In terms of road collaborative management, the accurate and effective prediction of vehicle traffic flow plays an important role in the construction and planning of signal intersections. The application of the AI prediction model in traffic flow performance prediction achieves positive results. However, there is still great uncertainty in determining which AI approach can effectively solve traffic congestion problems. Isaac Oyeyemi Olayode ^[10][36] et al. compared an artificial neural network trained by a particle swarm optimization model (ANN-PSO) and a heuristic artificial neural network model (ANN) for vehicle traffic flow prediction, using the South Africa transportation system as a case study. Results show that the ANN-PSO model is more efficient than the neural network model in predicting vehicle traffic flow at four-way signal intersections, and is robust enough to predict traffic flow. This research idea provides traffic flow information and guidance for the collaborative control system to optimize its travel time decision. In the multi-vehicle queue forming control system, the system needs to be specially designed when individual vehicles need to leave or join the fleet. Ying ZB et al. ^[21][65] added a dynamic AVP management protocol to the control system for how to effectively manage the addition of vehicles and the leaving of vehicles in the vehicle queue. Vehicles joining and leaving the queue will need to communicate with the queue leader, and all the messages will be related to the corresponding transactions specified by the smart contract.

2.3. Collaborative Positioning

3.3. Collaborative Positioning

The vehicle–road collaboration technology in the automotive internet integrates modern communication technology and network platforms. Through information sharing among vehicles, roads, and people, it realizes complex environmental sensing, collaborative decision making, and intelligent control functions to build a safe, comfortable, and energy-efficient automotive internet platform. Vehicle GPS technology is usually used to achieve positioning. Due to signal blockage and the multiplex effect, GPS positioning technology often suffers from missing signals or insufficient accuracy to achieve lane-level positioning accuracy and cannot meet the requirements of vehicle–road collaboration applications ^[22][74]. Cooperative positioning (CP) technology is another method to improve the positioning accuracy of vehicle–road collaboration networks ^[23][75]. Currently, using various methods to obtain more and more meters or even centimeter-level high precision position information research, based on the global navigation satellite system positioning technology, based on computer vision sensor positioning technology, based on the laser radar sensor positioning technology, and based on super broadband signal positioning technology, the four methods are the mainstream of the high-precision positioning technology route. The global navigation satellite system is the existing and widely used positioning technology in the field of road traffic, and the GPS, Beidou, and other systems are integrated to improve the positioning accuracy and reliability of the positioning system. Zeng Qinghua and others of Nanjing University of Aeronautics and Astronautics proposed that the positioning method of multi-constellation-combined navigation can improve the accuracy of users. ^[24][76]. Robert Odolinski and others of Otago University also did relevant research in order to reduce the cost of RTK and improve the positioning accuracy, and proposed that ^[25][77] uses the measurement antenna to improve the positioning accuracy of the receiver. However, the impact on the interference of streamers and tall buildings on satellite signals is ^[26][78], and satellite signals are vulnerable to road conditions and weather, which will cause signal drift and signal loss, affect normal driving, and even cause safety accidents, which cannot meet the conditions of the high reliability requirements of the positioning system. Computer visual positioning can be divided into: monocular visual positioning navigation, binocular visual positioning navigation, and multiocular visual positioning navigation ^{[27][28][29][30]}[79,80,81,82]. These positioning and navigation technologies achieved good results in the research of visual positioning. However, bad weather can lead to poor work normally, and existing technology can only well solve the identification of specific targets. The complex scene of the social road cannot well identify any problems, nor can it meet the requirements of real-time and the reliability of the positioning system. Lidar positioning technology uses adjacent point cloud data to derive the rigid body transformation ^[31][83] between two adjacent frames through feature extraction and a registration algorithm. Compared with other sensors, lidar has incomparable advantages in the unmanned positioning system, and the positioning algorithm, based on lidar sensors, plays an important role in the intelligent driving positioning module of ^[32][33][84,85]. Ultra wiband technology is a new wireless communication technology. Its positioning technology has low system complexity, low power consumption, good anti-interference ability, high multi-path resolution, and high positioning accuracy ^[34][35][86,87]. Kegen et al. ^[36][88] proposed that in order to locate and track the target, we applied the Kalman filtering algorithm in an ultra-wideband positioning system is applied. Car–car and car–road information collaborative interaction drives multiple collaborative positioning applications, for example, car–car collaborative positioning ^[37][89], based on roadside positioning enhancement ^[38][90], car–car collaborative integrity monitoring ^[39][91], and beyond visual distance detection perception ^[40][92]; at the same time, it also greatly expands the traditional bicycle autonomous positioning weight calculation based on the perceptual range in the weight distribution and actual satellite observation quality level to establish closer correlations. Liu ^[41][93] et al. established the overall framework of vehicle satellite positioning and collaborative positioning enhancement based on vehicle–road information interaction for the tracking and adaptation of navigation satellite positioning and weight allocation in a complex dynamic operating environment. Based on high-precision mapping and multi-sensor fusion positioning technology, Yao ^[42][94] and others complement the advantages of various sensing and positioning methods, such as the global navigation satellite system and roadside multi-sensor sensing, and realize the continuous tracking and high-precision positioning of vehicles in urban ground and underground garage scenes. In view of the problem of large positioning errors of unmanned vehicles in unstructured scenarios, combined with on-board lidar and a roadside binocular camera, the dual-layer fusion collaborative positioning algorithm is adopted to achieve high-precision positioning.

