Signal-free intersection cooperative passing real vehicle test method based on double-layer control architecture

Abstract

The invention relates to a non-signalized intersection cooperative passing real vehicle testing method based on a double-layer control framework, and belongs to the technical field of automatic driving testing. The method comprises a theoretical passing algorithm process and a real vehicle testing process. The theoretical part adopts a double-layer control framework, and the upper layer predicts the arrival time based on networked vehicle state data collected by an RSU, determines the passing priority in sequence, calculates a conflict point occupation time period and an entry forbidding time slot set, and allocates expected speed and road right information; and the lower-layer vehicle receives information through V2X, and autonomously plans the speed by using a DDPG algorithm, so that the oil consumption is reduced, and the passing efficiency is improved. The real vehicle test part is deployed at a real urban intersection, communication equipment deployment and experimental steps are designed, a test environment can be quickly established at low cost, the effectiveness, robustness and engineering feasibility of a control algorithm in a real environment can be comprehensively checked, real vehicle test verification is effectively supported, and intersection traffic efficiency is improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of automatic driving test, and relates to a cooperative passing test method for a signal-free intersection based on a double-layer control architecture. BACKGROUND

[0002] In recent years, with the rapid development of connected and automated vehicles (CAV) and vehicle-road coordination technology, cooperative passing control of signal-free intersections has become an important research direction in the field of intelligent transportation, and plays a key role in improving road network passing efficiency and ensuring driving safety. Through devices such as road side units (RSU), on-board units (OBU), and connected vehicle centers (CVC), the CVC can obtain real-time accurate position, speed, acceleration, and other state information of connected vehicles, and based on cooperative perception and decision-making algorithms, realize intelligent regulation and control of CAV driving trajectory and traffic signals. Traditional signal light control mainly uses a preset fixed timing scheme, which is difficult to adapt to the dynamic changes of traffic flow and is prone to waste green light time or vehicle queuing congestion. Since signal-free intersections lack centralized coordination mechanisms, they mainly rely on the experience and autonomous decision-making of drivers, which can easily cause vehicle conflicts, passing delays, and even traffic accidents due to improper passing and yielding. Especially in mixed traffic environments, the interaction between human-driven vehicles and automated vehicles is more complex, further increasing the difficulty of managing signal-free intersections. Existing researches mainly focus on joint optimization and utilization of vehicle trajectories and traffic signals in a fully connected environment, and use simulation for testing and verification.

[0003] Currently, the research on cooperative control of signal-free intersections mainly relies on simulation verification, and lacks real vehicle test verification. Although existing algorithms perform well in simulation, they do not fully consider complex factors such as communication delay, sensor error, and mixed traffic flow in real scenarios, and lack standardized real vehicle test schemes.

[0004] Existing researches mainly focus on joint optimization and utilization of vehicle trajectories and traffic signals in a fully connected environment, but have problems such as high computational complexity, insufficient real-time performance, and lack of real vehicle verification in real mixed traffic environments. Moreover, most of the researches are only implemented on simulation platforms, and the deployability of the algorithms on real vehicle platforms has not been fully verified. Due to the immaturity of intersection cooperative passing algorithms and the complexity of large-scale real vehicle deployment, there are few real vehicle verifications of signal-free intersection cooperative passing algorithms at home and abroad. SUMMARY

[0005] In view of this, the purpose of this invention is to provide a real-vehicle testing method for cooperative passage at unsignalized intersections based on a two-layer control architecture, which aims to improve the efficiency of cooperative passage of vehicles at intersections, ensure driving safety, and provide a specific real-vehicle testing method to overcome the shortcomings of existing technologies in supporting real-vehicle testing and the inability of algorithms to be implemented.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A real-vehicle testing method for cooperative passage at unsignalized intersections based on a two-layer control architecture is disclosed. This method includes at least a theoretical passage algorithm design process and a real-vehicle testing process. In the theoretical passage algorithm design process, a two-layer control framework is adopted, with the upper layer responsible for vehicle right-of-way allocation and the lower layer responsible for vehicle speed control. In the real-vehicle testing process, communication equipment is deployed for the theoretical passage algorithm, a real-vehicle testing environment is built, and the effectiveness, robustness, and engineering feasibility of the theoretical passage algorithm are verified. The upper-level right-of-way allocation is based on the real-time network vehicle status data collected by the RSU, predicting the arrival time of each vehicle at the intersection, and determining the passage priority according to the arrival order of the vehicles; it calculates the time period occupied by the conflict point of vehicles within the intersection, and calculates the set of prohibited entry time slots at the starting line of the intersection based on the spatiotemporal projection algorithm; based on this, it determines the feasible time period of the vehicles and allocates specific expected speed and right-of-way information to each vehicle; the lower-level vehicle control receives the speed and right-of-way information sent by the upper layer based on the vehicle-to-everything (V2X) communication, and uses the Deep Deterministic Policy Gradient (DDPG) algorithm deployed on the vehicles to plan the optimal speed trajectory curve; The actual vehicle testing process includes at least the following steps: Build a real vehicle test scenario, establish intersection spatial data, and deploy intelligent connected vehicle functional modules as well as roadside units and edge computing units; The On-Board Unit (OBU) establishes a V2X communication link with the Roadside Unit (RSU) to complete vehicle registration and continuously exchange data. During the journey, the vehicle collects its own motion status information and sends it to the RSU. The RSU forwards all vehicle motion status information to the Mobile Edge Computing (MEC). The MEC calculates the corresponding expected speed and right-of-way allocation information according to the expected speed and right-of-way information distribution strategy in the theoretical traffic algorithm. The edge computing unit (MEC) encapsulates the desired speed and right-of-way allocation information and forwards it to the corresponding vehicle through the roadside unit (RSU). After receiving its own desired speed and right-of-way information, the vehicle's onboard unit (OBU) generates the optimal speed trajectory curve according to the Deep Deterministic Strategy Gradient Algorithm (DDPG) in the theoretical communication algorithm and proceeds according to the optimal speed trajectory curve. After a vehicle passes through an intersection, the edge computing unit updates the right-of-way allocation status, clears the occupancy information of the vehicles that have passed, performs subsequent vehicle planning, and records the experimental data of each vehicle.

