In the previous issues of Easy Car Original, Easy Car Technology made an in-depth analysis from the auto-driving enterprise to the auto-driving technology, mainly focusing on the hardware equipment, software algorithms and the underlying architecture of the car. Today, we will broaden our horizons and talk about the two major development directions of auto-driving at present-bicycle intelligence and vehicle-road coordination.

bicycle intelligence: its center of gravity is more inclined to the degree of automation of vehicle driving, and the technical realization route of bicycle intelligence is also divided into two types. One is represented by Waymo, which senses with multi-line lidar and multiple sensors; The second category, represented by Tesla, is based on camera and visual perception.

vehicle-road coordination: on the basis of bicycle intelligence, the "people-vehicle-road-cloud" traffic participation factors are organically linked through the vehicle networking, which helps the self-driving vehicles to upgrade their abilities in environmental awareness, calculation and decision-making, control and execution, and accelerates the maturity of autonomous driving applications.

in short, the essence of bicycle intelligence and vehicle-road coordination is the allocation of technology and cost on the vehicle side and the road side.

Among them, bicycle intelligence is the solution promoted by most unmanned enterprises at home and abroad, but this does not mean that it will become the optimal solution to realize unmanned driving, nor can it be said that vehicle-road coordination is the optimal solution. Although the most ideal mode of L4-L5 level autonomous driving is to achieve a high degree of coordination of "car-side-road-side-cloud", from smart cars to smart roads, car-side intelligence and roadside intelligence cooperate and echo, but the development of car-side intelligence and roadside intelligence is not completely synchronized, and the route selection of autonomous driving faces the problem that different abilities such as perception ability and decision-making ability (computing power) are distributed on the car side and roadside, so the corresponding automatic driving costs are different.

At present, in the track of autonomous driving, the technical routes of various companies are not completely similar. Some of them focus on bicycle intelligence and arm the car to the teeth as much as possible. The hardware devices such as laser radar, millimeter wave radar, camera and high-precision positioning are all fully equipped and selected with full parameters, representing enterprises such as AutoX and Xiaoma Zhixing. Some bicycle intelligence and road coordination go hand in hand, and both hands should be grasped, representing enterprises such as Baidu Apollo and Mushroom Car Federation.

As a representative enterprise of bicycle intelligence, AutoX is quite prominent in China. Xiao Jianxiong, president of AutoX, once said: Automatic driving can't be realized solely by road-vehicle coordination, and it is difficult to achieve full coverage of road intelligence at present. In order to improve the safety of autonomous driving and driving experience, we must improve the technology of bicycle intelligence to cope with various extreme weather such as rain and fog, and ensure that users can get a safe Robotaxi experience.

At the same time, vehicle-road coordination and bicycle intelligence complement each other, and vehicle-road coordination is a beneficial supplement to bicycle intelligence. If road infrastructure construction is increased at the national level, it can help improve the road accuracy of autonomous driving and increase the safety of road traffic.

It can be seen that Xiao Jianxiong still pays more attention to bicycle intelligence. He thinks that roadside facilities have a series of problems, such as failure, repair and maintenance, which can only be regarded as an auxiliary function. The key point is to make the car the most intelligent and safe, improve sensor redundancy and be foolproof.

in July this year, AutoX released the fifth-generation fully unmanned system AutoX Gen5, which was built from scratch for unmanned driving. A * * * has more than 51 sensors, equipped with a huge sensor cluster, with a total pixel of 221 million pixels per frame, equipped with a high-definition 4D millimeter-wave radar and a resolution of 1.9 degrees. The lidar point cloud imaged every second has reached 15 million, and the computing platform supports the computing platform of 2211TOPS. This amazing hardware assembly shows AutoX's determination to make the car the smartest.

car-road coordination take Baidu Apollo as an example. Baidu's technical path is "smart car+smart road" to achieve the optimal solution of automatic driving.

In terms of bicycle intelligence, Baidu Apollo launched a new generation of unmanned car Apollo Moon, which was jointly built by BAIC Extreme Fox.

