Shulou(Shulou.com)06/03 Report--

Https://www.toutiao.com/a6679370418282824204/

With the passage of time, more and more human beings hand over mechanical objects to the machine system in modern life, and mobile cars are no exception, from the common control of auxiliary driving to the authorized control of self-driver. the test is whether the machine system can face and adapt to the outside non-institutional environment.

From the point of view of demand, urban residents' demand for self-driving exists, but they can not fully trust its safety and reliability. In the meantime, automakers' ability to limit self-driving because of possible legal liability has led to delays in the launch of high and new technologies.

Let's start with self-driving technology, talk about the demand of urban residents for self-driving, and the importance of supervision behind self-driving.

What is a driverless car?

Driverless vehicle (Self-driving Car) is a kind of outdoor wheeled mobile robot, which relies on the cooperation of artificial intelligence, sensor, positioning system and navigation system, so that the computer can operate the motor vehicle automatically and safely without any human active operation, which brings a new experience for human traffic safety and efficiency.

The evolution of self-driving is a process in which human beings gradually hand over the control and improve the safety factor.

The stage of self-driving evolution is a process in which the control of the vehicle is gradually handed over to the computer system by people, as shown in the picture above.

The self-driving distance is on the market, and there is also the problem of the supervision system.

It is also a process of increasing security (as shown in the figure above, from passive security to active security and then to preventive security).

Combined with the current general prediction cycle of the industry, Yiou think-tank analysis and judgment, some self-driving is expected to begin commercialization around 2025, completely self-driving commercialization will not be until after 2025, but before that, ADAS

(Advanced Driver AssistantSystem, Advanced driving Assistance system) will play an important role.

Technologies involved in driverless vehicles = environment awareness + positioning navigation + path planning + decision control

The technologies involved in self-driving can be divided into two levels: perception and decision-making. as shown in the following figure, on the one hand, the sensor data is used to obtain local data (data of the vehicle itself and its surrounding environment). On the other hand, the global data is constructed by combining high-precision map and weather data. The integration of the data will coordinate the application with the decision-making layer, assist the system to do positioning and navigation, and then combine the algorithm model to do path planning, control the steering and speed of the vehicle, and realize driving automation. The data obtained by the decision-making layer will also be fed back to the high-precision map.

Environment awareness level = perception of local data + assistance of global data

The perception function of the vehicle is mainly through the sensor to obtain data. The sensor is equivalent to the eyes of the driverless car and is used to observe the dynamic changes while driving. It is an indispensable and important part of the driverless vehicle. The commonly used sensors include camera, lidar, ultrasonic radar, GPS, gyroscope and so on. Camera and lidar are the two most important sensors.

Camera

At present, shooting through the camera, identifying the image and video, and determining the environment in front of the vehicle is the main perception way of self-driving cars, which is also one of the main research and development contents of many self-driving companies. As a widely used sensor, camera has the characteristics of low cost, large amount of information collection and so on. At present, car cameras are mainly divided into monocular and binocular.

Monocular camera, mainly based on the principle of machine learning, using a large number of data for training, can obtain road images, extract lane lines, and identify the environment. Although it needs a lot of data support and does not perform as well as binocular cameras in bad light conditions, its relatively cheap price and mature technology are also favored by some companies. On the other hand, based on the principle of parallax (as shown in the following picture), the binocular camera can measure the environment in front of the vehicle (trees, pedestrians, vehicles, potholes, etc.) in the case of insufficient data, and obtain accurate distance data. then the algorithm enhanced adjustment is used to obtain the depth of field of the surrounding environment, which is used to provide vehicle control to the unmanned driving system.

Radar

The working principle of lidar is to transmit electrical pulses into light pulses through the transmitting unit, and then the receiving unit restores the light pulses reflected from the target into electrical pulses. By calculating the time difference between the transmitted signal and the received signal, the relative distance between the contour edge of the object and the equipment in the field of view can be accurately measured. The contour information forms the so-called point cloud and draws a 3D environment map with an accuracy of centimeters. This is shown in the following figure.

