Multi-vehicle cooperative driving control method and system for unmanned vehicles

Abstract

The invention discloses a multi-vehicle cooperative driving control method and system for unmanned vehicles, and relates to the technical field of intelligent driving control, and the method comprises the steps: carrying out the multi-dimensional sensing of a plurality of unmanned vehicles, and obtaining a multi-dimensional sensing data set; environment modeling is carried out, and a dynamic traffic scene model is constructed; multi-vehicle path conflict analysis is carried out through the multi-vehicle cooperation platform, and a first path combination is generated; optimizing the first path combination according to the plurality of single-trajectory tracking results to generate a second path combination; and performing multi-vehicle intelligent cooperative driving control on the unmanned vehicle based on the multi-vehicle cooperative strategy. The technical problems that in the prior art, the environment sensing range of a single vehicle is limited, communication between vehicles is insufficient, complex and changeable traffic scenes are difficult to deal with, and consequently the multi-vehicle cooperative driving efficiency and safety are poor are solved, intelligent control over the unmanned vehicle in the dynamic traffic scene is achieved, and the driving safety is improved. The technical effect of improving the multi-vehicle cooperative driving efficiency and the driving safety is achieved.

Description

Technical Field

[0001] This application relates to the technical field of intelligent driving control, and particularly to a multi-vehicle collaborative driving control method and system for driverless vehicles. Background Art

[0002] The environmental information obtained by traditional single-vehicle intelligent driving through its own sensors (such as cameras, radars, and lidar) has certain range limitations, and it is difficult to comprehensively grasp all key information in the dynamic traffic scene. Moreover, in the dynamic traffic scene, multiple driverless vehicles need to cooperate to avoid path conflicts. Existing methods are mostly limited to short-distance point-to-point communication, and it is difficult to effectively integrate the state information of multiple vehicles for comprehensive decision-making. In addition, they perform poorly in terms of vehicle path planning and adjustment in the dynamic traffic environment, thus affecting the efficiency and safety of multi-vehicle collaborative control of driverless vehicles.

[0003] Therefore, in the current related technologies, there are technical problems such as limited single-vehicle environmental perception range, insufficient vehicle-to-vehicle communication, and difficulty in coping with complex and changeable traffic scenes, resulting in poor multi-vehicle collaborative driving efficiency and safety. Summary of the Invention

[0004] This application provides a multi-vehicle collaborative driving control method and system for driverless vehicles, which solves the technical problems in the prior art, such as limited single-vehicle environmental perception range, insufficient vehicle-to-vehicle communication, and difficulty in coping with complex and changeable traffic scenes, resulting in poor multi-vehicle collaborative driving efficiency and safety. It realizes the intelligent control of driverless vehicles in the dynamic traffic scene and achieves the technical effect of improving multi-vehicle collaborative driving efficiency and driving safety.

[0005] This application provides a multi-vehicle collaborative driving control method for driverless vehicles. The method includes: performing multi-dimensional sensing on multiple driverless vehicles through multi-sensors deployed on the vehicles to obtain a multi-dimensional sensing data set, where the multi-dimensional sensing data set includes multiple vehicle state data sets and multiple vehicle environment data sets; synchronizing the multiple vehicle state data sets to the vehicle networking to establish a multi-vehicle communication network, and based on the multiple vehicle environment data sets, performing environment modeling to construct a dynamic traffic scene model; integrating the dynamic traffic scene model with the multi-vehicle communication network to obtain a multi-vehicle collaborative platform, and through the multi-vehicle collaborative platform, performing multi-vehicle path conflict analysis on multiple driverless vehicles to generate a first path combination; traversing the first path combination to perform single-vehicle trajectory tracking to obtain multiple single-trajectory tracking results, and optimizing the first path combination according to the multiple single-trajectory tracking results to generate a second path combination; performing collaborative learning on multiple driverless vehicles according to the second path combination, formulating a multi-vehicle collaborative strategy based on the learning feedback results, and performing multi-vehicle intelligent collaborative driving control on the driverless vehicles based on the multi-vehicle collaborative strategy.

[0006] In a possible implementation, the multi-vehicle collaborative driving control method for the driverless vehicle further performs the following processes: performing synchronization processing based on the multiple vehicle state data sets to generate a multi-vehicle state alignment data set; taking the multiple driverless vehicles as multiple communication nodes, traversing the multiple communication nodes for communication association analysis to determine multiple correlation coefficients; connecting the multi-vehicle state alignment data set according to the multiple communication nodes and the multiple correlation coefficients to construct a multi-vehicle communication network; performing Gaussian process regression based on the multiple vehicle environment data sets to construct an environmental space distribution model; using a graph neural network to traverse the multiple vehicle environment data sets to establish the multi-vehicle communication network for interaction to obtain dynamic traffic flow parameters; mapping the dynamic traffic flow parameters to the environmental space distribution model to obtain the dynamic traffic scene model.

[0007] In a possible implementation, the multi-vehicle collaborative driving control method for the driverless vehicle further performs the following processes: taking the multiple driverless vehicles as multiple game participants for driving analysis to determine multiple driving paths, where the multiple driving paths have a corresponding relationship with the multiple game participants; extracting multiple starting positions and multiple ending positions, and merging the multiple driving paths according to the multiple starting positions and the multiple ending positions to construct a path strategy space for multiple vehicles; mapping the multiple game participants to the path strategy space, traversing the multiple starting positions and the multiple ending positions for intersection detection, and performing path conflict games according to the path intersection information to generate the first path combination.

[0008] In a possible implementation, the multi-vehicle collaborative driving control method for the driverless vehicle further performs the following processes: taking the multiple game participants as indexes, traversing the path strategy space for retrieval to determine multiple target driving paths, where the multiple target driving paths have a corresponding relationship with the multiple game participants; performing spatio-temporal analysis based on the multiple target driving paths to construct a time-space graph; performing intersection determination according to the multiple target driving paths to generate a determination result; if the determination result has an intersection, then performing overlap analysis according to the multiple starting positions and the multiple ending positions according to the time-space graph to generate an overlap signal; performing matching identification on the multiple target driving paths according to the overlap signal to determine multiple path conflict points, performing driving impact analysis on the multiple path conflict points, sorting the multiple path conflict points in descending order of priority according to the impact analysis result to determine a path conflict sequence; traversing the path strategy space according to the path conflict sequence to perform games on the multiple target driving paths to obtain a game path solution, and when the game path solution reaches a Nash equilibrium, then generating the first path combination.

[0009] In a possible implementation, the multi-vehicle cooperative driving control method for the driverless vehicle further performs the following processes: Simulate the driving of multiple driverless vehicles according to the first path combination to obtain a simulation driving result; Perform single-vehicle trajectory tracking based on the simulation driving result to generate M pieces of vehicle simulated driving trajectory data; Set expected trajectory data, and determine whether the M pieces of vehicle simulated driving trajectory data conform to the expected trajectory data. If the M pieces of vehicle simulated driving trajectory data do not conform to the expected trajectory data, generate a deviation prompt, perform a deviation calculation on the M pieces of vehicle simulated driving trajectory data according to the deviation prompt to generate multiple trajectory deviation values; Perform an overlap analysis on the M pieces of vehicle simulated driving trajectory data according to the time-space diagram based on the multiple trajectory deviation values, set a deviation critical value, extract the trajectory deviation values greater than or equal to the deviation critical value, and match and determine N pieces of vehicle simulated driving trajectory data, where N is a positive integer greater than or equal to 0 and less than or equal to M; Add the N pieces of vehicle simulated driving trajectory data to the multiple single-trajectory tracking results.