In terms of hardware, the fifth generation kit carried by Apollo Moon adopts a multi-redundant sensor combination of 1 main lidar, 13 cameras and 5 millimeter-wave radars. A low-cost forward lidar is installed in the front of the vehicle, which will be used in redundant systems in case of system failure. Although the use of lidar is reduced, Apollo Moon has increased the number of cameras, and at the same time greatly improved the image resolution and frame rate, and the visual perception ability is playing an increasingly important role.

In addition, the computing platform adopted by Apollo provides more computing power than 811TOPS, and more devices of vehicle specifications are used, thus realizing the integrated design of the main computing system and the backup safety system for unmanned. The water cooling design is adopted, which not only reduces the volume, simplifies the structure, but also has extremely low overall noise and very quiet interior.

On the roadside, Apollo Air is the only technology in the world that can realize the closed-loop automatic driving of open road L4 only through roadside perception.

Apollo Air technology can realize "vehicle-road-cloud" information interaction by using wireless communication technologies such as V2X and 5G without vehicle-side sensors and only relying on roadside light perception and traffic light information, thus enabling automatic driving.

compared with bicycle intelligence, the vehicle-road collaborative technology route can not only expand the vehicle perception range, ensure the safety of automatic driving, but also reduce the requirements for the vehicle-side perception system, thus further reducing the cost of bicycle automatic driving.

according to Baidu's own statistics, vehicle-road coordination can solve about 54% of the problems encountered by bicycle intelligence in road test, reduce the number of takeover by 62% and reduce the bicycle cost by 31%.

At present, Baidu has carried out the practice of vehicle-road coordination scheme in Beijing, Guangzhou and Shanghai.

In Beijing, Baidu has carried out intelligent transformation of vehicle-road coordination for 28 intersections with a distance of 1.2 km, and built a basic environment to support the test operation of L4 self-driving vehicles; The supporting platform of vehicle-road collaborative edge computing is built, and the service framework of edge computing is built, so as to realize the edge cloud integration functions such as equipment management and vehicle-road collaborative algorithm synchronization.

In August, 2121, Huangpu District, Guangzhou Development Zone and Baidu Apollo started the "New Infrastructure Project of Intelligent Transportation for Auto-driving and Vehicle-Road Coordination in Guangzhou Development Zone, Huangpu District, Guangzhou", covering 112 intersections and sections of Huangpu's 133km urban open road.

Shanghai Jiading Automobile City has developed an intelligent networked automobile test environment for open roads, with a construction mileage of 37.8km and a coverage area of about 65km2. Through the intelligent transformation of 56 intersections and key sections, more abundant test scenarios have been provided.

through the technical routes of these two companies, we can see that although each company has expressed its vision and goal of bicycle intelligence+vehicle-road cooperation and two-legged walking, it still has its own emphasis in the actual implementation process, which also leads the two sides to different forks in the road, and it remains to be seen which one can stand on the market in the future.

The limitations of bicycle intelligence are:

1. Over-the-horizon perception and visual blind spots cannot be perceived.

Both cameras and lidar detect electromagnetic waves in essence, which is similar to human visual senses. Where people can't see, these devices can't detect them. A typical example of the blind spot is the "ghost probe". As shown in the figure below, when pedestrians appear, it is too late to slow down.

2. perception of harsh environment.

There are many long tail problems to be solved in bicycle intelligence. For example, in rainstorm weather, the bicycle sensing system is almost out of order, the noise of laser radar is increased due to water reflection, the camera picture is blurred, and the confidence of target recognition is reduced.

In the dark scene, the visual perception conditions of bicycles are seriously insufficient, the exposure time is prolonged, and the photosensitive range is reduced. Because of the lack of color and semantic information fed back by the camera, the radar cannot identify obstacles.

3. The utilization rate of high-cost equipment is low.

it is very expensive to install laser radar and other equipment in a car. However, a car is stationary most of the time, and the driving time only accounts for a small part (stopping for one night to drive to work and stopping for one day to drive off work). Such expensive equipment has low utilization rate and is not cost-effective.