[0007] Furthermore, the vehicle priority allocation process in the upper-level right-of-way allocation adopts the First-Come First-Served (FCFS) principle, meaning that the earlier the arrival time, the higher the vehicle's priority. Vehicles are ranked according to their arrival time, with the arrival time at the intersection predicted based on the vehicle's real-time driving status on the road. :

[0008] in, For the feasible time period of the vehicle At the time of application, Indicates the vehicle's acceleration. Indicates the distance of the vehicle from the starting line of the intersection. This indicates the vehicle's current speed.

[0009] Furthermore, the calculation process for the feasible time period of vehicles during the upper-level right-of-way allocation is as follows: First, calculate the time period occupied by vehicle conflict points within the intersection. Establish a set of prohibited entry time slots for vehicles at the starting line of the intersection, where the time slot occupied by the vehicle conflict point is defined. The calculation method is as follows:

[0010] In the formula, These represent the four directions of entry at the intersection: north, south, east, and west. Lane numbering, The moment the vehicle first enters the point of conflict. For vehicle length, Indicates the speed of the vehicle within the intersection; Then, a projection algorithm is introduced to map the time periods occupied by conflict points of each vehicle within the intersection onto the time axis, forming a set of vehicle entry prohibition time slots. The projection algorithm is expressed as follows:

[0011] Among them, variables Representing the The distance from each conflict point to the starting line of the intersection , Total number of conflict points; superscript Indicates the direction of lane turning. To turn left, For straight ahead; Finally, based on the forbidden entry time slot set Determine the feasible time period for the vehicle. Among them, the reachability time interval of the vehicle is defined. It is represented as:

[0012] In the formula, Indicates V2X communication latency. This indicates the permissible range of fluctuation in vehicle acceleration. Indicates the permissible range of fluctuation in vehicle speed; According to time slot conflict constraints Determine the feasible time period for the vehicle. If the time interval for a vehicle to reach is occupied by the set of prohibited entry time slots, then the time interval for the vehicle to reach is determined by moving forward along the set of prohibited entry time slots.

[0013] Furthermore, the upper-level right-of-way allocation process also assigns a desired speed to each vehicle based on a determined feasible time period, which includes: First, determine the feasible time period for the vehicle. The midpoint is taken as the target entry time of the vehicle, that is:

[0014] In the formula, , Feasible time periods The upper and lower bounds; Then, calculate the vehicle's desired speed using the following formula:

[0015] In the formula, This represents the speed transition distance before a vehicle enters the intersection. It is used to constrain the vehicle's speed adjustment process before entering the intersection, ensuring that the vehicle reaches the preset speed upon entering the intersection. ; The average acceleration is calculated as follows:

[0016]

[0017] in, This represents the maximum acceleration of the vehicle. This represents the minimum acceleration of the vehicle.

[0018] Furthermore, the lower layer receives the desired speed assigned by the upper layer through the vehicle's OBU. After obtaining right-of-way information, speed planning is immediately performed. Given the initial state of the vehicle, the target state, and the trajectory of the vehicle in front, the optimal speed curve is generated.

[0019] Among them, vehicles The optimal velocity curve trajectory can be expressed as: ,vehicle The optimal speed curve trajectory of the target vehicle ahead is represented as: The trajectory generation function is denoted as Used to output the optimal velocity trajectory curve. Indicates vehicle The initial states are the initial time, initial velocity, and initial position, respectively; Indicates vehicle The target states are the target time, target speed, and target position; The lower-level vehicle control unit uses reinforcement learning to guide the vehicle agent to learn the optimal driving control strategy by designing a reward function; The vehicle status update results are fed back to the upper control center via V2X communication, thus forming a closed-loop control mechanism to realize dynamic coordinated passage of multiple vehicles within the intersection.

[0020] Furthermore, in the deep deterministic policy gradient algorithm for vehicle agents, the reward function comprehensively considers multiple dimensions such as collision risk, driving safety, motion stability, ride comfort, and task completion, collectively forming the system's total reward value to guide the agent's decision-making process. The total reward function... Defined as a linear combination of the following reward functions:

[0021] Collision Rewards This is used to penalize vehicles for causing collisions, in order to ensure the basic safety of the system; when a collision event is detected, a negative reward value will be given. Safety Rewards Used to guide vehicles to maintain a safe following distance from the vehicle in front; its calculation formula is as follows:

[0022] in, These are the weighting coefficients. This indicates the actual distance between the vehicle and the vehicle in front. This is a preset safety distance; Stability Rewards Aimed at smoothing vehicle motion, reducing drastic fluctuations in speed and acceleration, and improving ride comfort and energy efficiency, it is defined as:

[0023] in, and These are the weighting coefficients. and These represent the changes in velocity and acceleration between adjacent time steps, respectively. Comfort reward Used to optimize the riding experience, it reduces passenger discomfort by suppressing jerk fluctuations; it is defined as:

[0024] in, These are the weighting coefficients. Accelerate the vehicle; Task Constraints and Rewards Used to describe the completion status of an intelligent agent on a preset task, including achieving the expected speed and reaching the target time, it is defined as follows:

[0025] in, These are the weighting coefficients. At the current speed, For the desired speed; This is the actual arrival time. The time to reach the target.

[0026] Furthermore, during the actual vehicle testing process, the actual vehicle testing scenario can be carried out at any intersection in the city. A unified coordinate system is established with the center of the intersection to determine the location of each conflict point and its distance to each starting line, forming a spatial data benchmark for the intersection. The test vehicles were intelligent connected vehicles, deployed upstream of intersections at different entrances. Each vehicle was equipped with an On-Board Unit (OBU), GPS locator, IMU unit, vehicle speed sensor, and control interface module to achieve vehicle status acquisition, communication interaction, and control execution. Roadside units (RSUs) with independent power supply and communication capabilities are deployed near the intersection. The RSUs are connected to the edge computing units (MECs) via wired or wireless links. The edge computing units (MECs) are pre-loaded with intersection geometry information, lane parameters, and topology data required for projection algorithms.