34. Vehicle Communication

As one of the key technologies in this field, vehicle communication refers to the use of wireless communication, physical terminals, and intelligent sensors to realize V2V, vehicle–road communication (V2I), to enhance traffic efficiency, improve traffic safety and travel experiences, and construct the vehicle network (IOV) or vehicle self-organization network (VANET). The frequency band of on-board communication is mainly divided into low frequency, intermediate frequency, and high frequency. The application representative of low-frequency technology mainly includes automobile anti-theft and keyless systems. The product application of this technology mainly includes vehicle remote control keys; high-frequency communication mainly includes Bluetooth communication, mobile communication, dedicated short-range communication (DSRC), ultra-wideband (UWB) communication, etc.

3.1. Communication Security

4.1. Communication Security

With the rapid development of wireless communication technology, intelligent vehicle communication is becoming more vulnerable to potential security attacks ^[43][105]. Due to the openness of wireless channels, the signal exposed in the open environment is likely to be stolen, interfered with, or even modified by the attacker ^[44][106]. If the attacker maliciously impersonates the vehicle to release false information and mislead other vehicles to form incorrect judgments, serious consequences may result. At present, the homomorphic aggregation scheme ^[45][112], elliptic curve encryption algorithm ^[46][113], and Chinese residual theorem ^[47][114] are often used to achieve data aggregation. The homomorphic encryption algorithm is the most commonly used technical method of data aggregation ^[48][115] because it satisfies the cipheric homomorphism operation properties. Homomorphic encryption (HE) refers to the specific calculation of the ciphertext after homomorphic encryption, and the ciphertext calculation results are equivalent to the same calculation ^[49][116] after the corresponding homomorphic data. Although the scheme in literature ^[50][51][117,118] can meet the basic privacy protection requirements and efficiency requirements in the internet of vehicles, it still has some deficiencies. The homomorphic encryption algorithm used in the scheme includes the elliptic curve-based encryption algorithm and the Paillier homomorphic encryption algorithm. The scheme based on the Paillier homomorphic encryption algorithm is not based on an elliptical curve. To protect the privacy of the internet of vehicles, there should be the following points: First, ensure that all the vehicles received by the nodes in the vehicle ad hoc network send and receive messages can be verified. The authentication is a method to determine whether the information received is true when the receiving vehicle successfully receives the information sent to it, as well as a method to determine whether the vehicle sending the message is a registered vehicle in the network. Second, ensure the integrity of the messages in the vehicle self-group network. Integrity refers to the message, from sending to receiving, not being tampered with by unauthorized vehicle nodes, add, delete, or packaging. However, a message integrity defect is not necessarily caused by an attacker, but also may be caused by the roadside units, relay vehicle node, routing, and other network line hardware or software equipment failure leading to timeout, packet loss, and other phenomena. Third, ensure that the communication of vehicles in the ad hoc network is confidential. Communication confidentiality means that only the parties sending and receiving the information can correctly interpret the information content, and other vehicle nodes or devices that relay the message cannot obtain the true content of the message. Fourth, ensure the traceability of messages in the vehicle ad hoc network. Traceability means that any message sent, received, relayed, or forwarded by the sender, relay, or receiver of the information will be recorded and retained. The vehicle cannot change or delete the sending, receiving, or relay records, nor can it deny the message sent, received, or relay by itself. Traceability plays an important role in investigating traffic crimes and confirming vehicle tracks. Fifth, it must be ensured that the spatiotemporal correlation of the vehicle’s location is cut off for the attacker in the network. The spatial and temporal correlation of the location is cut off, which means that the attacker cannot know the name, position, and corresponding time of the vehicle at the same time. Once the attacker knows the three points, the location information of the owner or the vehicle can be judged according to the spatial and temporal correlation of the vehicle. There are two kinds of types of privacy protection for in-car ad hoc networks, namely the protection of information privacy and location privacy.

