Vehicle-road-cloud integrated multi-vehicle cooperative simulation decision method and system

Abstract

This application discloses a multi-vehicle cooperative simulation decision-making method and system integrating vehicle, road, and cloud technologies. It employs a fast filtering algorithm based on intent spectrum and conflict cost matrix to compress the combinatorial explosion problem into a feasible solution space, significantly improving cloud-based solution efficiency. A cooperative weight bidding mechanism is introduced to resolve conflicts while considering system benefits and vehicle preferences, ensuring the fairness and optimality of the scheduling scheme. Vehicles perform local replanning and safety verification based on cloud-based suggestions, adhering to cooperative intent while maintaining local real-time response capabilities. The system's cooperative status is monitored through a consistency index; when a deviation is triggered, only the deviating vehicle needs to provide a new intent spectrum, allowing for rapid cloud-based optimization and simplified scheduling. This avoids large-scale global replanning, enabling low-latency online dynamic adjustment of cooperative strategies and significantly improving the system's long-term stable operation in complex traffic scenarios.

Description

Technical Field

[0001] This application relates to the field of vehicle-road-cloud integration technology, specifically to a multi-vehicle collaborative simulation decision-making method and system for vehicle-road-cloud integration. Background Technology

[0002] With the development of intelligent connected vehicle technology, multi-vehicle collaborative planning has become crucial for improving traffic efficiency and safety. Existing collaborative solutions mainly fall into two categories: centralized scheduling, where the cloud plans trajectories for all vehicles, ensuring global optimization but facing high computational complexity and poor dynamic scalability, making it difficult to handle dense traffic scenarios; and distributed decision-making, where each vehicle plans independently based on its own perception and coordinates through vehicle-to-vehicle communication, but this is prone to a "prisoner's dilemma" due to conflicting individual intentions, making it difficult to form a globally consistent collaborative strategy. Furthermore, existing methods generally lack explicit modeling and game theory mechanisms for vehicle driving intentions, and struggle to quickly resolve intention conflicts and achieve online adaptive adjustments in dynamic traffic flows, resulting in insufficient collaborative efficiency and robustness in complex scenarios.

[0003] Regarding the above-mentioned solutions, the inventors of this application have discovered that the above-mentioned technology has at least the following technical problems:

[0004] 1. Existing technologies lack a fast collaborative filtering mechanism based on the conflict cost matrix. When faced with combinations of multiple vehicle intentions, they typically employ exhaustive search or simple pairwise collision detection, which easily leads to combinatorial explosion, resulting in excessively high computational complexity and large response latency, making it difficult to meet millisecond-level real-time requirements. Existing methods often directly output a unique solution, lacking redundancy and compressing the decision space for subsequent optimization and scheduling. Furthermore, traditional conflict judgments are mostly limited to simple collision detection between pairs of vehicles, lacking the ability to globally quantify and evaluate multi-vehicle intention conflicts in complex interaction scenarios.

[0005] 2. Existing technologies lack effective weighted bidding mechanisms in collaborative scheduling, typically relying on fixed rules or simple priorities for decision-making, making it difficult to balance individual utility with system benefits. This can easily lead to "aggressive" vehicles continuously seizing resources, resulting in a lack of fairness; or excessive accommodation of weaker vehicles, causing a significant drop in overall traffic efficiency. The final selected scheduling scheme often contains implicit conflicts on a global scale, making it difficult to guarantee comprehensive benefits.

[0006] 3. Existing technologies mostly employ "cloud-based remote operation" or vehicle-independent planning models, with control commands directly issued from the cloud. This is highly susceptible to communication delays and uncertainties, making it difficult to guarantee the real-time performance and stability of control. The vehicle side lacks a hierarchical planning mechanism that combines cloud guidance with local constraints, making it difficult to use travel time windows and trajectory suggestions as hard constraints for local replanning. Furthermore, the lack of high-frequency reporting and cloud-consistency monitoring mechanisms limits monitoring granularity to the "individual vehicle" level, failing to perceive the overall operational status from a "system collaboration" perspective. Minor deviations can easily accumulate into serious conflicts, and the cloud cannot detect collaboration failure risks in advance, resulting in delayed risk discovery. Summary of the Invention

[0007] To address the aforementioned technical shortcomings, the purpose of this application is to provide a multi-vehicle collaborative simulation decision-making method and system that integrates vehicle, road, and cloud technologies.