[0027] Furthermore, during vehicle operation, the onboard unit periodically collects the vehicle's motion status information, including position, speed, and acceleration information, and generates standardized data packets. :

[0028] Vehicles transmit data packets via V2X communication The data packet is sent to the Roadside Unit (RSU); the RSU forwards the data packet to the Edge Computing Unit (MEC), which calculates the vehicle's right-of-way information and expected speed according to the theoretical traffic algorithm.

[0029] Furthermore, the edge computing unit (MEC) encapsulates the generated expected speed and right-of-way allocation information into data packets. :

[0030] Edge computing units (MECs) transmit data packets through roadside units (RSUs). Issued to the corresponding vehicles.

[0031] Furthermore, the vehicle received the data packet. Then, the system obtains the desired speed and right-of-way allocation information, calls the DDPG algorithm deployed on the vehicle, generates speed planning results based on the allocated passage time slots and desired speed, and outputs control commands to the underlying execution system accordingly, so that the vehicle enters and passes through the intersection at the desired speed. After the vehicle successfully passes through the intersection, the system records the experimental data, the MEC updates the right-of-way allocation status, clears the occupancy information of the vehicles that have passed, and provides resource scheduling support for subsequent vehicles to enter the intersection.

[0032] The beneficial effects of this invention are as follows: This invention proposes a two-layer control architecture that combines upper-layer right-of-way allocation with lower-layer speed planning. Through hierarchical division of labor and information interaction mechanisms, it achieves coordinated control from global planning to vehicle execution. The upper layer is responsible for global right-of-way allocation, while the lower layer completes speed planning and control execution based on instructions from the upper layer. This improves the system's coordination, real-time performance, and safety, providing a unified system framework for the engineering application of cooperative control at unsignalized intersections.

[0033] This invention proposes a complete real-vehicle testing scheme for the cooperative control algorithm at unsignalized intersections. By building a test environment that includes multiple intelligent connected vehicles, GPS, and V2X communication devices, the algorithm's traffic efficiency, safety, and robustness are verified in real-world road scenarios.

[0034] This invention proposes a real-vehicle testing method for cooperative control at unsignalized intersections in actual traffic environments. The method is validated on a test platform containing multiple intelligent connected vehicles, V2X communication and positioning equipment, and the performance of the algorithm in terms of real-time performance, reliability and adaptability is evaluated. This provides a feasible approach for iterative optimization and engineering deployment of the algorithm.

[0035] The proposed vehicle testing solution is simple, efficient, and supports rapid adaptation to different vehicle models and roadside equipment. It can adapt to real urban intersection road environments, significantly improving the engineering applicability and promotional value of this vehicle testing solution.

[0036] Overall, this invention can effectively improve the traffic efficiency of unsignalized intersections. It can flexibly conduct real-vehicle experiments in actual road intersection scenarios, which is convenient for verifying the real-time performance and reliability of the algorithm. It significantly improves the overall efficiency and safety level of cooperative passage at unsignalized intersections, and has important application and promotion value and industrialization prospects.

[0037] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0038] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein: Figure 1 This is a simplified schematic diagram of the two-layer control architecture according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating the actual vehicle testing process according to an embodiment of the present invention; Figure 3 This is a schematic diagram of a real-vehicle testing scenario according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the actual vehicle deployment in a real vehicle testing scenario according to an embodiment of the present invention; Figure 5 The speed curve of vehicle ROS during real-vehicle testing according to an embodiment of the present invention; Figure 6 The acceleration curve of the vehicle ROS during real-vehicle testing according to an embodiment of the present invention; Figure 7 The speed curve of the Apollo vehicle under this embodiment of the invention during real-vehicle testing; Figure 8 This is a curve showing the communication latency versus distance during a real-vehicle test in an embodiment of the present invention. Figure 9 This is a comparison curve of the three-dimensional trajectories of the vehicle Apollo and ROS inside the intersection in an embodiment of the present invention. Detailed Implementation

[0039] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0040] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0041] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0042] Please see Figures 1-9 This is a real-vehicle testing method for cooperative passage at unsignalized intersections based on a two-layer control architecture.

[0043] Example This embodiment first describes a real-vehicle testing method for cooperative passage at unsignalized intersections based on a two-layer control architecture, such as... Figure 1As shown, it includes at least a theoretical traffic flow algorithm design process and a real-vehicle testing process. In the theoretical traffic flow algorithm design process, a two-layer control framework is adopted: the upper layer is responsible for vehicle right-of-way allocation, and the lower layer is responsible for vehicle speed control. The upper-layer right-of-way allocation is based on the real-time network vehicle status data collected by the RSU, predicting the arrival time of each vehicle at the intersection, and determining the passage priority according to the arrival order. It calculates the time period occupied by conflict points within the intersection and calculates the prohibited entry time slot set at the intersection starting line based on a spatiotemporal projection algorithm. Based on this, it determines the feasible time period for each vehicle and assigns specific expected speed and right-of-way information to each vehicle. The lower-layer vehicle control receives speed and right-of-way information sent from the upper layer via V2X and uses the Deep Deterministic Strategy Gradient (DDPG) algorithm deployed on the vehicles to perform autonomous, safe, and efficient speed planning, enabling vehicles to arrive at the intersection within a specified time, thereby reducing vehicle fuel consumption and improving intersection traffic efficiency. In the real-vehicle testing process, communication equipment is deployed for the theoretical traffic flow algorithm, a real-vehicle testing environment is built, and the effectiveness, robustness, and engineering feasibility of the theoretical traffic flow algorithm are verified.