3.2. Control Strategy for Communication Delay

4.2. Control Strategy for Communication Delay

At present, the limited computing resources of vehicles are unable to meet the computing resource requirements of many delay-sensitive messages ^[52][124]. In order to cope with the expanding computing requirements of this vehicle terminal, the existing cloud computing technology can process a large amount of data information, which effectively reduces the local computing burden to a certain extent. However, when the security message is sent to the cloud server for processing through the core network, the processing delay of the message may be greatly affected. At the same time, the transmission of security messages often displays the phenomenon of redundant propagation, which causes a broadcast storm during message transmission and leads to the poor performance of message transmission. Generally, the transmission of emergency messages involves directional propagation, and the emergency messages are broadcast to the farthest receiving vehicles within the communication range ^[53][125]. The existence of communication delay will affect the following performance of the vehicle, and even threaten the driving safety. Therefore, it is necessary to design ^[54][126] for CACC control strategy for communication delay. On the one hand, the internal performance of the network can be studied from the perspective of communication, and the communication efficiency and quality can be improved by designing a reasonable network scheduling algorithm and communication protocol, so as to reduce its adverse impact on the control system. On the other hand, the optimization design of the control strategy under limited communication restriction gradually became a research hotspot, and a variety of solutions ^{[55][56][57][58]}[127,128,129,130] were formed.

45. Test Method and Evaluation

4.1. Real Vehicle Road Test Platform

5.1. Real Vehicle Road Test Platform

The real vehicle road test of an intelligent network includes three test scenarios: a closed scene test, a semi-open road test, and an open road test. Firstly, the representative closed test sites in foreign countries include the following: the Asta Zero test site in Sweden, which contains complete experimental facilities and has the capacity to test vehicle dynamics, driver behavior, and V2X technology; the Mria City Circuit test site in the UK, whose main features are simulated signal masking and various V2X communication facilities, with the flexible design of traffic lights and transmitting towers, and is oriented towards the testing of intelligent transportation systems and intelligent networked vehicles ^[59][158]; the Willow Run test site in Michigan, U.S., which is suitable for the extreme testing of V2X technology and autonomous driving technology, etc. ^[60][159]. The closed test sites also occupy a large proportion of the domestic real-world test sites. On July 30, 2021, the Ministry of Industry and Information Technology, the Ministry of Public Security, and the Ministry of Communications jointly issued the “Management Specification for Road Testing and Demonstration Application of Intelligent Connected Vehicles (Trial)” ^[61][160], adding the demonstration application of manned vehicles and special operating vehicles, and opening some expressway tests. At present, there are around 50 test sites in a state of completion or under construction, 30 of which are equipped with testing capability for intelligent networked vehicles. The intelligent driving test base of the Ministry of Transport in Beijing contains traffic scenes of various road shapes and surfaces, such as urban and rural roads, high-speed roads, and their ramps, and is equipped with street lights, traffic lights, weather simulation equipment, and other facilities, which can be used for intelligent driving, intelligent road networks, and other tests. Changsha, Wuhan, and other cities within intelligent network pilot demonstration areas can simulate a variety of road conditions, including wet roads, mountain roads, woodlands, high-speed roads, masonry, bridges, etc.; equipped with intelligent sensors and other monitoring equipment, they can be used for intelligent networked vehicle testing. Secondly, the domestic open test sites mainly include the following: the Shanxi (Yangquan) autonomous driving vehicle–road cooperation demonstration area, where roadside sensing, collection, and transmission systems are being built, and the deployment of the vehicle–road cooperation cloud control platform and autonomous driving vehicle supervision platform based on Baidu’s public cloud was completed, so as to realize object detection technology based on environmental sensing and V2X communication technology to support L4-level autonomous driving vehicle over the horizon, etc. The Yongchuan Baidu Western Autonomous Driving Open Test Base is an open test and demonstration operation base for L4-level autonomous driving. The base deploys 5G communication in the road network environment in all aspects and includes more than 30 typical open road test scenarios in the mountain city, such as interchanges, tunnels, and bridges, with the unique traffic terrain of Chongqing, and it can accommodate 200 intelligent driving vehicles for testing at the same time. The efficient data transmission based on 5G technology and intelligent data processing based on artificial intelligence technology are of great significance for the application of intelligent connected vehicles. The Changsha test demonstration area has a wealth of intelligent connected vehicle test scenarios, and 5G widely covers the intelligent connected vehicle test area. It is the first ^[62][161] demonstration area to carry out high-speed tests and manned tests in China.