[0010] In a possible implementation, the multi-vehicle cooperative driving control method for the driverless vehicle further performs the following processes: Perform local control on multiple driverless vehicles based on the N pieces of vehicle simulated driving trajectory data to generate abnormal control information; Update the N pieces of vehicle simulated driving trajectory data according to the abnormal control information to generate N pieces of trajectory update data; Perform a fitness evaluation on the N pieces of trajectory update data according to the first path combination to obtain multiple fitness values; Replace and optimize the first path combination with the N pieces of trajectory update data according to the multiple fitness values to generate the second path combination.

[0011] In a possible implementation, the multi-vehicle cooperative driving control method for the driverless vehicle further performs the following processes: Train the states of multiple driverless vehicles according to the second path combination through reinforcement learning to construct a state information space; Train the actions of multiple driverless vehicles according to the second path combination through reinforcement learning to construct an action feedback space; Introduce a reward function to perform a fusion analysis on the state information space and the action feedback space to determine the global cooperative learning results of multiple driverless vehicles; Perform multi-agent reinforcement learning according to the global cooperative learning results to generate the learning feedback results, where the learning feedback results include positive feedback learning results and negative feedback learning results; Implement behavior rewards based on the positive feedback learning results and implement punishments based on the negative feedback learning results to generate a multi-vehicle cooperation strategy, and send the multi-vehicle cooperation strategy to multiple driverless vehicles for multi-vehicle intelligent cooperative driving control.

[0012] The present application also provides a multi-vehicle cooperative driving control system for driverless vehicles, including: a multi-dimensional sensing data set acquisition module, configured to perform multi-dimensional sensing on multiple driverless vehicles through multi-sensors deployed on the vehicles to obtain a multi-dimensional sensing data set, where the multi-dimensional sensing data set includes multiple vehicle state data sets and multiple vehicle environment data sets; a dynamic traffic scene model construction module, configured to synchronize the multiple vehicle state data sets to a vehicle network to establish a multi-vehicle communication network, perform environment modeling based on the multiple vehicle environment data sets, and construct a dynamic traffic scene model; a path conflict analysis module, configured to integrate the dynamic traffic scene model with the multi-vehicle communication network to obtain a multi-vehicle cooperation platform, perform multi-vehicle path conflict analysis on multiple driverless vehicles through the multi-vehicle cooperation platform, and generate a first path combination; a path combination optimization module, configured to traverse the first path combination to perform single-vehicle trajectory tracking, obtain multiple single-trajectory tracking results, optimize the first path combination according to the multiple single-trajectory tracking results, and generate a second path combination; a multi-vehicle cooperation strategy formulation module, configured to perform cooperative learning on multiple driverless vehicles according to the second path combination, formulate a multi-vehicle cooperation strategy based on the learning feedback results, and perform multi-vehicle intelligent cooperative driving control on the driverless vehicles based on the multi-vehicle cooperation strategy.

[0013] It is intended to perform multi-dimensional sensing on multiple driverless vehicles through the multi-vehicle cooperative driving control method and system for driverless vehicles proposed in the present application to obtain a multi-dimensional sensing data set; perform environment modeling to construct a dynamic traffic scene model; perform multi-vehicle path conflict analysis through a multi-vehicle cooperation platform to generate a first path combination; optimize the first path combination according to multiple single-trajectory tracking results to generate a second path combination; and perform multi-vehicle intelligent cooperative driving control on the driverless vehicles based on a multi-vehicle cooperation strategy. This solves the technical problems in the prior art, such as the limited single-vehicle environment perception range, insufficient vehicle-to-vehicle communication, and difficulty in coping with complex and changeable traffic scenarios, resulting in poor multi-vehicle cooperative driving efficiency and safety. It realizes the intelligent control of driverless vehicles in a dynamic traffic scene and achieves the technical effect of improving multi-vehicle cooperative driving efficiency and driving safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings of the embodiments of the present disclosure will be briefly introduced below. Flowcharts are used in this application to illustrate the operations performed by the systems according to the embodiments of the present application. It should be understood that the operations described above or below do not necessarily need to be executed precisely in order. On the contrary, according to the need, they can be executed in reverse order or simultaneously. At the same time, other operations can also be added to these processes, or one or several operations can be removed from these processes.

[0015] Figure 1Schematic flowchart of the multi-vehicle collaborative driving control method for the driverless vehicle provided by the embodiment of the present application.

[0016] Figure 2 Schematic structural diagram of the multi-vehicle collaborative driving control system for the driverless vehicle provided by the embodiment of the present application.

[0017] Explanation of reference numerals: multi-dimensional sensing data set acquisition module 10, dynamic traffic scene model construction module 20, path conflict analysis module 30, path combination optimization module 40, multi-vehicle collaborative strategy formulation module 50. Detailed implementation manners

[0018] The above description is only an overview of the technical solution of the present application. In order to be able to understand the technical means of the present application more clearly, it can be implemented according to the content of the description. And in order to make the above and other purposes, features and advantages of the present application more obvious and understandable, the following specifically gives the detailed implementation manners of the present application.

[0019] In order to make the purpose, technical solution and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations of the present application. All other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0020] In the following description, "some embodiments" are involved, which describe a subset of all possible embodiments. However, it can be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict. The terms "first\second" involved are only used to distinguish similar objects and do not represent a specific order for the objects. The terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or server that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or are inherent to these processes, methods, products or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application.

[0021] The embodiment of the present application provides a multi-vehicle collaborative driving control method for a driverless vehicle, as Figure 1 shown, the method includes: Step S100, performing multi-dimensional sensing on multiple driverless vehicles through multi-sensors deployed on the vehicles to obtain a multi-dimensional sensing data set, where the multi-dimensional sensing data set includes multiple vehicle state data sets and multiple vehicle environment data sets.

[0022] Preferably, multiple different types of sensors are installed on the driverless vehicle, including but not limited to radar (monitoring speed and position at a long distance), lidar (high-precision ranging and 3D modeling, LiDAR), camera (collecting visual information for object recognition and lane detection), ultrasonic sensor (detecting obstacles at a short distance), GPS / IMU (providing vehicle position and motion state information), etc. Multidimensional data is collected through the collaboration of different sensors to comprehensively perceive the environment where the vehicle is located and its own state, obtaining a multi-dimensional sensing data set, including multiple vehicle state data sets and multiple vehicle environment data sets. Among them, the vehicle state data set refers to the state information of the driverless vehicle itself, usually including the current position of the vehicle (GPS position, on-vehicle positioning system data), the speed and acceleration of the vehicle, the heading angle and driving direction of the vehicle, the steering angle, wheel speed, operating state of the power system, etc., as well as the braking state and throttle control of the vehicle; the vehicle environment data set refers to the information of the vehicle's surrounding environment, including interaction data with the external environment, usually obtained through on-vehicle sensors (such as radar, lidar, camera, ultrasonic sensor, etc.), including road conditions (whether there are obstacles, lane lines, traffic signs, etc.), detection of other vehicles, pedestrians, and obstacles around, environmental changes (such as weather conditions, road conditions, traffic lights, etc.), and the state and trend of the surrounding traffic flow (speed and driving direction of other vehicles, etc.). By comprehensively perceiving the vehicle state and environment, it helps to understand the dynamic traffic scene in real time, provides support for the vehicle's path planning and decision-making, and realizes the intelligent collaboration and safe and efficient driving of the driverless vehicle.

[0023] Step S200, synchronize the multiple vehicle state data sets to the vehicle network to establish a multi-vehicle communication network, and perform environmental modeling based on the multiple vehicle environment data sets to construct a dynamic traffic scene model.