[0044] Currently, research on cooperative control at unsignalized intersections is mostly limited to simulation verification, lacking a systematic real-vehicle testing scheme. While existing methods perform well in simulation environments, they do not fully consider practical factors such as communication latency, sensor errors, and mixed traffic flow, making it difficult to directly apply the algorithms in real traffic environments. Furthermore, existing research largely relies on joint optimization models in fully connected scenarios, which suffer from high computational complexity and insufficient real-time performance, and has not yet formed a generalizable real-vehicle verification system.

[0045] To address the aforementioned issues, this invention proposes a real-vehicle testing method for cooperative passage at unsignalized intersections based on a two-layer control architecture. In terms of overall architecture, vehicle control is implemented at two levels: upper-layer right-of-way allocation and lower-layer speed planning, effectively improving intersection traffic efficiency. Furthermore, this invention also proposes a feasible real-vehicle solution for cooperative passage at intersections, whose effectiveness can be verified through real-vehicle experiments.

[0046] Specifically, the theoretically accepted algorithm design process includes the following steps: The upper-level vehicle network control center will distribute the allocated desired speed and right-of-way information to the vehicle's On-Board Unit (OBU); this specifically includes the following steps: To enable coordinated passage of vehicles within an intersection, the upper-level right-of-way allocation unit first predicts the arrival time of vehicles at the intersection based on their real-time driving status on the road. :

[0047] in, For the feasible time period of the vehicle At the time of application, Indicates the vehicle's acceleration. Indicates the distance of the vehicle from the starting line of the intersection. This indicates the vehicle's current speed.

[0048] Vehicle passage priority is determined according to the first-come, first-served (FCFS) principle.

[0049] The allocation of right-of-way information specifically includes the following follow-up steps: Calculate the time period occupied by vehicle conflict points within the intersection Establish a set of prohibited entry time slots for vehicles at the starting line of the intersection; Vehicle conflict point occupancy time period The calculation method is as follows:

[0050] in, These represent the four directions of entry at the intersection: north, south, east, and west. Lane number. The moment the vehicle first enters the point of conflict. For vehicle length, This represents the vehicle's speed within the intersection. For time mapping calculations, it is assumed that the vehicle travels at approximately a constant speed within the intersection. Driving.

[0051] To avoid conflicts in vehicle passage time slots at intersections, the upper-level right-of-way allocation unit introduces a projection algorithm during the right-of-way allocation process. This algorithm maps the time periods occupied by conflict points of each vehicle within the intersection onto the time axis, forming a set of time slots where vehicles are prohibited from entering. This will determine the feasible time period for vehicles. This algorithm can quickly determine the start and end times of available time slots for vehicles, thereby accurately determining the time periods for multiple vehicles to travel and ensuring that vehicle traffic within an intersection does not conflict.

[0052] The projection algorithm is expressed as follows:

[0053] Among them, variables Representing the Distance from each conflict point to the starting line of the intersection superscript Indicates the direction of lane turning ( To turn left, (For straight ahead).

[0054] This projection algorithm allows us to obtain the time period occupied by vehicle conflict points within the intersection, shifted along the time axis. The time period, i.e. the set of time slots during which vehicles are prohibited from entering at the starting line of the intersection.

[0055] Feasible time period for the determined vehicle ; To further determine the feasible time period for the vehicle This invention defines the vehicle reachable entry time interval. :

[0056] in, Indicates V2X communication latency. This indicates the permissible range of fluctuation in vehicle acceleration. This indicates the permissible range of fluctuation in vehicle speed.

[0057] To ensure that no time slot conflicts occur within the intersection, the feasible time period for vehicles should meet the following constraints:

[0058] This means that the time interval for a vehicle to enter the intersection must not overlap with the set of prohibited entry time slots at the intersection's starting line. The time interval that satisfies this constraint is the vehicle's feasible time interval. If the entire time interval for a vehicle to reach is occupied by the set of prohibited entry time slots, then the feasible time interval for the vehicle is determined by moving forward along the set of prohibited entry time slots.

[0059] After obtaining the feasible time period for vehicles through a projection algorithm, in order to improve the stability of traffic flow and the utilization rate of time slots, this invention selects the midpoint of the interval as the target entry time for vehicles, that is:

[0060] in, , They are respectively The upper and lower boundaries.

[0061] The desired speed of the vehicle can then be determined:

[0062] in, This represents the speed transition distance before a vehicle enters the intersection. It is used to constrain the vehicle's speed adjustment process before entering the intersection, ensuring that the vehicle reaches the preset speed upon entering the intersection. . For average acceleration, Calculate using the following formula:

[0063]

[0064] in, This represents the maximum acceleration of the vehicle. This represents the minimum acceleration of the vehicle.

[0065] Through the above process, the upper-level control unit can allocate a non-conflicting feasible time period and desired speed to each vehicle, and lock the time slot when the vehicle approaches the entrance, ensuring that the vehicle can reach the predetermined speed. Vehicles can smoothly enter the intersection. This ensures that vehicles can reach the intersection smoothly within the feasible time frame allocated by the higher-level decision-making, thus guaranteeing the safety and coordination of overall traffic.

[0066] At the lower level, vehicles use the deployed Deep Deterministic Strategy Gradient (DDPG) algorithm for speed planning based on received speed and right-of-way information and their own vehicle status. The lower layer receives the desired speed assigned by the upper layer through the vehicle's OBU. After obtaining right-of-way information, speed planning is immediately performed to ensure that the vehicle can reach the intersection within the feasible time period allocated by the upper layer. The goal of speed planning is to generate the optimal speed curve given the initial state, target state, and trajectory of the preceding vehicle.

[0067]

[0068] Among them, vehicles The optimal velocity curve trajectory can be expressed as: ,vehicle The optimal speed curve trajectory of the target vehicle ahead is represented as: The trajectory generation function is denoted as... Used to output the optimal velocity trajectory curve. Indicates vehicle The initial states are the initial time, initial velocity, and initial position, respectively; Indicates vehicle The target states are the target time, target speed, and target position.