Lidar has a long penetration distance, and high-performance lidar can reproduce 3D scene with precision of up to centimeter within 200 meters, thus helping the unmanned system to plan the driving route in advance. At present, multi-line lidar is likely to be a necessary sensor for unmanned vehicles in the future, and is closely related to high-precision maps and core algorithms of driving systems.

At present, there is no mature mass production scheme for multi-line lidar at the vehicle gauge level. although mechanical rotary multi-line lidar has been widely used, it is large and expensive. Smaller and lower-cost pure solid-state lidar has not yet seen mature products.

Millimeter wave radar, ultrasonic radar: in addition to lidar, in recent years, millimeter wave radar and ultrasonic radar have gradually become self-driving vehicles, participating in multi-sensor information fusion sensing equipment. Among them, the most famous example is Tesla's scheme of millimeter wave radar + camera instead of using lidar at all in his smart car. In addition, intelligent driving giants such as Bosch and mainland China also have more profound technical accumulation and application experience in low-cost sensor devices such as millimeter wave radar and ultrasonic radar. In China, millimeter wave radar manufacturers such as Xingyi Road are also actively carrying out technological development to catch up with the level of international giants.

Self-driving positioning and navigation

Self-driving accurately perceives its relative position in the global environment through positioning technology, regards itself as a particle and organically combines itself with the environment. Navigation technology helps driverless cars "know" their speed, direction, path and other information.

In practical application, the two are combined by information fusion technology, so that the environmental information and body information are integrated into a systematic whole.

Among them, high-precision map is the basis of self-driving navigation and follow-up path planning. in recent years, satellite navigation and 3D environment modeling based on lidar have become more and more mature, and the quality of high-precision map mapping has been gradually improved. this provides a great help for the research and development of self-driving. Domestic high-precision maps, with Baidu Map, Amap, Siwei Tuxin and other companies as the main force, while abroad, Here, TomTom and other companies have been praised.

Self-driving route planning, decision control

Path planning technology can provide the optimal driving path for self-driving. In the process of driverless vehicle driving, starting from the travel demand, on the basis of high-precision map, a collision-free and passable path from the starting point to the target point is drawn according to the road network and macro traffic information (including the calculation of road length, speed, road section grade, traffic port waiting time, etc.). Then according to the vehicle in the driving process of the local environment data, their own state data to make the optimal path choice. Thanks to the lidar, the algorithm can plan paths on larger scales, slowly changing maps and longer paths, as shown in the following figure, without waiting until the last minute to find a problem with the path.

The algorithm provides the underlying support for self-driving technology to detect and track dynamic obstacles.

In the perception level of self-driving, deep learning mainly deals with the local data collected by camera and radar (combined with global data), based on the rich information of dynamic images and the difficulty of manual modeling. Deep learning can maximize its advantages.

At the decision-making level, the first thing to solve in the research process of self-driving technology is safety, but lidar can only provide sparse environmental information, while self-driving on the road is facing a dynamic change. therefore, to improve the accuracy of dynamic obstacle detection and tracking and reduce the false detection rate is an urgent problem to be solved in the environmental perception of self-driving vehicles.

In order to avoid collision with dynamic obstacles during driving, the self-driving system needs the assistance of algorithms to achieve the following three conditions:

First of all, it is necessary to reliably detect the dynamic obstacles that have an impact on driving, which requires the sensor to accurately measure the position changes of the obstacles and be able to extract the features of the obstacles for the matching between obstacles at different times, so as to complete the tracking of the same obstacle.

Secondly, the motion path of dynamic obstacles must be predicted.

Finally, it is necessary to identify the types of dynamic obstacles, and different obstacles have different motion characteristics, which directly affects the obstacle avoidance strategy adopted by driverless vehicles.

In addition to perception and decision-making, self-driving also involves vehicle control, automotive dynamics, automotive engineering and many other technical disciplines, but also needs the support of vehicle control (brake, steering, lighting, throttle, etc.) accessories.

Self-driving enterprise map

Automation, as one of the future urban mobile travel trends, helps the supply side to cope with the demand on the demand side.