[0024] Preferably, multiple vehicle state datasets are synchronized to the vehicle Internet of Things to establish a multi-vehicle communication network, ensuring that each vehicle can understand the real-time status of other vehicles, that is, sharing its own vehicle state with other vehicles through the communication network. For example, data such as vehicle speed and position obtained by Vehicle A will be transmitted to the vehicle Internet of Things in real time for reference and collaborative planning by other vehicles such as Vehicle B and Vehicle C. Among them, the vehicle Internet of Things (V2X) refers to a system in which vehicles exchange data with other vehicles, infrastructure (such as traffic lights, road signs), pedestrians, and the cloud through wireless communication technologies (such as 5G, LTE, Wi-Fi, etc.), enabling real-time transmission and sharing of information and supporting the cooperative driving system. Multiple driverless vehicles establish a multi-vehicle communication network through the vehicle Internet of Things, enabling each vehicle to exchange dynamic state information, and then achieving more accurate cooperative decision-making and path planning; then, an environmental model is built through the collected multiple vehicle environment datasets. Specifically, object detection based on sensor data (detecting surrounding vehicles, pedestrians, obstacles, etc.) and path planning models (such as lane recognition, obstacle avoidance, etc.) are used to extract information such as objects, traffic signs, and traffic lights through computer vision, and a virtual environment model reflecting the real world is constructed, including a static environment (such as road structure, traffic signs, terrain, etc.) and a dynamic environment (such as the dynamic changes of pedestrians, other vehicles, obstacles, etc.), that is, a dynamic traffic scene model is constructed, that is, a dynamic change scene model constructed using the state information of the vehicle (such as vehicle position, speed, etc.) and environmental data (such as traffic flow, obstacle position, etc.), which not only includes static road information but also reflects the dynamic changes in the traffic environment in real time. For example, traffic density (changes in traffic flow), traffic signals (changes in traffic lights, markings), and emergencies (reactions and avoidance during emergency stops, accidents). By constructing a dynamic traffic scene model, driverless vehicles can more accurately predict environmental changes, achieve cooperative control between vehicles, ensure that driverless vehicles can understand the changes in the surrounding traffic environment in real time and accurately, and make intelligent path planning and decisions, enabling multiple driverless vehicles to effectively cooperate and interoperate in a complex traffic environment.

[0025] Preferably, step S200 further includes step S210 of performing synchronization processing on the multiple vehicle state data sets to generate a multi-vehicle state alignment data set; step S220 of using multiple driverless vehicles as multiple communication nodes, traversing the multiple communication nodes for communication correlation analysis to determine multiple correlation coefficients; step S230 of connecting the multi-vehicle state alignment data set according to the multiple communication nodes and multiple correlation coefficients to construct a multi-vehicle communication network; step S240 of performing Gaussian process regression on the multiple vehicle environment data sets to construct an environmental space distribution model; step S250 of using a graph neural network to traverse the multiple vehicle environment data sets to establish the multi-vehicle communication network for interaction to obtain dynamic traffic flow parameters; and step S260 of mapping the dynamic traffic flow parameters to the environmental space distribution model to obtain the dynamic traffic scenario model.

[0026] Preferably, performing synchronization processing on the multiple vehicle state data sets includes timestamp alignment (aligning vehicle state data at different time points to the same time reference point) and data interpolation (filling in missing state information through linear interpolation, spline interpolation, etc.) to generate a multi-vehicle state alignment data set. Regarding multiple driverless vehicles as a communication node in a vehicle-to-everything (V2X) network, they can exchange information with other vehicles or infrastructure nodes (such as roadside units). Then, by analyzing the communication situation between different vehicles (such as information exchange frequency, communication quality, relative position, etc.), the communication correlation between each vehicle is determined, that is, a parameter measuring the communication quality between two vehicles, which may include signal strength, distance, signal transmission delay, network latency, etc., to obtain multiple correlation coefficients. Among them, the correlation coefficient represents the communication strength or reliability between two nodes. Then, the multi-vehicle state alignment data set is connected according to the multiple communication nodes and multiple correlation coefficients, that is, based on the communication relationship and relative position between vehicles, the state alignment data between vehicles is organized into a communication network, that is, a multi-vehicle communication network. Each node in the network represents a vehicle, and the connection weight between nodes is determined by the correlation coefficient obtained from the communication correlation analysis. Among them, if the communication signal between two vehicles is strong or the correlation is high, their connection will have a larger weight; otherwise, the connection weight will be smaller.

[0027] Preferably, Gaussian process regression is performed based on multiple vehicle environment datasets, that is, a model is constructed according to the environmental data collected by the vehicle (such as traffic flow, road quality, obstacle distribution, etc.) to describe the distribution of these environmental factors in space, and then an environmental space distribution model is obtained, which reflects the environmental characteristics of different regions, such as whether a certain section of the road is congested or there are obstacles. Then, the graph neural network traverses all the vehicles and their environmental data in the entire communication network for interaction and information dissemination. Specifically, through the learning of the graph neural network, the system can obtain dynamic parameters such as traffic flow, vehicle speed distribution, and traffic density, that is, dynamic traffic flow parameters. The graph neural network can combine the vehicle states and environmental data in the communication network to obtain the dynamic changes of traffic flow in real time. Among them, the graph neural network is a type of deep learning model for graph data structure learning, which is particularly suitable for tasks with complex relationships between nodes. Finally, the dynamic traffic flow parameters (such as traffic density, traffic congestion degree, etc.) obtained through the graph neural network are mapped into the environmental space distribution model constructed by Gaussian process regression, that is, the vehicle and environmental data are combined to form a dynamic traffic scenario model, which not only considers static road information (such as the number of lanes, road signs, etc.), but also dynamically integrates the changes in traffic flow and real-time information of traffic events (such as accidents, congestion), improving the safety, efficiency, and flexibility of fleet operation.

[0028] Step S300, integrate the dynamic traffic scenario model with the multi-vehicle communication network to obtain a multi-vehicle collaboration platform, and perform multi-vehicle path conflict analysis on multiple driverless vehicles through the multi-vehicle collaboration platform to generate a first path combination.

[0029] Preferably, integrate the dynamic traffic scenario model (the global view of the entire traffic scenario) with the multi-vehicle communication network (real-time data synchronization and information interaction), that is, input the data of the multi-vehicle communication network into the dynamic traffic scenario model to obtain a multi-vehicle cooperation platform, which can plan the driving path of each vehicle according to real-time traffic information and vehicle positions, and can identify whether there are possible conflicts in the planned paths of multiple vehicles (such as multiple vehicles driving into the same intersection, collision during overtaking, etc.), analyze possible risk points and propose solutions, and monitor and update the paths in real time to adapt to sudden changes (such as traffic accidents or road closures). Then, perform multi-vehicle path conflict analysis on multiple driverless vehicles through the multi-vehicle cooperation platform. Specifically, when the planned paths of multiple vehicles overlap in space and time, which may lead to collisions or blockages, path conflict analysis is required. The current position, speed, acceleration, driving direction of each vehicle, and road environment information (such as the number of lanes, road conditions) are used as input data and input into the dynamic traffic scenario model to simulate the driving trajectories of each vehicle and detect whether there are path overlaps or time conflicts, and at the same time propose optimized paths to avoid conflicts, such as adjusting the speed of some vehicles, changing lanes, and changing the right of way; after conflict analysis, preliminarily plan paths for all driverless vehicles to obtain a first path combination. Among them, the path of each vehicle is globally optimized to avoid conflicts with the paths of other vehicles, and at the same time meet traffic rules and goals (such as the shortest path, the highest efficiency), thereby providing an optimized driving plan for the vehicles, which helps to improve the safety, efficiency, and coordination of the driverless vehicle fleet.