4.2. Virtual Test Platform

5.2. Virtual Test Platform

The simulation test experiments of intelligent network connected systems seek to establish a real static scene and carry out dynamic scene modeling according to the actual situation through computer simulation technology, so as to realize the model and algorithm of the network-connected vehicle. The system can carry out a variety of test verification experiments in the simulated traffic scene, reducing the dependence on real vehicle experiments to a certain extent, such as the current mainstream of intelligent driving vehicle sensor physical model simulation authenticity, intelligent transportation system V2X model construction, and the construction of dynamic traffic flow in the simulation scene. At present, a new generation of intelligent driving simulation systems integrating physical characteristic information is being gradually developed, which cannot only verify and iterate intelligent driving algorithms more effectively, but also meet the overall test requirements of intelligent driving simulation platforms for physical information systems more comprehensively. With the rapid development of advanced driver assistance systems (ADAS) and intelligent driving, the development of simulation software underwent the following stages: The early simulation test software mainly used dynamic simulation. The Simulink module of control design simulation software MATLAB was used to build the vehicle dynamics model and carry out real-time simulation, such as Carmaker ^[63][162], CarSim, Panosim ^[64][163], and other simulation software. With the further development of the ADAS function, simulation and test software to assist this function began to appear, such as Prescan ^[65][164]. In recent years, the ability of unreal engines to restore the virtual environment became stronger, and more researchers paid attention to the open-source simulators, such as AirSim ^[66][165], CARLA ^[67][166], and Unity. However, due to the diversity and functional complexity of the current mainstream intelligent driving simulation tools, most of them cannot support multi-agent co-simulation or simulation in large-scale scenarios, so the test verification of the off-vehicle networking system in large-scale complex traffic environments is not yet realized. At present, the optimization control of urban traffic in intelligent vehicle network systems is mainly concentrated within single-point traffic control, lacking real-time linkage control. However, the simulation of intelligent driving in large-scale open scenes shifted from macro-centralized control to meso-edge-side coordination control, and then to micro-single-car intelligent control. Therefore, it is necessary to develop simulation tools that can support this hierarchical control model. A microscopic traffic flow simulation model takes vehicles as the description unit, which can describe in detail the car-following and lane-changing behavior between vehicles. At present, the more common ones include SUMO ^[68][167] and PTV-Vissim ^[69][168], both of which have corresponding interfaces for secondary development. An intelligent driving simulation simulator based on SUMO and Unity3D ^[70][169] was proposed through the TraCI protocol. It can be seen that the use of intelligent driving simulation tools to provide single-vehicle intelligent decision making, traffic flow simulation models to provide specific traffic scene modeling and design, and the combination of the two through an interface, can support intelligent networked vehicle collaborative simulation in large-scale open scenes.

4.3. Test Method and Evaluation

5.3. Test Method and Evaluation

The test and evaluation of intelligent connected vehicles is an important stage in the development of its vehicle functions. Similar to human drivers, the test method can be divided into three parts: the perception function test, the decision function test based on perception information, and the action function test. Vargas et al. ^[71][174] proposed a conceptual sensor testing framework for automatic driving vehicles, which is oriented towards different types of sensors and communication mechanisms, and provides a means of performing test scenarios similar to those occurring in the physical world. Wei et al. ^[72][175] implemented a parallel computing framework and system for intelligent driving tests and verification. It constructs a set of intelligent test models, which enables the system to develop a cognitive mechanism of automatic self-upgrading under the guidance of human experts, and further improves the ability of intelligent driving vehicles to adapt to complex environments. In order to speed up the scene testing, virtual environment simulation can be used. The mainstream virtual environment simulation software includes Prescan, Carmaker, dSpace, etc. At present, most of the tests only cover intelligent driving, and the test methods for intelligent driving vehicles in static, dynamic, and uncertain environments are not perfect. The vehicle decision-making ability test and V2X-based traffic integration test also need to be improved. Domestic and foreign scholars and institutions carried out a lot of basic research work in simulation test methods, test scenarios, simulation modeling, tool chain reliability, and other aspects of ^{[73][74][75][76]}[176,177,178,179]. For the different stages of the product, different types of simulation and simulation test methods are not only applied to the development process of intelligent and connected vehicles, but also gradually play an important role in the product verification, confirmation, and evaluation. The UN is in the automatic lane keeping system for L3 automated lane keeping systems (ALKS)-type approval regulations for test verification, and put forward the relevant requirements for the simulation tools and models ^[77][78][180,181]. Japan explicitly introduced the software in the rings in its software-in-the-loop (SIL) and hardware-in-the-loop (HIL) to test the ^[79][182]. The new test assessment method proposed by the United Nations Informal Working Group on Autonomous Driving Verification Methods requires the use of a proven simulation tool chain to conduct simulation tests to evaluate the safety of autonomous driving systems, and proposes SIL testing for driving safety and critical safety scenario assessment ^[80][183]. In the draft regulation on the type approval requirements of autonomous driving systems, the EU made it clear that simulation, closed sites, and practical law can be adopted, and roads introduced the United Nations study of ^[81][184] on the credibility of simulation tests. The vehicle dynamics simulation model and test methods developed by ISO provide the basic ^[82][83][185,186] for the vehicle dynamics simulation test and verification.