[0069] The vehicle status update results are fed back to the upper control center via V2X communication, thus forming a closed-loop control mechanism to realize dynamic coordinated passage of multiple vehicles within the intersection.

[0070] To achieve the aforementioned trajectory planning, the lower-level vehicle control unit of this invention employs a reinforcement learning method based on a deep deterministic policy gradient algorithm. A well-designed reward function guides the vehicle agent to learn the optimal driving control strategy. The reward function comprehensively considers multiple dimensions, including collision risk, driving safety, motion stability, ride comfort, and task completion, to form the system's total reward value, thus guiding the agent's decision-making process.

[0071] Total reward function Defined as a linear combination of the following reward functions:

[0072] Collision Rewards This is used to penalize vehicle collisions to ensure the basic safety of the system; when a collision event is detected, a large negative reward value will be given, defined as follows:

[0073] Safety Rewards It is used to guide vehicles to maintain a safe following distance from the vehicle in front, preventing potential dangerous behaviors such as rear-end collisions; its calculation formula is as follows:

[0074] in, These are the weighting coefficients. This indicates the actual distance between the vehicle and the vehicle in front. This is the preset safety distance. Actual distance. Less than the safety distance When a vehicle is considered to be in potential danger, the reward value will generate a negative reward based on the distance; the larger the distance, the greater the penalty, thus incentivizing vehicles to maintain a sufficient safe distance.

[0075] Stability Rewards Aimed at smoothing vehicle motion, reducing drastic fluctuations in speed and acceleration, and improving ride comfort and energy efficiency, it is defined as:

[0076] in, and These are the weighting coefficients. and These represent the changes in velocity and acceleration between adjacent time steps, respectively. This reward incentivizes the vehicle to take smooth and stable control actions by penalizing the squared terms of the state changes.

[0077] Comfort reward Used to optimize the riding experience by reducing passenger discomfort by suppressing excessive acceleration. It is defined as:

[0078] in, These are the weighting coefficients. To accelerate the vehicle.

[0079] Task Constraints and Rewards Used to describe the completion status of an intelligent agent on a preset task, including achieving the expected speed and reaching the target time, it is defined as follows:

[0080] in, These are the weighting coefficients. At the current speed, For the desired speed; This is the actual arrival time. The target arrival time is the incentive. This incentive penalizes the deviation between the current state and the target state, driving the vehicle to efficiently complete the preset task objective.

[0081] During real vehicle testing, such as Figure 2 As shown, the specific steps include: S21. Build a real vehicle test scenario, establish intersection spatial data, deploy intelligent connected vehicle functional modules, and deploy roadside units and edge computing units; S22. Start the vehicle. The On-Board Unit (OBU) and the Roadside Unit (RSU) establish a V2X communication link to complete vehicle registration and continuously exchange data. S23. Drive the vehicle and collect its motion status information during driving, and send it to the roadside unit (RSU); S24. The Roadside Unit (RSU) forwards the motion status information of all vehicles to the Edge Computing Unit (MEC). The Edge Computing Unit calculates the corresponding expected speed and right-of-way allocation information according to the expected speed and right-of-way information distribution strategy in the theoretical traffic algorithm. S25. The edge computing unit (MEC) encapsulates the desired speed and right-of-way allocation information and forwards it to the corresponding vehicle through the roadside unit (RSU). S26. After receiving its desired speed and right-of-way information, the vehicle's on-board unit (OBU) generates the optimal speed trajectory curve according to the Deep Deterministic Strategy Gradient Algorithm (DDPG) in the theoretical communication algorithm, and proceeds according to the optimal speed trajectory curve. S27. After a vehicle passes through an intersection, the edge computing unit updates the right-of-way allocation status, clears the occupancy information of the vehicles that have passed, performs subsequent vehicle planning, and records the experimental data of each vehicle.

[0082] In step S21 of this embodiment, the experimental scenario is conducted at any intersection in the city. A unified coordinate system is established with the center of the intersection to determine the location of each conflict point and its distance to each starting line, thus forming a spatial data benchmark for the intersection.

[0083] The test vehicles were intelligent connected vehicles, deployed upstream of intersections at different entrances. Each vehicle was equipped with an onboard unit (OBU) (with V2X communication capabilities), GPS, IMU, vehicle speed sensors, and a control interface module to achieve vehicle status acquisition, communication interaction, and control execution.

[0084] Roadside units (RSUs) with independent power and communication capabilities are deployed near the intersection. The RSUs are connected to the edge computing units (MECs) via wired or wireless links. The MECs contain pre-set intersection geometry information, lane parameters, and topology data required for projection algorithms.

[0085] More specifically, in this embodiment, the experimental location was selected at an open T-shaped intersection, approximately a 7 m × 7 m square, with a lane width of 3.5 m. Two intelligent connected vehicles were used in the experiment: vehicle A (an Apollo vehicle) traveled straight from west to east, and vehicle B (a ROS car) turned left from south to north. Vehicle A was positioned at a distance from the starting line of the intersection. Vehicle B is a distance from the starting line of the intersection. Set speed transition distance The initial speed of vehicle A is... Assuming vehicle B travels at a constant speed in a straight line through the intersection, its initial speed is... Based on the intersection geometry, the distances from the conflict point to each entrance line are determined as follows: (Vehicle A) and (Vehicle B), the vehicle is within the intersection at... The vehicle travels at a constant speed. In this embodiment, the vehicle is at a low speed, and the effect of acceleration is not considered. .

[0086] The vehicles are equipped with the following modules: On-Board Unit (OBU), which is responsible for vehicle-side communication and computing; V2X communication module, which supports DSRC / 5G-V2X protocol; GPS and IMU module, which provides high-precision vehicle positioning and attitude information; power module, which supplies power to each unit; and low-level control system, which executes acceleration and deceleration control commands.

