1 week, 5 days ago

See a lot of friends in the discussion "is not" problem. As a practitioner in the auto driving industry, I can definitely tell you that 5g will have a profound impact on the development of auto driving. 

I found that when you talk about 5g technology, you are easy to fall into two misunderstandings. One is to discuss 5G under the premise that the 5G base station has not yet been popularized in large scale. Another is to think that 5G is to enable autopilot to connect to 4G faster 5G network, and then achieve faster and more precise V2X Vehicle to Everything function. 

However, v2x only uses the most basic "low delay" feature of 5g technology, and 5g has two other important features, namely, high capacity and high rate. These three characteristics complement each other, which will not only have a huge impact on the development of autonomous driving, but also subvert some existing industries. 

Next, I will talk about how 5g affects the development of automatic driving in three time lines: initial stage, development stage and mature stage. 


 
 initial stage of automatic driving

At the initial stage of automatic driving, that is, the current stage of automatic driving. 5g is a mature v2x function. 

At this stage, there are two top players who can best represent the current situation of autopilot technology. One is Google's L4 driverless car, which represents the most cutting-edge technology, and the other is Tesla Model Series, which has the strongest sense of technology in mass production vehicles. 

Google unmanned vehicle is equipped with self-developed lidar system, millimeter wave radar system and visual perception system. Equipped with a variety of advanced sensor systems, it is to obtain 360 ° environmental information around the vehicle and drive more safely. 


 360 ° environmental awareness. 


 https://www.tesla.com/autopilot Although Google's driverless car and Tesla model can achieve 360 ° environmental awareness and automatic driving in specific scenes, the real road and traffic conditions are far more complex than imagined. For example, pedestrian (vehicle) crossing scene (commonly known as ghost probe) or non sight distance (beyond the sensor sensing range or in the sensor sensing blind area) scene such as Daqu mountain road. 

This kind of complex scene is difficult to deal with only through single vehicle sensors, which will bring a lot of security risks to automatic driving. Therefore, it is necessary to use the communication technology outside the vehicle sensor to ensure the safety of the self driving vehicle in the complex scene. The vehicle connected to 5g network can realize real-time 
 beyond visual range perception 
 ability through v2x technology and 5g "low delay" feature. 


 https://pic1.zhimg.com/v2-2026a4ed9b3fee5274d5b1e8f6f9c498_ b. The "low delay" feature of GIF "ALT =" "> 
 big curvature mountain road perception blind area 
 
 
 5g not only helps to realize the perception ability of automatic driving beyond visual range, but also can realize remote driving control on the vehicle with 5g communication module and wire control function, as shown below. 


 
 5g remote driving control 
 
 
 according to the current laws and regulations, the large-scale road test of automatic driving must be equipped with a safety officer on the driver's seat to ensure the safety of the test. With the development of automatic driving, the car will no longer be equipped with steering wheel, brake pedal and driver's seat, and the configuration of safety officer will be gradually reduced. This process will also make the remote driving control vehicle to be popularized and applied. On the other hand, remote driving control will have a profound impact on the industry of "Valet driving". In the future, Valet driving can really be done at home. 


 
 development period of automatic driving

At present, the technology of autonomous driving, which is mainly based on perception, positioning, planning and control, is basically mature in Lucian, and most of the automatic driving companies are also based on this set of technical routes to achieve automatic driving. How does 5g enable automatic driving when all the automatic driving schemes are mature? This starts with the problem that autonomous driving can't solve. 

Taking the location problem as an example, the current mainstream of automatic driving positioning scheme is to give a rough positioning by the principle of 
 triangle positioning 
 of GPS system, and then use the sensor's perception of the environment (lane line, road boundary, feature points, etc.) to match the corresponding environmental features in high-precision map to achieve high-precision positioning. 


 http://slideplayer.com/slide/5717261/ However, GPS antenna will not receive satellite signal in some seriously blocked scenes, such as under the tunnel, basement and viaduct. If the autopilot system starts in such a scenario, it will not be able to get rough positioning information. The main reason is that the signal source of GPS system is in the sky, once blocked, it will lose the signal source. 

 so how can 5G solve the problem of the location of autopilot under the tunnel, underground storage and viaducts? It depends on the high precision positioning technology based on 5g. 

