Autonomous driving faces a difficult road to truth

Changes in every industry will gradually move from madness to sanity, and the same is true for the autonomous driving industry. Since the emergence of the automobile industry, the pursuit of autonomous driving has never stopped. Automotive automation experiments can even be traced back to 1920. Automotive engineers have never stopped on the road to making technology serve people. The efforts of generations of automotive engineers have only one purpose, which is to "liberate human hands." And in 2023, a hundred years later, autonomous driving seems to be still far away from us, and the road to breakthrough for autonomous driving is still far away.

On the road to pursue the truth, there will be many difficulties, and the process of pursuing the truth will also have many challenges. The development of autonomous driving is not an exploration model that goes all the way to the end. It has gone through path selection, verification, and then giving up. The development of self-driving cars is not just about considering the feasibility of technology. As a key technology for serving public travel, how to make it acceptable to mass consumers is very important. The requirements of self-driving cars are also very important. The path to truth is difficult.

LiDAR, the non-essential choice of technology

The realization of autonomous driving is inseparable from three very important core technologies: perception, decision-making, and execution. From the perspective of perception, more complete traffic information will allow autonomous vehicles to complete driving tasks more safely. As an important technology that allows self-driving cars to clearly see the road conditions, the current self-driving perception is mainly through the installation of sensing hardware on the car. Lidar has the advantages of high sensing accuracy, complete detection information, and strong anti-interference ability. It has always been in an advantageous position in the selection of autonomous driving perception hardware, and autonomous driving companies represented by Waymo use lidar as the main perception hardware.

Many new domestic car-making forces are also very fond of lidar. Different car companies also have differences in the installation location, quantity, and scanning scheme of lidar. Among them, car companies represented by Weilai ET7, Ideal L9, and Zhiji L7 installed lidar on the roof of the car, car companies represented by Jiar S and Xiaopeng G9 installed lidar on the bumper, and Jidu was the first Brands that install lidar on the front cover. In terms of the number of installations, the number of lidars installed by manufacturers ranges from 1 to 4, and generally more than 2 lidars are installed.

Lidar is a key component that promotes autonomous driving from 0 to 1. It is also an important promotional accessory for many car companies to demonstrate their autonomous driving capabilities. The loading of lidar seems to have been equated with autonomous driving. Any autonomous driving technology is inseparable from lidar. For cars equipped with lidar, the focus of publicity must be L2 or even L3 autonomous driving. .

In a world where there are dragons, there will be dragon slayers, and the same goes for the autonomous driving industry. When everyone agrees that lidar is an essential accessory for autonomous driving, the emergence of Tesla has opened up new possibilities for autonomous driving. Due to the high cost of lidar, the purchase cost of many cars equipped with lidar has discouraged many people. This has also led to the failure of technical solutions that use lidar as the main sensing hardware to spread to mid- and low-end models with a high probability. reason.

The autonomous driving mode in which on-board cameras serve as the main sensing hardware opens the door to the concept that lidar is indispensable in the autonomous driving industry. By adopting a purely visual sensing solution that uses on-board cameras as the main sensing hardware on vehicles, Tesla has The road to self-driving will make self-driving cars more like “human beings”. The image data of the traffic environment is collected through on-board cameras and processed through a complex perceptual neural network architecture to construct a real-world three-dimensional vector space so that self-driving cars can understand and recognize the traffic environment.

The emergence of Tesla has made it a reality for autonomous driving to get rid of lidar, and has also made lidar a non-essential choice for the realization of autonomous driving technology. It has also opened a hole in the selection of autonomous driving perception hardware and proposed another possibility. .

High-precision maps, taking center stage

High-precision maps, as an important technology to make up for the lack of perception hardware and allow autonomous vehicles to drive more safely, have been playing a very important role since the emergence of autonomous driving. In the development of autonomous driving in recent years, high-precision maps seem to It's a necessity. As the "eye of God", high-precision maps allow self-driving cars to perceive distances that cannot be detected by other sensing hardware, allowing self-driving cars to understand the road conditions ahead and prepare in advance.