[0087] The RSU is deployed on the north side of the intersection, with a communication radius of approximately 100 m, and is connected to the MEC (Edge Computing Unit) via a wired connection. The MEC is pre-loaded with intersection topology information, lane parameters, and projection algorithm models. Figure 3 The diagram illustrates the process in this real-vehicle test scenario. Figure 4 A schematic diagram of the actual vehicle deployment in this real-vehicle test scenario is shown. The main parameters of the vehicle and environment are shown in Table 1.

[0088] Table 1

[0089] In step S22 of this embodiment, after the vehicle starts, the onboard sensors acquire the vehicle's motion status information; the onboard unit (OBU) completes initialization and establishes a V2X communication link with the roadside unit (RSU), realizing the connection with the edge computing unit (MEC). After successful vehicle registration, heartbeat communication is maintained to ensure continuous data interaction capability.

[0090] In step S23 of this embodiment, during vehicle operation, the on-board unit periodically collects the vehicle's motion state information, including position, speed, and acceleration, and generates standardized data packets. :

[0091] Vehicles transmit data packets via V2X communication Send to RSU.

[0092] In step S24 of this embodiment, after the RSU receives the status data of multiple vehicles via V2X communication, it directly forwards the data to the MEC. The MEC processes the vehicle status information inside and outside the intersection in a unified manner and establishes a unified coordinate system. Then, it executes the following steps according to the theoretical algorithm: Calculate the distance of the vehicle from the starting line of the intersection based on the direction the vehicle is coming from. Calculate the predicted arrival time of vehicles at the intersection. Vehicle priority is determined based on the vehicle's predicted arrival time. Vehicle right-of-way allocation is determined according to the FCFS (Forward, Forward, Forward) principle. Specifically, the right-of-way allocation process includes: calculating the time period occupied by vehicle conflict points within the intersection. Based on the projection algorithm, calculate the set of prohibited entry time slots for vehicles at the starting line of the intersection. ; Calculate the vehicle's reachable entry time range To ensure that vehicles do not experience time slot conflicts within the intersection, the feasible time period for vehicles should meet the following constraints; thus, the feasible time period for vehicles can be determined. ; Furthermore, the desired vehicle speed is determined according to the theoretical algorithm, which first calculates the target time for the vehicle to enter the intersection. Then, combine the target time to determine the desired speed. .

[0093] More specifically, in this embodiment, the data of vehicle A and vehicle B are processed uniformly by MEC to calculate the distance to the starting line of the intersection:

[0094] Assuming the vehicle application period is requested from the beginning, there is The estimated time for the vehicle to arrive at the starting line of the intersection is:

[0095] Vehicle A is expected to arrive at the intersection first. According to the FCFS principle, MEC assigns vehicle A a higher priority.

[0096]

[0097] For car A, the projection algorithm is used. Since the process has just started and there are no vehicles at the intersection, the time period occupied by the vehicle conflict point is considered. Empty, the set of prohibited entry time slots for vehicle A at the starting line of the intersection. for:

[0098] The calculated set of prohibited entry time slots is negative, so vehicle A continues to move at its original speed.

[0099] Based on the current speed of vehicle A Distance of vehicle A from the intersection The distance from the starting line of the intersection to the conflict point is 4.95m. Calculate the time period during which vehicle A occupies the conflict point (since this embodiment only uses two intelligent connected vehicles, and the Apollo vehicle moves at a constant speed in a straight line and has a higher priority, we directly predict the time interval during which vehicle A occupies the conflict point, instead of waiting until vehicle A moves to the conflict point to calculate it). :

[0100] For vehicle B, the distance from the intersection starting line to the conflict point is 1.77m. Using the projection algorithm, calculate the set of prohibited entry time slots for vehicle B at the intersection starting line. :

[0101] Based on the current speed of vehicle B Distance of vehicle from intersection Calculate the reachability window for vehicle B to reach the intersection. Assuming a 30ms communication delay in data transmission and a dynamic speed variation range within 0.1m / s, we have:

[0102]

[0103] According to the constraints:

[0104] The feasible time period for vehicle B is obtained. :

[0105] Because vehicle B has a lower priority, its feasible time period is selected later. The final feasible time period for vehicle B is:

[0106] Find the midpoint of the feasible time period:

[0107] Calculate the desired speed of vehicle B:

[0108] In step S25 of this embodiment, the MEC encapsulates the generated desired speed and right-of-way allocation information to form a data packet. :

[0109] MEC transmits data packets via RSU. The data is then sent to the corresponding vehicle. The RSU ensures low latency and high reliability during the data transmission process, and a retransmission mechanism can be implemented when necessary to ensure reliable data transmission.

[0110] In step S26 of this embodiment, the vehicle receives the data packet. Then, the desired speed and right-of-way allocation information are obtained, the DDPG algorithm deployed on the vehicle is invoked, and speed planning results are generated based on the allocated passage time slots and desired speed. Based on this, control commands are output to the underlying execution system so that vehicles enter and pass through the intersection at the desired speed, thereby avoiding conflicts and improving traffic efficiency.

[0111] In step S27 of this embodiment, after the vehicle successfully passes through the intersection, the system records the experimental data, the MEC updates the right-of-way allocation status, clears the occupancy information of the passed vehicles, and provides resource scheduling support for subsequent vehicles entering the intersection. The recorded experimental data includes the speed and acceleration curves of each vehicle, the vehicle trajectory inside the intersection, and statistics on the stability and latency of the traffic link. Figure 5 The speed curve of the vehicle during real-world testing using ROS is shown. The desired speed set in the example was approximately 0.7 m / s, while the measured average speed was approximately 0.6 m / s. This deviation is mainly due to communication latency and the fact that the actual vehicle speed fluctuation range is greater than the theoretical assumption, resulting in a wider calculated vehicle reachable time window. This shifts the allocated feasible time period later on the time axis, ultimately leading to a lower-than-expected speed. Furthermore, the actual speed is difficult to accurately track the reference value due to factors such as the discrete execution characteristics of the ROS vehicle's underlying control, motor overshoot, and changes in road friction. Nevertheless, the vehicle successfully completed intersection passage within the safe time window, verifying that the proposed method has good tolerance to system execution errors.