[0030] Further, step S300 further includes step S310, performing driving analysis on multiple driverless vehicles as multiple game participants to determine multiple driving paths, where there is a corresponding relationship between the multiple driving paths and the multiple game participants; step S320, extracting multiple starting positions and multiple ending positions, and merging the multiple driving paths according to the multiple starting positions and the multiple ending positions to construct a path strategy space for multiple vehicles; step S330, mapping the multiple game participants to the path strategy space, traversing the multiple starting positions and the multiple ending positions for intersection detection, and performing path conflict games according to the path intersection information to generate the first path combination.

[0031] Preferably, the path planning and conflict resolution of driverless vehicles are analyzed and optimized through game theory. Specifically, each driverless vehicle is regarded as an independent participant in the game. The participants in game theory need to make decisions on path selection based on their goals (such as the shortest path, the lowest energy consumption, etc.) and their interaction relationships with other vehicles. The driving analysis of each vehicle includes the current starting position and the target ending position, environmental constraints (such as road conditions, traffic rules, obstacles, etc.), the behaviors of other vehicles and their possible impacts on its own decisions, and then generate one or more feasible paths for each vehicle, that is, determine multiple driving paths, considering the specific driving goals and constraint conditions of the vehicle. Each path corresponds to a game participant. Among them, the starting position of each vehicle is the position where the current vehicle is located, and the ending position is the target position of its driving task (such as the destination or the traffic task target). Different vehicles may have different starting and ending positions. Classify and merge the paths of all vehicles according to their starting and ending positions to form a comprehensive multi-vehicle path strategy space, which not only includes the feasible paths of each vehicle but also reflects the possible overlapping or conflict relationships between different paths. Each element in the strategy space can be represented as the decision of a certain vehicle on a certain path.

[0032] Preferably, map the game participants to the path strategy space, that is, each vehicle (game participant) corresponds to one or more path choices, and these choices are concretized as path points in the strategy space. Then, detect whether there are spatial and temporal intersections between multiple vehicle paths, such as whether two vehicles will pass through the same road section and the same time period. If an intersection exists, there may be a collision or blockage. Record the path overlap information between each pair of vehicles, including the specific position and time of the overlap. Finally, conduct a path conflict game based on the path intersection information. Specifically, each vehicle, as a game participant, optimizes its path selection through a game theory model to avoid conflicts with other vehicles. Each vehicle has multiple path strategies to choose from (strategy set) in the game. The goal of conflict resolution is to find a path combination that minimizes the total cost (such as time, energy consumption, or safety risk) of all vehicles. When a path conflict is detected, the vehicle can avoid the conflict by adjusting the speed, changing the driving order, or re-planning the path. Finally, obtain the preliminary optimal path selection for each vehicle, generate the first path combination, ensure that there is no path conflict (or the conflict is minimized) and the overall driving efficiency is maximized, so as to ensure the safe and efficient cooperative driving of driverless vehicles.

[0033] Further, step S330 further includes step S331, using the multiple game participants as indexes to traverse the path strategy space for retrieval, determining multiple target driving paths, where there is a corresponding relationship between the multiple target driving paths and the multiple game participants; step S332, performing spatio-temporal analysis based on the multiple target driving paths to construct a time-space graph; step S333, performing an intersection determination according to the multiple target driving paths to generate a determination result; step S334, if there is an intersection in the determination result, then perform an overlap analysis on the multiple starting positions and the multiple ending positions according to the time-space graph to generate an overlap signal; step S335, performing a matching identification on the multiple target driving paths according to the overlap signal to determine multiple path conflict points, performing a driving impact analysis on the multiple path conflict points, and sorting the multiple path conflict points in descending order of priority according to the impact analysis result to determine a path conflict sequence; step S336, traversing the path strategy space according to the path conflict sequence to play a game on the multiple target driving paths to obtain a game path solution, and when the game path solution reaches a Nash equilibrium, then generate the first path combination.

[0034] Preferably, each driverless vehicle is regarded as a game participant (index), corresponding to its specific path selection in the path strategy space. By traversing the path strategy space, the target driving path of each vehicle is retrieved and determined. These paths meet specific conditions (such as the lowest cost, the highest safety, etc.) and have a one-to-one correspondence with the game participants. Then, spatio-temporal analysis is performed on the target paths of each vehicle, combining information such as the driving position, speed, and acceleration of the vehicle at different time periods to evaluate the distribution of the path in the spatio-temporal dimension, and a time-space graph is constructed to show the driving trajectory of each vehicle in a specific spatio-temporal area. Among them, the time axis represents the time distribution of the vehicle on the path, and the space axis represents the distribution of the path in space (such as road coordinates, lane positions, etc.); then, an intersection determination is performed on the multiple target driving paths, that is, to detect the overlapping situation of the target driving paths of different vehicles in the time-space graph, that is, whether there are two or more vehicles at the same moment and the same location, and a determination result is generated, including the existence of an intersection or the non-existence of an intersection.

[0035] Preferably, if there is an intersection in the determination results, overlapping analysis is performed, that is, for the overlapping regions existing in the path intersection, according to the starting position, ending position, and the information of the time-space graph, the nature and severity of the overlap are analyzed. For example, the time range of the overlap (such as short or continuous), the spatial range of the overlap, such as whether it is on the same road segment, and then an overlap signal (an identifier for the path conflict area) is generated, including information such as the starting point, ending point of the conflict, and the severity of the conflict. Among them, there is only a conflict when both the time and space in the path overlap. If either the space or time does not overlap, the path is not in conflict; then, the specific positions of the path conflicts are identified by matching the overlap signal, that is, which space-time points have conflicts, multiple path conflict points are determined, and the driving impact analysis is performed on them to evaluate the impact degree of these conflict points on vehicle driving. For example, high-impact conflict points may lead to serious accidents or long delays, and low-impact conflict points only have a slight impact on the local path. Then, the conflict points are sorted in descending order according to the impact degree to generate a path conflict sequence, and the high-priority conflict points are resolved first.

[0036] Preferably, the multiple target driving paths are gamed by traversing the path strategy space according to the path conflict sequence. Specifically, based on the path conflict sequence, the conflict points are processed one by one. The vehicle, as a game participant, resolves the conflict by adjusting the path selection. Each vehicle selects a path and expects other vehicles to select paths that do not conflict with itself, including the vehicle adjusting the path strategy (such as changing lanes, changing speed) at the conflict point. Each vehicle adjusts its own strategy according to the strategies of other vehicles to obtain a game path solution, that is, a state where each vehicle cannot further optimize its path decision without changing its own strategy, that is, a stable optimal solution is reached. When the game path solution reaches the Nash equilibrium, that is, each vehicle selects an optimal path, and in this state, no vehicle can obtain a better result by unilaterally changing its path, it is considered that the paths of all vehicles have reached the best collaborative state, and thus the first path combination is output. There is no conflict in the path selection of all vehicles, and the overall efficiency of path planning is the highest and the cost is the lowest, ensuring that the driverless vehicles can drive efficiently and safely in complex traffic scenarios.

[0037] Step S400, traverse the first path combination for single-vehicle trajectory tracking to obtain multiple single-trajectory tracking results, and optimize the first path combination according to the multiple single-trajectory tracking results to generate a second path combination.

[0038] Preferably, all vehicle paths in the first path combination are traversed for single-vehicle trajectory tracking (calculating the specific execution trajectory for each vehicle's path), that is, based on the preliminary path of each vehicle, calculating the change trajectory of its position, speed, acceleration and other states at each moment during execution, and then evaluating whether the path is feasible and safe, and further evaluating the dynamic performance of the vehicle when executing the path, so as to obtain multiple single-trajectory tracking results. The path of each vehicle may change its execution trajectory due to factors such as traffic conditions and mutual influence between vehicles. Among them, the single-trajectory tracking result refers to the tracking result of the path simulation of each vehicle, including the position trajectory (the position of the vehicle at different time points), speed change (the change of the vehicle's speed over time), acceleration and braking state (the acceleration, deceleration and braking behavior of each vehicle when executing the path), and the influence of other vehicles and the environment, such as the distance to the vehicle in front and the presence of obstacles. The multiple single-trajectory tracking results correspond to the tracking result set of all vehicles, reflecting the dynamic response of each vehicle when executing the first path combination.