 5g uses new coding methods, beamforming, large-scale antenna array, millimeter wave spectrum and other key technologies, which has large bandwidth, is conducive to parameter estimation, and provides support for high-precision distance measurement. By introducing large-scale antenna technology, the base station can be equipped with 128 antenna units, which provides the basis for high-precision angle measurement. 5g will realize dense networking, significantly improve the density of base stations, and user signals can be received by multiple base stations at the same time, which will be conducive to the cooperation of multiple base stations to achieve high-precision positioning. 

Reference source: 
 new mobile communication technology, realizing indoor high-precision positioning in 5g Era


It can be seen from the references that the key factor to realize high precision positioning with 5g is the 5g base station with high density of 
 and 
. Through the communication between 5G antennas and multiple base stations, when knowing the true longitude and latitude position of the base station, and with the low delay of 5G, the position and posture information of the self driving vehicle can be solved in real time. The most important point is that it is no longer restricted by occlusion. 

For autonomous driving, perception problem, in a sense, is the location problem of traffic participants. When the 5g based real-time high-precision positioning technology is mature and widely used, the traffic participants in the automatic driving environment can share their own positioning in real time and interact with other traffic participants, and many perception problems caused by occlusion will be solved. 

5g's high-precision positioning technology will not only promote automatic driving, but also have a far-reaching impact on commodity guidance, personalized advertising push, disaster rescue and other industries in indoor scenes, so it will not be launched here. 


 
 autopilot maturity period 
 


Once autonomous driving enters the mature stage, technical problems will no longer be obstacles. At this time, we need to focus on commercial issues - how to use existing technology to reduce costs and improve profits on the premise of ensuring functions. 

 autopilot system costs mainly in two blocks. One is the vehicle sensor mainly composed of camera, lidar and millimeter wave radar, the other is the domain controller of automatic driving, which is often referred to as the computing platform. 

No matter how to reduce the cost, we can't pay attention to the sensor, because the sensor is the input source of sensing information. Then we can only find a way from the domain controller of autopilot. 

Reference here 
 @ Zhuge Youyu 
, that is to say, the higher the level of automatic driving, the higher the requirement of computing power. The computing power is not only reflected in the more powerful CPU and GPU, but also requires more stable power drive and more reasonable cooling structure. Therefore, the domain controller occupies a large part of the cost of the automatic driving system. 


 
 reference source: https://www.zhihu.com/question/376327323/answer/1064216156 The price performance ratio of the domain controller is very low. It has very powerful computing resources, but it can only be used when driving automatically, and the rest of the time is idle due to parking. From the perspective of resource utilization, this is a great waste. 

In addition, with the higher resolution of the vehicle camera, the more laser harness of the lidar, the more data the vehicle end needs to process, and the higher calculation force requirements. In the future, even if the domain controller is mass-produced, its cost is not easy to reduce. As shown below, it is NVIDIA's continuous 5 generations of automatic driving domain controller with increasing computing power (a PX2 needs 200000 RMB when it is just launched). 


 https://www.fudzilla.com/news/45943-nvidia-announces-automotive-soc-orin In the face of the high cost of domain controller, how to use 5g technology to solve this problem? Here is one of the ways I envision. 

 is equipped with a low cost controller to deal with emergencies on the autopilot car. The computing resources with huge computing power are deployed in the cloud. When the vehicle is driving, the computing resources are applied to the cloud, and the resource is released when the automatic driving or stopping is stopped. This way can not only reduce the hardware cost of the self driving car itself, but also achieve the multiple use of cloud resources. The core idea of this approach is similar to the hot "cloud computing" at this stage. 

However, there is a defect in the current "cloud computing". For large-scale data, it is necessary to upload the data to the platform in advance, and then process it, which is difficult to achieve real-time. borrow

Comment

No Comment Yet.