The demand side and supply side of urban mobile travel in the future

Urbanization and population growth will increase the average urban population density by at least 30%. To this end, the demand for mobility in densely populated cities will double (if the per capita mileage remains stable and the ratio of car ownership to GDP growth remains at historical levels). There is no doubt that the demand for mobility has doubled, and the traffic jams (especially during commuting hours) will greatly reduce people's traffic efficiency.

From the perspective of residents of (China's first-tier cities), there is traffic efficiency on the one hand and safety and reliability on the other. Then there is global regulation of emissions based on livability and sustainability, as well as support for renewable energy, in an attempt to improve air quality.

In its "Prospect for Future Travel (Mobility)" report, McKinsey puts forward three mobile trends of electrification, sharing and automation, which are judged according to the specific urban conditions in China (urban population density, economic development, road infrastructure, etc.). In the future, Chinese cities (which will be reflected in the super-first-tier cities such as Beijing, Shanghai and Shenzhen in the relatively short term) will gradually transition from "clean energy and sharing systems" to "seamless mobility". Sharing has become the backbone of the public transport system, while electrification and automation are the technological upgrading of these shared vehicles for the sake of improving air quality and traffic efficiency.

Autopilot is to access big data into the original vehicle system to achieve automation. as shown in the picture above, automation, as one of the future urban mobile travel trends, helps the supply side to cope with the three major demands of the demand side.

Three business models for the future of self-driving

From the perspective of business model, after the future technology is mature, self-driving cars will be commercialized in the form of "selling products" or "providing services" when they are launched to the C-end. Sell to high net worth people as private self-driving vehicles in the form of products, while the services provided can be divided into B2B2C (the middle B side serves as a shared mode vehicle operator, providing self-driving vehicles to C side), and B2B2B (relatively closed, uncomplicated road conditions, such as trucks and trucks are handed over to the driving system on the middle highway).

At present, scenes with relatively uncomplicated road conditions, such as highway sections, are commercialized faster. At present, society is in the stage of transition from auxiliary driving ADAS to partial self-driving and completely self-driving. From the point of view of the demand of urban residents, for the sake of "improving traffic efficiency" and "improving air quality", they all hope that self-driving technology will be put on the market as soon as possible, but they are hesitant whether self-driving is "safe and reliable".

Nancy G. Leveson, a professor at MIT Massachusetts Institute of Technology in the United States, pointed out that the safety problem of ADAS is not in the individual program components, but in the integration of the system. A research report by the German T ü V Safety Certification Agency pointed out that when the driving assistance system begins to show some (semi) automatic behavior, it will sometimes be accompanied by a number of unstable "unnecessary system behavior". In serious cases, there will be consequences that threaten personal safety.

According to the current research, the transition from assisted driving to self-driving is a process of constantly improving adaptation to the unstructured environment, in which there are hidden dangers and hidden dangers caused by errors: 1, hackers invade the intranet or unduly interfere with vehicle sensors; 2, security accidents caused by incomplete understanding of environmental conditions.

Under the trend of data opening and sharing, such as vehicle networking, the safety hidden danger coefficient of type 1 is also increasing. the second type of error can be subdivided into: (1) safety hidden dangers such as the sudden start of the vehicle active braking system for no reason; (2) the safety hidden danger caused by the wrong "classification" and "understanding" caused by the technical level of the system is not reached. (3) there is unknowability in the self-driving system using machine learning, which may lead to the final behavior deviating from the expectations of the automaker. These hidden dangers may cause or directly lead to traffic accidents at any time.

Except in the mode of human and machine control, it is difficult to judge common negligence. In fact, in the process of transition from auxiliary driving to partially self-driving and completely self-driving, there is a trend that the responsibility of traffic accidents has shifted from human beings to car manufacturers, then out of consideration of huge legal liability, self-driving car manufacturers may consider limiting car capacity because of safety risks, resulting in high and new technologies not being fully put into the society.

In the final analysis, whether the driverless car can successfully enter the society or not does not depend on the technological maturity, but on the consideration of bottom-up social acceptance and top-down policy and legislative regulation. As Dr. Medford, head of safety at Google's driverless car program, puts it, "even the best car safety technology can't guarantee saving every life. The limit to the effectiveness of safety technology is the way people use it (or don't use it)."

The self-driving distance is on the market, and there is also a problem with the regulatory system.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.