[0039] Preferably, by analyzing the tracking results of multiple vehicle trajectories, the feasibility and safety of the current path combination are evaluated, including discovering potential problems based on the single-vehicle trajectory tracking results and adjusting the path to improve system efficiency and safety. For example, reducing the collision risk (avoiding collisions or being too close between vehicles by adjusting the path), improving traffic flow (reducing congestion or unnecessary deceleration, stopping, etc., and optimizing traffic flow), and improving energy efficiency (reducing the energy consumption of vehicles by path adjustment, such as reducing hard braking and acceleration), and then generating a second path combination. That is, by optimizing the first path combination, considering the problems and potential risks found in the trajectory tracking results, providing a more efficient and safer driving path, better solving the conflicts in the path, reducing unnecessary acceleration and deceleration, improving the coordination between vehicles and the smoothness of the whole system in terms of performance, and ensuring that the driverless vehicle can perform cooperative driving control more precisely and intelligently in a complex and dynamic traffic environment.

[0040] Further, step S400 further includes step S410 of simulating the driving of multiple driverless vehicles according to the first path combination to obtain a simulation driving result; step S420 of performing single-vehicle trajectory tracking based on the simulation driving result to generate M pieces of vehicle simulation driving trajectory data; step S430 of setting expected trajectory data, determining whether the M pieces of vehicle simulation driving trajectory data conform to the expected trajectory data, and if the M pieces of vehicle simulation driving trajectory data do not conform to the expected trajectory data, generating a deviation prompt, calculating the deviation of the M pieces of vehicle simulation driving trajectory data according to the deviation prompt to generate multiple trajectory deviation values; step S440 of performing overlapping analysis on the M pieces of vehicle simulation driving trajectory data according to the time-space diagram based on the multiple trajectory deviation values, setting a deviation critical value, extracting the trajectory deviation values greater than or equal to the deviation critical value, and matching and determining N pieces of vehicle simulation driving trajectory data, where N is a positive integer greater than or equal to 0 and less than or equal to M; step S450 of adding the N pieces of vehicle simulation driving trajectory data to the multiple single-trajectory tracking results.

[0041] Preferably, simulating the driving of multiple driverless vehicles according to the first path combination means performing driving simulation in a virtual environment to simulate the actual driving process of each vehicle on this path and obtaining a simulation driving result, including the trajectory and state data (such as speed, position, acceleration, etc.) collected by each vehicle during the simulation execution process. Then, tracking the simulation driving result of each vehicle one by one to generate the specific simulation driving trajectory data of this vehicle, including the change of the position of the vehicle over time during the simulation driving and dynamic parameters such as speed, acceleration, and steering angle. Through tracking, each vehicle generates a set of simulation driving trajectory data, that is, M sets of trajectory data are generated, where M is a positive integer representing the total number of driverless vehicles; presetting expected trajectory data, which represents the driving trajectory of the vehicle under ideal conditions and is used to evaluate whether the actual simulation driving of the vehicle reaches the expected goal.

[0042] Preferably, comparing the simulation trajectories of M vehicles with the expected trajectory to determine whether the simulation driving trajectory data conforms to the expected trajectory data includes comparing whether it conforms to the expectation in terms of space (such as position, path) and whether it meets the requirements in terms of time (such as speed, time to reach the target point). If it does not conform, a deviation prompt is generated, that is, the deviated area or parameters are marked, such as position deviation, time deviation, etc. Then, according to the deviation prompt, the deviation between the simulation trajectory and the expected trajectory of each vehicle is quantified to generate multiple trajectory deviation values, and the trajectory deviation value represents the degree of trajectory deviation, including space deviation and time deviation.

[0043] Preferably, overlapping analysis is performed on the M vehicle simulated driving trajectory data based on multiple trajectory deviation values. Specifically, the trajectory deviation values are mapped to a time-space graph to analyze the specific distribution and impact of the deviated trajectories, with a focus on the regions with larger deviation values to determine the critical time periods and spatial points of the problem. A deviation threshold value (marking severely deviated trajectories) is set, and the trajectories with deviation values greater than or equal to the threshold value are extracted. N vehicle simulated driving trajectory data (N is a positive integer less than or equal to M, representing the number of vehicles meeting the conditions), that is, the part with larger deviation values selected from the M vehicle simulated driving trajectory data, are added to the previous single-vehicle trajectory tracking results as the input for further analysis and optimization. By comprehensively considering all problem trajectories, the accuracy of the overall path planning and driving control is improved, thereby ensuring that the driverless vehicle can gradually optimize its driving behavior and enhance the overall coordination and execution efficiency.

[0044] Further, step S400 further includes step S460 of locally controlling multiple driverless vehicles based on the N vehicle simulated driving trajectory data to generate abnormal control information; step S470 of updating the N vehicle simulated driving trajectory data according to the abnormal control information to generate N trajectory update data; step S480 of performing fitness evaluation on the N trajectory update data based on the first path combination to obtain multiple fitness values; and step S490 of replacing and optimizing the first path combination with the N trajectory update data according to the multiple fitness values to generate the second path combination.

[0045] Preferably, local control of the N vehicle trajectory data (trajectories deviating from the expected trajectory or showing abnormal behavior) selected from the simulation includes solving local trajectory deviation problems, such as adjusting the vehicle speed, steering angle, or acceleration / deceleration, to improve the conformity of the trajectory with the expected path and correct potential conflicts or efficiency degradation problems caused by abnormal behavior, generating abnormal control information for describing how to adjust the driving behavior of the N vehicles. For example, it is prompted that the first vehicle should reduce its speed to better match the planned path, and the second vehicle is advised to turn to avoid leaving the predetermined lane; according to the abnormal control information, the simulated trajectory data of the N vehicles are updated, that is, the trajectory point positions, speeds, and accelerations are updated, generating new trajectory data called trajectory update data, making the trajectories of the vehicles closer to the expected trajectory and reducing the deviation degree.

[0046] Preferably, the fitness of the N trajectory update data is evaluated according to the first path combination to measure the matching degree between the updated trajectory data and the first path combination. Among them, the evaluation indicators may include trajectory deviation, time consistency, and overall safety. A fitness value is calculated for the trajectory update data of each vehicle, that is, N fitness values are obtained. The higher the fitness value, the more the updated trajectory meets the requirements of the planned path. Finally, the N trajectory update data are used to replace and optimize the first path combination according to the fitness value, including locally replacing the original trajectory in the first path combination and replacing the trajectory with a large deviation with the updated trajectory data. For example, according to the fitness value, the updated trajectory with a higher fitness is selected for replacement. By replacing the trajectory update data, the deviation of the overall path planning is reduced, the global adaptability and safety of the path are improved, and then the second path combination is generated. The overall path better conforms to the actual driving conditions of the vehicle, reduces deviation, solves the problem of path deviation of driverless vehicles in complex scenarios, and improves the safety and efficiency of vehicle cooperative driving through trajectory update and global optimization.

[0047] Step S500, perform cooperative learning on multiple driverless vehicles according to the second path combination, formulate a multi-vehicle cooperation strategy based on the learning feedback results, and perform multi-vehicle intelligent cooperative driving control on the driverless vehicles based on the multi-vehicle cooperation strategy.