To talk about the limitations of bicycles, we must talk about two examples, namely, Uber's auto-driving accident and Tesla's auto-driving accident, which are typical manifestations of the limitations of bicycle intelligence. The data shows that traditional cars have an accident about every 511,111 miles, and bicycle intelligent self-driving cars have an accident about every 42,111 miles.

therefore, there should be coordination between vehicles and roads. Install expensive equipment on the road, and the equipment on the road will sense it (and sometimes do some calculation work). Vehicle-road coordination is to "tell" the situation around the car by "road", for example, there is a car 211 meters ahead, pay attention to slowing down; There is a car accident 5 kilometers ahead of the route, so make an early detour. (Vehicle-road coordination is a good application tool for 5G because the vehicle speed is very fast, which requires a transmission mode with high bandwidth and low delay)

So the vehicle and the road become a unified whole.

Roadside devices have collected all the information of vehicles, and these data can be reported to a unified center, which will analyze and apply them according to these data. This center is the so-called "cloud brain":

1. All vehicles can be informed in time when there is an accident or congestion, and the travel time can be calculated according to the traffic conditions after setting the destination;

2. Predict in advance when and where congestion will occur through big data and give an early warning;

3. According to the predicted traffic flow, give travel suggestions, and when to start and which road will be the smoothest.

from the point of view of road-vehicle coordination facilities, intelligent road signs, signal lights and other road-vehicle coordination facilities can ensure that bicycle intelligence can obtain external data and information, and ensure unified traffic scheduling and safe driving. In an ideal state, vehicle-road coordination can really solve these weaknesses of bicycle intelligence.

Limitations of vehicle-road coordination:

Vehicle-road coordination is also dependent on bicycle intelligence, and one of them is very important. The network security protection of vehicle-road coordination and bicycle intelligence is not unbreakable.

For hackers, the destruction of bicycle intelligent network may only be a traffic accident of several cars, but the security failure of vehicle-road cooperative network may bring paralysis or even more serious to the whole traffic network. Therefore, once the vehicle-road coordination is invaded, the importance of bicycle intelligence is reflected. For the time being, not to mention how to realize the problem of no traffic jam, and to ensure the safety of vehicles when driving can only be taken over by bicycle intelligence.

In addition, in order to realize L5 autonomous driving in the whole area, it is necessary to lay intelligent devices on all sections, so it takes a lot of manpower and material resources to repair, maintain and test these roadside devices alone, regardless of when the policies can be completely matched, and if there is something wrong with the devices at a certain intersection, it depends on whether the vehicles are smart enough to deal with it.

From the perspective of two-way communication, every bicycle in the vehicle-road coordination is a very important part of the system and an important source of platform data. If the vehicles are not intelligent, the vehicle networking will not land and will lose its meaning.

generally speaking, according to the U.S. department of transportation, the core value of car networking is to improve consumers' travel safety and reduce traffic accidents. Then, a glimpse of the whole leopard, from the most important network security and road safety point of view to see the car-road coordination and bicycle intelligence, the two must be integrated and developed in the future.

for the United States, the field of artificial intelligence is leading the world, with sufficient talent reserves and strong basic scientific research strength. The number of artificial intelligence enterprises in the United States ranks first in the world, covering the basic layer, technical layer and application layer. In addition, the United States has developed integrated circuit technology, and the field of high-end chip design has always maintained a leading position, laying a good foundation for the development of high-performance vehicle chips. On the other hand, the United States lags behind China in communication industry and 5G field, and the investment in infrastructure is generally led by the market rather than the government, so the promotion of networking is slow. Whether it is the "Google School" or the "Tesla School" of bicycle intelligence, the core capabilities behind it are artificial intelligence algorithms and decision-making chips, which is the strategic advantage of the United States.

for China, the communication enterprises represented by Huawei are leading the world in 5G technology, and there are a large number of 4G and 5G base stations with wide coverage. By the end of 2121, there will be more than 611,111 5G base stations in China. In February, 2121, the Strategy for Innovative Development of Intelligent Vehicles predicted that by 2125, the construction of intelligent transportation systems and related facilities in smart cities will make positive progress, and the wireless communication networks for vehicles (LTE-V2X, etc.) will achieve regional coverage. In addition, judging from the road conditions in China, the total mileage of expressway in China ranks first in the world, and the total highway <: a class="hidden" href="/bentianlicheng/" tit