[0008] To solve the above technical problems, this application adopts the following technical solution: In the first aspect, this application provides a multi-vehicle cooperative simulation decision-making method integrating vehicle, road and cloud, which includes the following steps: S1, the simulated vehicle generates a local fusion situation map based on local sensor and V2X communication data, and constitutes the local candidate intent spectrum of the simulated vehicle;

[0009] S2. Based on the local candidate intent spectrum of each vehicle, analyze and derive a set of feasible intent combinations;

[0010] S3. Based on each feasible intention combination, analyze and obtain the precise spatiotemporal resource set occupied by each vehicle in the combination, and generate the corresponding resource scheduling package.

[0011] S4. Based on resource scheduling packages, the cloud analyzes and derives the optimal collaborative scheduling scheme.

[0012] S5. Based on the travel time window and suggested reference trajectory allocated to the vehicle in the optimal collaborative scheduling scheme, and combined with the latest local environment information, the vehicle end performs constrained trajectory replanning and verification, and generates and prepares to execute an accurate and feasible trajectory.

[0013] S6. Based on the vehicle's precise and feasible trajectory, the cloud and the vehicle work together to monitor the execution status and analyze and derive collaborative consistency indicators.

[0014] S7. Analyze the collaborative consistency indicators and trigger the dynamic optimization process.

[0015] Preferably, the simulated vehicle generates a local fusion situation map based on local sensor and V2X communication data, and constitutes a local candidate intent spectrum for the simulated vehicle, including:

[0016] S201, Based on the multi-source sensor data and V2X communication data, perform data fusion operation to generate a local fusion situation map describing the vehicle's surrounding environment and dynamic target status;

[0017] S202, Based on the local fusion situation map, perform a multi-strategy trajectory generation operation on the simulated vehicle to obtain a candidate trajectory set containing at least two candidate trajectories and their corresponding driving intention labels;

[0018] S203, based on a preset cost evaluation function, perform an initial preference weight calculation operation on each candidate trajectory in the candidate trajectory set to obtain the initial preference weight of each candidate trajectory;

[0019] S204. Based on the candidate trajectory set, the corresponding driving intention labels and the initial preference weights, perform an intention spectrum structured encapsulation operation to generate and upload the local candidate intention spectrum of the simulation vehicle.

[0020] Preferably, the step of analyzing and deriving a set of feasible intent combinations based on the local candidate intent spectrum of each vehicle includes:

[0021] S301, based on the local candidate intent spectrum of each vehicle, perform intent tag extraction operation to obtain a global intent tag set containing each candidate driving intent of each vehicle;

[0022] S302, Based on the global intent tag set and intent conflict cost matrix, run a fast collaborative filtering algorithm to obtain the global intent combination space;

[0023] S303, based on the intent conflict cost matrix and the global intent combination space, perform a total conflict cost enumeration calculation operation to obtain the total conflict cost corresponding to each possible global intent combination;

[0024] S304, based on a preset conflict cost threshold, perform a cost filtering operation on each total conflict cost to select a set of feasible intent combinations.

[0025] Preferably, the step of performing a total conflict cost enumeration calculation to obtain the total conflict cost corresponding to each possible combination of global intentions includes:

[0026] Through calculation formula Determine the total cost of conflict ,in and This is represented as a vehicle index. This represents the total number of vehicles. Represented as the first The driving intent label selected by the vehicle in the current global intent combination. Represented as the first The driving intent label selected by the vehicle in the current global intent combination. Indicated as driving intent label and driving intention label The fundamental conflict between them represents value. This is represented as an optional weighting coefficient used to adjust the first... Vehicles and the first The importance of conflicts between vehicles.

[0027] Preferably, the step of analyzing the collaborative consistency index and triggering the dynamic optimization process includes:

[0028] S901 performs dynamic optimization trigger judgment operations based on system-level coordination consistency index and actual trajectory status of a single vehicle to identify vehicles with actual trajectory status where the coordination consistency index is less than a preset first threshold and the expected state deviation of the optimal coordination scheduling scheme is greater than a preset second threshold.