[0048] Preferably, cooperative learning is performed on multiple driverless vehicles according to the second path combination, that is, each vehicle starts to learn and adjust its own behavior according to the execution path and environmental feedback. Specifically, each vehicle performs the driving task according to the second path combination, and real-time collects data during the execution process, including speed, acceleration, relative position with other vehicles, obstacle handling, etc. Each vehicle generates feedback information according to the performance of its own execution path (such as whether there is a collision, whether the traffic flow is smooth, energy consumption, etc.) and the interaction with other vehicles. The vehicle shares the data and status in its execution path in real time through the vehicle network (V2X), not only learns from its own path feedback, but also obtains information from the execution of other vehicles, and learns the strategies and optimization measures taken by other vehicles in path execution. Through information exchange among multiple vehicles, all vehicles can continuously adjust and optimize their strategies in various scenarios to achieve more efficient cooperation. Among them, cooperative learning refers to the process in which multiple vehicles share information, learn each other's behaviors and feedbacks during the execution of tasks to improve the performance of the overall system. Each vehicle not only adjusts its behavior according to its own feedback in path execution, but also learns according to the feedback information of other vehicles, and finally forms an optimized cooperation strategy.

[0049] Preferably, feedback results are generated based on the performance of the driverless vehicles during actual driving (such as whether they pass through intersections smoothly, whether collisions occur, whether energy is saved, etc.). For example, safety feedback (such as whether collisions occur, whether there is a relatively high collision risk), efficiency feedback (such as whether the shortest driving time and minimum energy consumption are achieved), coordination feedback (such as whether vehicles cooperate smoothly with each other and whether traffic congestion is avoided), etc. A multi-vehicle collaborative strategy is formulated based on the feedback results of multiple driverless vehicles, that is, how multiple vehicles jointly adjust their strategies according to each other's states and feedback to ensure that all vehicles in the fleet are coordinated and achieve common goals (such as safety, efficiency, fluency, etc.), including path coordination, conflict avoidance, efficiency improvement, and real-time adjustment according to changing traffic conditions, environmental factors, and the behaviors of other vehicles to adapt to real-time traffic conditions. Finally, the multi-vehicle collaborative strategy is applied to the driving decisions of each vehicle for intelligent collaborative driving control, enabling it to drive intelligently and coordinately according to real-time situations, thereby ensuring the efficient and safe operation of multiple driverless vehicles as a whole.

[0050] Further, step S500 further includes step S510 of training the states of multiple driverless vehicles according to the second path combination through reinforcement learning to construct a state information space; step S520 of training the actions of multiple driverless vehicles according to the second path combination through reinforcement learning to construct an action feedback space; step S530 of introducing a reward function to perform fusion analysis on the state information space and the action feedback space to determine the global collaborative learning results of multiple driverless vehicles; step S540 of performing multi-agent reinforcement learning according to the global collaborative learning results to generate the learning feedback results, where the learning feedback results include positive feedback learning results and negative feedback learning results; step S550 of implementing behavior rewards based on the positive feedback learning results and implementing punishments based on the negative feedback learning results to generate a multi-vehicle collaborative strategy, and sending the multi-vehicle collaborative strategy to multiple driverless vehicles for multi-vehicle intelligent collaborative driving control.

[0051] Preferably, the state training of multiple driverless vehicles is carried out according to the second path combination through reinforcement learning, that is, the states that the vehicles may experience during actual driving are simulated and trained, learning how to perceive and express these states, obtaining various possible states of multiple driverless vehicles, forming a state information space, which defines the state distribution when the vehicle interacts with the environment; then, the action training is carried out based on the second path combination through reinforcement learning, enabling the vehicle to learn to take adaptive actions in different states, obtaining the actions that the vehicle may take and the results generated in different states, constituting an action feedback space, including the effects of actions (such as avoiding conflicts and improving efficiency), and whether the actions conform to the global cooperation goal; then, a reward function is introduced to integrate the state information space and the action feedback space. Among them, the reward function defines the reward obtained by the vehicle after taking a certain action, which is used to guide the learning process. The reward may be based on path efficiency (such as completing the path in the shortest time), safety (such as avoiding collisions and maintaining a reasonable vehicle distance), and energy consumption (such as reducing sudden acceleration or braking). The reward function comprehensively analyzes the state information space and the action feedback space, evaluates the performance of each state-action pair, and the output result is the global value of the vehicle taking a specific action in different states, thereby determining the global cooperative learning result of multiple driverless vehicles, that is, the set of the best behavior strategies for multi-vehicle cooperation, which reflects the global optimal strategy of the entire vehicle fleet in a dynamic traffic scenario.

[0052] Preferably, multi-agent reinforcement learning is carried out according to the global cooperative learning result. Specifically, multi-agent reinforcement learning (MARL) regards multiple driverless vehicles as independent agents, collaboratively optimizing their learning processes. Each vehicle acts as an agent, interacts with other vehicles and the environment, and learns its optimal strategy. Vehicles enhance the overall performance through information sharing and cooperative learning, compete for learning under limited resources (such as road space), optimize the conflict resolution ability between vehicles, and then generate learning feedback results, including positive feedback learning results (the actions executed by the vehicle conform to the global goal and obtain rewards) and negative feedback learning results (negative feedback learning results: the actions executed by the vehicle deviate from the goal and trigger penalties). Then, rewards and penalties are implemented based on the feedback results. The vehicles with positive feedback learning results are rewarded to strengthen their strategies, and the vehicles with negative feedback learning results are penalized to suppress their incorrect behaviors. The global cooperative strategy among multiple vehicles is optimized according to the reward and penalty results, enabling the entire vehicle fleet to perform better in a dynamic environment. Finally, the generated optimized cooperative strategy is distributed to each vehicle, enabling it to guide its behavior according to the strategy during actual driving. For example, adjusting the driving path, optimizing speed and acceleration, avoiding conflicts, and maintaining the vehicle fleet formation, realizing intelligent cooperative driving control, that is, the vehicle autonomously adjusts its behavior based on the cooperative strategy in a dynamic environment, ensuring the intelligent cooperative driving of the driverless vehicle fleet and ensuring a safe, efficient, and reliable driving experience.

[0053] In the above text, with reference to Figure 1A multi-vehicle collaborative driving control method for driverless vehicles according to an embodiment of the present invention is described in detail. Next, reference will be made to Figure 2 Describe a multi-vehicle collaborative driving control system for driverless vehicles according to an embodiment of the present invention.

[0054] The multi-vehicle collaborative driving control system for driverless vehicles according to an embodiment of the present invention is used to solve the technical problems existing in the prior art, such as limited single-vehicle environmental perception range, insufficient vehicle-to-vehicle communication, and difficulty in coping with complex and changeable traffic scenarios, resulting in poor multi-vehicle collaborative driving efficiency and safety. It realizes the intelligent control of driverless vehicles in dynamic traffic scenarios and achieves the technical effect of improving multi-vehicle collaborative driving efficiency and driving safety. As Figure 2 shown, the multi-vehicle collaborative driving control system for driverless vehicles includes: a multi-dimensional sensing data set acquisition module 10, a dynamic traffic scenario model construction module 20, a path conflict analysis module 30, a path combination optimization module 40, and a multi-vehicle collaborative strategy formulation module 50.

[0055] The multi-dimensional sensing data set acquisition module 10 is used to perform multi-dimensional sensing on multiple driverless vehicles through multi-sensors deployed on the vehicles to obtain a multi-dimensional sensing data set, and the multi-dimensional sensing data set includes multiple vehicle state data sets and multiple vehicle environment data sets; the dynamic traffic scenario model construction module 20 is used to synchronize the multiple vehicle state data sets to the vehicle network to establish a multi-vehicle communication network, and perform environmental modeling based on the multiple vehicle environment data sets to construct a dynamic traffic scenario model; the path conflict analysis module 30 is used to integrate the dynamic traffic scenario model with the multi-vehicle communication network to obtain a multi-vehicle collaboration platform, and perform multi-vehicle path conflict analysis on multiple driverless vehicles through the multi-vehicle collaboration platform to generate a first path combination; the path combination optimization module 40 is used to traverse the first path combination to perform single-vehicle trajectory tracking, obtain multiple single-trajectory tracking results, and optimize the first path combination according to the multiple single-trajectory tracking results to generate a second path combination; the multi-vehicle collaborative strategy formulation module 50 is used to perform collaborative learning on multiple driverless vehicles according to the second path combination, formulate a multi-vehicle collaborative strategy according to the learning feedback results, and perform multi-vehicle intelligent collaborative driving control on the driverless vehicles based on the multi-vehicle collaborative strategy.