Compared with ordinary navigation maps, high-precision maps have the advantages of high precision, multiple data, and accurate positioning, such as road curvature, slope, lane line position, type, width, traffic lights, traffic signs, roadside landmarks and other elements , can all be presented in high-precision maps. High-precision maps are like a crutch for self-driving cars, allowing them to walk more smoothly.

Entering 2023, the autonomous driving industry has also ushered in a new round of development. It is self-evident that high-precision maps are irreplaceable for autonomous vehicles, but its development is even more difficult. In the autonomous driving industry in 2023, the concept of "emphasis on perception and light on maps" has gradually become mainstream. The "first echelon" of domestic driving assistance such as Xiaopeng, Huawei, NIO, Ideal, Horizon, and Haomo Zhixing have successively announced that they will We will continue to develop autonomous driving (advanced assisted driving) technology using a strategy that focuses on the vehicle's own sensors and is supplemented by high-precision maps.

The reason why many car companies have such a unified view is mainly because of the difficulty and high cost of constructing high-precision maps. According to the "White Paper on High-Precision Maps for Intelligent Connected Vehicles", using traditional surveying and mapping vehicles, decimeter-level maps The surveying and mapping efficiency is about 500 kilometers per vehicle per day, and the cost is about 10 yuan per kilometer. In addition, when artificial congestion occurs on the road and road signs are replaced, it is also a problem for high-precision maps to respond in time. In order for self-driving cars to drive safely under any circumstances, high-precision maps need to be updated in a timely manner. This A series of requirements further increase the cost of autonomous driving implementation.

Both Tesla FSD and Mobileye's REM adopt the concept of crowdsourcing maps, that is, every vehicle equipped with the system is a mapping vehicle. The data collected during driving is uploaded to the cloud to form a new dynamic layer. The more vehicles equipped with this system, the more accurate the map will be and the more timely it will be updated. Although this solution can solve the problem that high-precision maps cannot be updated in time, it brings more problems. Since high-precision maps contain very rich traffic information, if used improperly, important information will be leaked. For this reason Domestic laws have put forward more stringent requirements for the collection of high-precision maps, which has also made the collection of high-precision maps more difficult.

If autonomous driving can only be implemented using high-precision maps, autonomous driving will not be possible in some places that do not allow the collection of high-precision maps, which will also prevent the true popularization of autonomous driving technology. High-precision maps are both a crutch and a prison for autonomous driving. Only "light on maps and heavy on perception" can autonomous driving truly come to fruition. This is the main reason why many car companies have proposed this solution.

The development path of autonomous driving that relies on high-precision maps has reached a dead end. If it continues, it will be difficult to see the exit. Perhaps the only way to break the situation is to find other directions.

Intelligent network connection, a consistent choice

Intelligent network connection or bicycle intelligence, perhaps since the emergence of autonomous driving, has been a hot topic. The "Guidelines for the Construction of the National Internet of Vehicles Industry Standard System (Intelligent Connected Vehicles) (2023 Edition)" issued by the Ministry of Industry and Information Technology and the National Standardization Administration Committee in July 2023 mentioned that by 2025, the system will be able to support combined driving Intelligent connected car standard system with common functions for assisted and autonomous driving. By 2030, an intelligent connected vehicle standard system that can support the coordinated development of bicycle intelligence and network empowerment will be fully formed. At this point, the development of intelligent network will have a further clear direction.

Intelligent connected cars are cars that have the functions of environmental perception, intelligent decision-making and automatic control, or interaction with external information, or even collaborative control. The intelligent connected vehicle standard system is horizontally based on three levels: intelligent sensing and information communication layer, decision-making control and execution layer, resource management and application layer, and vertically based on functional safety and expected functional safety, network security and data security general specification technology. Support, form a core technical architecture of "three horizontals and two verticals", fully present the technical logic of the standard system, and clarify the status and role of each standard in the intelligent networked automobile industry technology system. At the same time, it combines the technical relevance of intelligent connected cars with mobile terminals, infrastructure, smart cities, travel services and other related elements to reflect the characteristics of cross-industry collaboration and jointly build an organic whole of collaborative development with intelligent connected cars as the core to better Give full play to the top-level design and guidance role of the intelligent connected vehicle standard system.