[0112] Figure 6 The acceleration curve of the vehicle's ROS during real-world testing is shown. The acceleration is at... The acceleration fluctuated continuously within the range of 0.7 to +0.7 m / s², with local spikes. Because the expected speed generated by the projection algorithm varies slightly in each control cycle, the vehicle needs to continuously accelerate and decelerate to track it, resulting in continuous acceleration jitter. Furthermore, the ROS algorithm uses encoder speed measurement and calculates acceleration differentially; quantization noise and the discrete execution characteristics of the motor further amplify this fluctuation. Although the acceleration variation exceeded the ideal model, it did not affect the vehicle's sequential passage through the intersection, indicating that the proposed method still has good stability under conditions of frequent speed adjustments and actuator nonlinearity.

[0113] Figure 7 The speed curve of the Apollo vehicle during real-world testing is shown. After a brief acceleration, the vehicle stabilized within the range of approximately 1.5–1.6 m / s, exhibiting slight periodic fluctuations. This result is consistent with the planning for vehicles with straight-ahead priority in the embodiment, i.e., vehicles pass through the intersection at a near-constant speed.

[0114] Figure 8 The curves showing the communication latency versus distance during real-vehicle testing are presented. The measured latency is approximately 80–120 ms, higher than the theoretically assumed 30 ms, mainly due to wireless link attenuation, retransmission, and device processing delay. Although the higher latency leads to deviations in arrival time calculation, it does not affect the time-window-based interval decision-making process. Experiments demonstrate that the system can still achieve safe passage, reflecting the robustness of the method to communication fluctuations.

[0115] Figure 9 The diagram shows a comparison of the 3D trajectories of vehicles Apollo and ROS within the intersection. The trajectory of vehicle Apollo is a stable straight line, while the trajectory of vehicle ROS is a left-turn trajectory. The two trajectories completely intersect in the time dimension, with no overlapping points of conflict.

[0116] The vehicle verification phase of this invention does not rely on complex hardware or high-precision control platforms, and the testing process has low engineering implementation difficulty. The core determinants of system performance lie in the computability, feasibility, and tolerance for dynamic errors of the upper-level projection algorithm. The two-layer control method proposed in this invention has a clear structure, low computational load, and can be decoupled from the vehicle's conventional speed controller, thus enabling rapid deployment on various vehicle platforms with different performance levels.

[0117] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A signal-free intersection coordinated passing real vehicle test method based on a double-layer control architecture, characterized in that: The method comprises at least a theoretical passing algorithm design process and a real vehicle test process, wherein in the theoretical passing algorithm design process, a double-layer control framework is adopted, the upper layer is responsible for vehicle right distribution, and the lower layer is responsible for vehicle speed control; in the real vehicle test process, a communication equipment is deployed for the theoretical passing algorithm, a real vehicle test environment is built, and the effectiveness, robustness and engineering feasibility of the theoretical passing algorithm are verified; The upper layer right distribution is based on the real-time collected state data of the networked vehicles by the RSU, predicts the time of each vehicle to reach the intersection, determines the passing priority according to the vehicle arrival order, calculates the conflict point occupation time period of the vehicles in the intersection, and calculates the prohibited entry time slot set at the starting line of the intersection based on the space-time projection algorithm; thus, the feasible time period of the vehicle is determined, and specific expected speed and right information is allocated to each vehicle; the lower layer vehicle control is based on the speed and right information sent by the upper layer through the vehicle networking communication V2X, and the deep deterministic policy gradient algorithm DDPG deployed on the vehicle is adopted to plan the optimal speed trajectory curve; In the real vehicle test process, the following steps are included at least: A real vehicle test scene is built, intersection space data is established, and intelligent networked vehicle function modules, roadside units and edge computing units are deployed; The OBU and the RSU establish a V2X communication link to complete vehicle registration and continuously interact with each other; the vehicle collects its own motion state information during driving and sends it to the RSU; the RSU forwards the motion state information of all vehicles to the MEC, which calculates the corresponding expected speed and right allocation information according to the expected speed and right information in the theoretical passing algorithm; The MEC encapsulates the expected speed and right allocation information and forwards it to the corresponding vehicle through the RSU; after receiving its own expected speed and right information, the vehicle generates an optimal speed trajectory curve according to the deep deterministic policy gradient algorithm DDPG in the theoretical communication algorithm and passes through the intersection according to the optimal speed trajectory curve; After the vehicle passes through the intersection, the MEC updates the right allocation state, clears the occupation information of the passed vehicles, plans for subsequent vehicles, and records the experimental data of each vehicle.

2. The method according to claim 1, wherein the method is characterized in that: The vehicle passing priority in the upper layer right-to-use allocation process adopts the principle of first-come-first-served, that is, the earlier the arrival time, the higher the vehicle passing priority, and the passing priority of the vehicle is sorted according to the arrival time, wherein the arrival time of the vehicle at the intersection is predicted based on the real-time driving state of the vehicle on the road : wherein, a time period in which the vehicle is drivable an application time point, an acceleration of the vehicle, a distance of the vehicle from a start line of the intersection, a current speed of the vehicle.

3. The method according to claim 2, wherein the method is characterized in that: The calculation process of the feasible time period of the vehicle in the upper layer right distribution process is as follows: Firstly, a vehicle conflict point occupation time period at the intersection is calculated A set of forbidden entry time slots of vehicles at the intersection starting line is established, wherein the vehicle conflict point occupation time period is calculated in the following manner: wherein, N, S, E and W represent the north, south, east and west approach directions of the intersection, respectively, L is the lane number, t is the time when the vehicle first enters the conflict point, L is the vehicle length, V is the speed of the vehicle within the intersection. Then, a projection algorithm is introduced to map the conflict point occupation time period of each vehicle in the intersection to the time axis to form a set of vehicle prohibited entry time slots wherein the projection algorithm is represented as follows: where the variable represents the distance from the conflict point to the beginning of the intersection, , is the total number of conflict points; the superscript indicates the turning direction of the lane, is left turn, is straight. Finally, according to the set of forbidden entry time slots determining a feasible time period for the vehicle wherein a reachable entry time interval for the vehicle is defined which is expressed as: In the formula, denotes a V2X communication latency, denotes a fluctuation range of vehicle acceleration allowed, denotes a fluctuation range of vehicle speed allowed; According to time slot conflict constraint Determining a feasible time period for a vehicle If all the reachable entry time intervals of the vehicle are forbidden to be occupied by the forbidden entry time slot set, the feasible time period for the vehicle is determined backward along the forbidden entry time slot set.