[0056] Next, the specific configuration of the dynamic traffic scenario model construction module 20 will be described in detail. The dynamic traffic scenario model construction module 20 further includes: performing synchronization processing on the multiple vehicle state data sets to generate a multi-vehicle state alignment data set; using multiple driverless vehicles as multiple communication nodes, traversing the multiple communication nodes for communication association analysis to determine multiple correlation coefficients; connecting the multi-vehicle state alignment data set according to the multiple communication nodes and multiple correlation coefficients to construct a multi-vehicle communication network; performing Gaussian process regression on the multiple vehicle environment data sets to construct an environmental space distribution model; using a graph neural network to traverse the multiple vehicle environment data sets to establish the multi-vehicle communication network for interaction to obtain dynamic traffic flow parameters; mapping the dynamic traffic flow parameters to the environmental space distribution model to obtain the dynamic traffic scenario model.

[0057] Next, the specific configuration of the path conflict analysis module 30 will be described in detail. The path conflict analysis module 30 further includes: using multiple driverless vehicles as multiple game participants for driving analysis to determine multiple driving paths, where the multiple driving paths have a corresponding relationship with the multiple game participants; extracting multiple starting positions and multiple ending positions, and merging the multiple driving paths according to the multiple starting positions and the multiple ending positions to construct a path strategy space for multiple vehicles; mapping the multiple game participants to the path strategy space, traversing the multiple starting positions and the multiple ending positions for intersection detection, and performing a path conflict game according to the path intersection information to generate the first path combination.

[0058] Next, the specific configuration of the path conflict analysis module 30 will be further described in detail. The path conflict analysis module 30 further includes: using the multiple game participants as indexes, traversing the path strategy space for retrieval to determine multiple target driving paths, where the multiple target driving paths have a corresponding relationship with the multiple game participants; performing spatio-temporal analysis based on the multiple target driving paths to construct a time-space graph; performing intersection determination according to the multiple target driving paths to generate a determination result; if there is an intersection in the determination result, then perform overlapping analysis according to the multiple starting positions, the multiple ending positions, and the time-space graph to generate an overlapping signal; performing matching identification on the multiple target driving paths according to the overlapping signal to determine multiple path conflict points, performing driving impact analysis on the multiple path conflict points, sorting the multiple path conflict points in descending order of priority according to the impact analysis result to determine a path conflict sequence; traversing the path strategy space according to the path conflict sequence to perform a game on the multiple target driving paths to obtain a game path solution, and when the game path solution reaches a Nash equilibrium, then generate the first path combination.

[0059] Next, the specific configuration of the path combination optimization module 40 will be described in detail. The path combination optimization module 40 further includes: simulating the driving of multiple driverless vehicles according to the first path combination to obtain a simulation driving result; performing single-vehicle trajectory tracking based on the simulation driving result to generate M vehicle simulation driving trajectory data; setting expected trajectory data, and determining whether the M vehicle simulation driving trajectory data conforms to the expected trajectory data. If the M vehicle simulation driving trajectory data does not conform to the expected trajectory data, a deviation prompt is generated, and deviation calculation is performed on the M vehicle simulation driving trajectory data according to the deviation prompt to generate multiple trajectory deviation values; based on the multiple trajectory deviation values, overlapping analysis of the M vehicle simulation driving trajectory data is performed according to the time-space diagram, a deviation threshold is set, and trajectory deviation values greater than or equal to the deviation threshold are extracted, and N vehicle simulation driving trajectory data are matched and determined, where N is a positive integer greater than or equal to 0 and less than or equal to M; adding the N vehicle simulation driving trajectory data to the multiple single-trajectory tracking results.

[0060] Next, the specific configuration of the path combination optimization module 40 will be further described in detail. The path combination optimization module 40 further includes: performing local control on multiple driverless vehicles based on the N vehicle simulation driving trajectory data to generate abnormal control information; updating the N vehicle simulation driving trajectory data according to the abnormal control information to generate N trajectory update data; performing fitness evaluation on the N trajectory update data based on the first path combination to obtain multiple fitness values; replacing and optimizing the first path combination with the N trajectory update data according to the multiple fitness values to generate the second path combination.

[0061] Next, the specific configuration of the multi-vehicle cooperation strategy formulation module 50 will be described in detail. The multi-vehicle cooperation strategy formulation module 50 further includes: constructing a state information space by performing state training on multiple driverless vehicles according to the second path combination through reinforcement learning; constructing an action feedback space by performing action training on multiple driverless vehicles according to the second path combination through reinforcement learning; introducing a reward function to perform fusion analysis on the state information space and the action feedback space to determine the global cooperation learning result of multiple driverless vehicles; performing multi-agent reinforcement learning according to the global cooperation learning result to generate the learning feedback result, where the learning feedback result includes a positive feedback learning result and a negative feedback learning result; implementing behavior rewards based on the positive feedback learning result and implementing punishments based on the negative feedback learning result to generate a multi-vehicle cooperation strategy, and sending the multi-vehicle cooperation strategy to multiple driverless vehicles for multi-vehicle intelligent cooperation driving control.

[0062] The multi-vehicle cooperative driving control system of the driverless vehicle provided by the embodiments of the present invention can execute the multi-vehicle cooperative driving control method of the driverless vehicle provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects for executing the method.

[0063] Although this application makes various references to certain modules in the system according to the embodiments of the present application, however, any number of different modules can be used and run on the user terminal and / or the server. The included individual units and modules are only divided according to the functional logic, but are not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific names of the functional units are only for the convenience of mutual distinction and do not limit the protection scope of the present invention.

[0064] The above specific embodiments do not constitute a limitation to the protection scope of this application. Those skilled in the art should understand that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principle of this application shall be included within the protection scope of this application.

Claims

1. A multi-vehicle cooperative driving control method for an unmanned vehicle, characterized in that: The method comprises: Performing multi-dimensional sensing on multiple unmanned vehicles by using multiple sensors deployed on the vehicles to obtain a multi-dimensional sensing data set, wherein the multi-dimensional sensing data set includes multiple vehicle state data sets and multiple vehicle environment data sets; Synchronizing the multiple vehicle status data sets to the Internet of Vehicles to establish a multi-vehicle communication network, performing environmental modeling based on the multiple vehicle environment data sets, and constructing a dynamic traffic scenario model; Integrating the dynamic traffic scenario model with the multi-vehicle communication network to obtain a multi-vehicle collaborative platform, performing multi-vehicle path conflict analysis on multiple unmanned vehicles through the multi-vehicle collaborative platform to generate a first path combination; Traversing the first path combination to perform single-vehicle trajectory tracking to obtain multiple single-trajectory tracking results, optimizing the first path combination according to the multiple single-trajectory tracking results to generate a second path combination; Collaborative learning is performed on multiple unmanned vehicles according to the second path combination, a multi-vehicle collaborative strategy is formulated according to the learning feedback results, and multi-vehicle intelligent collaborative driving control is performed on the unmanned vehicles based on the multi-vehicle collaborative strategy.

