of traditional taxis, it will not reach the inflection point until 2025 to 2027. The driving force behind this is the further increase of manpower cost in the future. At the same time, the transformation cost of autopilot system is gradually reduced, which makes the balance between taxi driver’s manpower cost and automatic driving transformation cost gradually broken. It is estimated that the cost advantage will be further highlighted after the cancellation of security personnel and large-scale deployment in 2025 or so, and the subversion of travel service will be brought about by Robo-taxi. .

This means that although the prospect of the automatic driving market is very good, there is still a long way to go before it can be truly implemented. However, with the improvement of supporting laws and regulations and the maturity of products and technologies, the difficulty of commercialization will naturally decrease, and the large-scale application of automatic driving will be just around the corner.

Editor in charge: Tzh

Leave a Reply

Your email address will not be published. Required fields are marked *