4. The method according to claim 3, wherein the method is characterized in that: The upper layer right distribution process also allocates expected speed to each vehicle according to the determined feasible time period of the vehicle, which includes: First, the midpoint of the feasible time period of the vehicle is taken as the target entry time of the vehicle, i.e.: wherein , are the upper and lower bounds of the feasible time period , respectively. Then, the expected speed of the vehicle is calculated as follows: In the formula, represents the speed transition distance before the vehicle enters the intersection, which is used to constrain the speed adjustment process of the vehicle before the entrance of the intersection, so that the vehicle can reach the preset speed when entering the intersection ; is the average acceleration, which is calculated in the following manner: wherein is a maximum value of the vehicle acceleration, is a minimum value of the vehicle acceleration.

5. The method according to claim 4, wherein the method is characterized in that: The lower layer receives the desired speed assigned by the upper layer through the vehicle OBU After receiving the road right information, the speed planning is performed immediately, and the optimal speed curve is generated under the conditions of the initial state, target state of the given vehicle and the trajectory of the preceding vehicle: wherein the optimal speed trajectory of the vehicle may be represented as wherein the optimal speed trajectory of the vehicle in front of the vehicle is represented as ; the trajectory generation function is denoted as for outputting the optimal speed trajectory curve, denotes the initial state of the vehicle , respectively, the initial time, the initial speed and the initial position; denotes the target state of the vehicle , respectively, the target time, the target speed and the target position; The lower layer vehicle control unit uses reinforcement learning method to guide the vehicle agent to learn the optimal driving control strategy by designing reward function; The vehicle state update result is fed back to the upper control center through V2X communication, thereby forming a closed-loop control mechanism and realizing dynamic cooperative passing of multiple vehicles in the intersection.

6. The method according to claim 5, wherein the method is characterized in that: In the deep deterministic policy gradient algorithm of the vehicle agent, a reward function is used to comprehensively consider collision risk, driving safety, motion stability, ride comfort and task completion degree, and the total reward value of the system is formed to guide the decision-making process of the agent, wherein the total reward function is defined as a linear combination of the following reward functions: collision reward to penalize the behavior of the vehicle in case of collision, in order to guarantee the basic safety of the system; a negative reward value will be given when a collision event is detected; Safety reward For guiding the vehicle to keep a safe following distance with the front vehicle; its calculation formula is as follows: wherein, is a weight coefficient; represents an actual distance from the ego vehicle to the preceding vehicle, is a preset safety distance; Stability reward The stability reward is defined as: wherein, and are weight coefficients, and respectively represent the velocity change amount and the acceleration change amount between adjacent time steps. Comfort rewards To optimize the ride experience, reduce passenger discomfort by suppressing jerk fluctuations, defined as: wherein, is a weight coefficient, is the vehicle jerk; Task constraint reward To describe the completion of the agent on the preset task, including the achievement of the expected speed and the arrival of the target time, which is defined as follows: wherein, is a weight coefficient, is a current speed, is a desired speed; is an actual arrival time, is a target arrival time.

7. The method according to claim 1, wherein the method is characterized in that: In the real test process, the test scene can be carried out at any intersection in the city, a unified coordinate system is established at the center of the intersection, the positions of the conflict points and their distances to the starting lines are measured, and the intersection space data benchmark is formed; The test vehicle is an intelligent networked vehicle, which is arranged at the upstream of intersections with different entrances; The vehicle is equipped with an on-board unit OBU, a GPS locator, an IMU unit, a vehicle speed sensor and a control interface module, which are used to realize vehicle state acquisition, communication interaction and control execution; A roadside unit RSU with independent power supply and communication capability is deployed near the intersection, and the RSU is connected with an edge computing unit MEC through a wired or wireless link; the edge computing unit MEC is preloaded with topological data of intersection geometric information, lane parameters and projection algorithm requirements.

8. The method according to claim 7, wherein the method is a real vehicle test method for coordinated passing at a signal-free intersection based on a double-layer control architecture. During the vehicle driving, the on-board unit periodically collects the motion state information of the vehicle, including position, speed and acceleration information, and generates a standardized data packet : Vehicles transmit data packets via V2X communication Send to the roadside unit (RSU); The roadside unit RSU forwards the data packet to the edge computing unit MEC, and the edge computing unit MEC calculates the right-of-way information and the expected speed of the vehicle according to the theoretical traffic algorithm.

9. The method according to claim 8, wherein the method is characterized in that: The edge computing unit MEC data encapsulates the generated expected speed and right-of-way allocation information to form a data packet : The edge computing unit MEC transmits data packets to the corresponding vehicle via the road side unit RSU and issues them to the corresponding vehicle.

10. The method according to claim 9, wherein the method is characterized in that: The vehicle receives the data packet After that, the expected speed and the right-of-way allocation information are obtained, the DDPG algorithm deployed at the vehicle end is called, the speed planning result is generated based on the allocated passing time slot and the expected speed, and control instructions are output to the underlying execution system according to the speed planning result, so that the vehicle enters and passes through the intersection at the expected speed; after the vehicle passes through the intersection smoothly, the system records experimental data, the MEC updates the right-of-way allocation state, and the occupation information of the vehicle that has passed through is cleared to provide resource scheduling support for subsequent vehicles entering the intersection.

Images