2. The multi-vehicle cooperative driving control method of an unmanned vehicle according to claim 1, characterized in that: The method comprises: synchronizing the plurality of vehicle status data sets to a vehicle network to establish a multi-vehicle communication network, performing environmental modeling based on the plurality of vehicle environment data sets, and constructing a dynamic traffic scene model. Perform synchronization processing based on the multiple vehicle status data sets to generate a multi-vehicle status alignment data set; Taking a plurality of unmanned vehicles as a plurality of communication nodes, traversing the plurality of communication nodes to perform communication correlation analysis, and determining a plurality of correlation coefficients; Connecting the multi-vehicle state alignment data sets according to the multiple communication nodes and the multiple correlation coefficients to construct a multi-vehicle communication network; Performing Gaussian process regression based on the multiple vehicle environment data sets to construct an environment space distribution model; Using a graph neural network to traverse the multiple vehicle environment data sets to establish the multi-vehicle communication network for interaction, and obtain dynamic traffic flow parameters; The dynamic traffic flow parameters are mapped to the environmental space distribution model to obtain the dynamic traffic scene model.

3. The multi-vehicle cooperative driving control method of an unmanned vehicle according to claim 1, characterized in that: The dynamic traffic scenario model is integrated with the multi-vehicle communication network to obtain a multi-vehicle collaborative platform, and a multi-vehicle path conflict analysis is performed on a plurality of unmanned vehicles through the multi-vehicle collaborative platform to generate a first path combination, the method comprising: Performing driving analysis on a plurality of unmanned vehicles as a plurality of game participants to determine a plurality of driving paths, wherein the plurality of driving paths correspond to the plurality of game participants; Extracting a plurality of starting positions and a plurality of end positions, merging the plurality of driving paths according to the plurality of starting positions and the plurality of end positions, and constructing a path strategy space for multiple vehicles; The multiple game participants are mapped to the path strategy space, the multiple starting positions and the multiple end positions are traversed to perform intersection detection, and a path conflict game is performed according to the path intersection information to generate the first path combination.

4. The multi-vehicle cooperative driving control method for an unmanned vehicle as claimed in claim 3, characterized in that: The method includes mapping the multiple game participants to the path strategy space, traversing the multiple starting positions and the multiple end positions to perform intersection detection, performing path conflict game according to path intersection information, and generating the first path combination. Using the multiple game participants as indexes, traversing the path strategy space for searching, and determining multiple target driving paths, wherein the multiple target driving paths correspond to the multiple game participants; Performing spatiotemporal analysis based on the multiple target driving paths to construct a time-space graph; Performing intersection determination according to the multiple target driving paths and generating a determination result; If there is an intersection in the determination results, performing overlap analysis according to the multiple starting positions and the multiple end positions according to the time-space diagram to generate an overlap signal; Matching and identifying the multiple target driving paths according to the overlapping signals, determining multiple path conflict points, performing driving impact analysis on the multiple path conflict points, sorting the multiple path conflict points in descending order of priority according to the impact analysis results, and determining a path conflict sequence; The path strategy space is traversed according to the path conflict sequence to perform a game on the multiple target driving paths to obtain a game path solution, and when the game path solution reaches a Nash equilibrium, the first path combination is generated.

5. The multi-vehicle cooperative driving control method of an unmanned vehicle as claimed in claim 4, characterized in that: Traversing the first path combination to perform single-vehicle trajectory tracking to obtain multiple single-trajectory tracking results, the method includes: Performing simulated driving on multiple unmanned vehicles according to the first path combination to obtain simulated driving results; Performing single-vehicle trajectory tracking based on the simulated driving results to generate M vehicle simulated driving trajectory data; Setting expected trajectory data, determining whether the M vehicle simulated driving trajectory data conform to the expected trajectory data, generating a deviation prompt if the M vehicle simulated driving trajectory data do not conform to the expected trajectory data, performing deviation calculation on the M vehicle simulated driving trajectory data according to the deviation prompt, and generating a plurality of trajectory deviation values; Based on the multiple trajectory deviation values, the M vehicle simulated driving trajectory data are overlapped and analyzed according to the time-space graph, a deviation critical value is set, a trajectory deviation value greater than or equal to the deviation critical value is extracted, and N vehicle simulated driving trajectory data are matched and determined, where N is a positive integer greater than or equal to 0 and less than or equal to M; The N vehicle simulated driving trajectory data are added to the multiple single trajectory tracking results.

6. The multi-vehicle cooperative driving control method of an unmanned vehicle as claimed in claim 5, characterized in that: The first path combination is optimized according to the plurality of single trajectory tracking results to generate a second path combination, the method comprising: Performing local control on a plurality of unmanned vehicles based on the N simulated driving trajectory data of the vehicles to generate abnormal control information; The N simulated driving trajectory data of the vehicles are updated according to the abnormal control information to generate N trajectory update data; Performing fitness evaluation on the N trajectory update data based on the first path combination to obtain multiple fitness values; The N trajectory update data are replaced and optimized for the first path combination according to the multiple fitness values ​​to generate the second path combination.

7. The multi-vehicle cooperative driving control method of an unmanned vehicle as claimed in claim 1, characterized in that: The method includes: performing collaborative learning on multiple unmanned vehicles according to the second path combination, formulating a multi-vehicle collaborative strategy according to the learning feedback result, and performing multi-vehicle intelligent collaborative driving control on the unmanned vehicles based on the multi-vehicle collaborative strategy. Performing state training on multiple unmanned vehicles according to the second path combination through reinforcement learning to construct a state information space; Performing action training on multiple unmanned vehicles according to the second path combination through reinforcement learning to construct an action feedback space; A reward function is introduced to fuse and analyze the state information space and the action feedback space to determine a global collaborative learning result of multiple unmanned vehicles; Perform multi-agent reinforcement learning according to the global collaborative learning result to generate the learning feedback result, wherein the learning feedback result includes positive feedback learning result and negative feedback learning result; Behavior rewards are implemented based on the positive feedback learning results, and penalties are implemented based on the negative feedback learning results, so as to generate a multi-vehicle collaborative strategy, and send the multi-vehicle collaborative strategy to multiple unmanned vehicles for multi-vehicle intelligent collaborative driving control.

8. A multi-vehicle cooperative driving control system for an unmanned vehicle, characterized in that: The system is used to implement the multi-vehicle cooperative driving control method for an unmanned vehicle according to any one of claims 1 to 7, and the system comprises: A multi-dimensional sensing data set acquisition module, used to perform multi-dimensional sensing on multiple unmanned vehicles through multiple sensors deployed on the vehicles to obtain a multi-dimensional sensing data set, wherein the multi-dimensional sensing data set includes multiple vehicle state data sets and multiple vehicle environment data sets; A dynamic traffic scene model building module, used to synchronize the multiple vehicle status data sets to the Internet of Vehicles to establish a multi-vehicle communication network, perform environmental modeling based on the multiple vehicle environment data sets, and build a dynamic traffic scene model; A path conflict analysis module, used for integrating the dynamic traffic scenario model with the multi-vehicle communication network to obtain a multi-vehicle collaborative platform, performing multi-vehicle path conflict analysis on a plurality of unmanned vehicles through the multi-vehicle collaborative platform, and generating a first path combination; a path combination optimization module, configured to traverse the first path combination to perform single-vehicle trajectory tracking, obtain multiple single-trajectory tracking results, optimize the first path combination according to the multiple single-trajectory tracking results, and generate a second path combination; A multi-vehicle collaborative strategy formulation module is used to perform collaborative learning on multiple unmanned vehicles according to the second path combination, formulate a multi-vehicle collaborative strategy based on the learning feedback results, and perform multi-vehicle intelligent collaborative driving control on the unmanned vehicles based on the multi-vehicle collaborative strategy.

Images