[0029] S902, based on the latest local fusion situation map of the deviation vehicle, perform a new local candidate intent spectrum generation operation to reflect the current situation, so as to generate and upload the new local candidate intent spectrum;

[0030] S903, the cloud runs a fast scheduling fine-tuning algorithm based on the new local candidate intent spectrum and the original optimal collaborative scheduling scheme to generate a new collaborative scheduling scheme after fine-tuning.

[0031] S904, the fine-tuned new collaborative scheduling scheme is sent to the relevant vehicles to achieve online dynamic adjustment.

[0032] This application provides a system for a vehicle-road-cloud integrated multi-vehicle cooperative simulation decision-making method in its second aspect, including:

[0033] Preferably, the local candidate intent spectrum generation module is used to generate a local fusion situation map based on local sensor and V2X communication data of the simulated vehicle, and to form the local candidate intent spectrum of the simulated vehicle.

[0034] The feasible intent combination set generation module analyzes and derives the feasible intent combination set based on the local candidate intent spectrum of each vehicle.

[0035] The precise spatiotemporal resource set and resource scheduling package generation module analyzes and derives the precise spatiotemporal resource set occupied by each vehicle in the combination based on each feasible intention combination, and generates the corresponding resource scheduling package.

[0036] The optimal collaborative scheduling scheme generation module in the cloud analyzes and derives the optimal collaborative scheduling scheme based on resource scheduling packages.

[0037] The module for generating and preparing to execute a precise feasible trajectory is used by the vehicle to perform constrained trajectory replanning and verification based on the travel time window and suggested reference trajectory allocated to the vehicle in the optimal cooperative scheduling scheme, combined with the latest local environment information, and to generate and prepare to execute a precise feasible trajectory.

[0038] The collaborative consistency index generation module, based on the vehicle's precise feasible trajectory, coordinates with the cloud and the vehicle to monitor the execution status and analyze and derive collaborative consistency indices.

[0039] The dynamic optimization module is used to analyze the collaborative consistency index and trigger the dynamic optimization process.

[0040] The beneficial effects of this application are as follows:

[0041] 1. The vehicle-road-cloud integrated multi-vehicle cooperative simulation decision-making method and system provided in this application adopts a fast filtering algorithm based on intent spectrum and conflict cost matrix to compress the combinatorial explosion problem into the feasible solution space, significantly improving the cloud-based solution efficiency; it introduces a cooperative weight bidding mechanism to resolve conflicts while taking into account system benefits and vehicle preferences, ensuring the fairness and optimality of the scheduling scheme; the vehicle side performs local replanning and safety verification based on cloud suggestions, which both follows the cooperative intent and maintains local real-time response capability; the system cooperative status is monitored through consistency index, and when a deviation is triggered, only the deviating vehicle needs to feed back a new intent spectrum, and the cloud quickly runs to simplify scheduling for fine-tuning, avoiding large-scale global replanning, realizing low-latency online dynamic adjustment of cooperative strategies, and significantly improving the long-term stable operation capability of the system in complex traffic scenarios.

[0042] 2. This application introduces a fast collaborative filtering algorithm based on a conflict cost matrix. Utilizing constraint propagation and pruning techniques, it enumerates and filters all feasible intent combinations with low conflict costs from the possibilities of combinatorial explosion. This solves the combinatorial explosion problem, enabling the elimination of obviously high-conflict combinations in a very short time, ensuring millisecond-level response of cloud computing and meeting real-time requirements. Instead of directly providing a unique solution, it outputs a "set of feasible intent combinations," preserving redundancy and providing a broader decision space for subsequent optimization and scheduling. Through a predefined or learned conflict cost matrix, it achieves a quantitative evaluation of multi-vehicle intent conflicts in complex interaction scenarios, which is more global than traditional pairwise collision detection.

[0043] 3. This application designs a collaborative weight bidding mechanism; it regards candidate scheduling schemes as bidding targets, and selects the optimal collaborative scheduling scheme that comprehensively considers individual utility and system benefits through a three-round mechanism of "ranking-conflict detection-optimization"; it achieves a balance between fairness and efficiency: through the design of comprehensive collaborative weights, it prevents "aggressive" vehicles from always occupying resources (considering individual utility), and avoids a significant decrease in overall traffic efficiency due to favoring weaker vehicles (considering system benefits); it ensures that the finally selected scheme is globally conflict-free and has the best comprehensive benefits within the given candidate set.

[0044] 4. This application adopts a hierarchical planning model of "cloud-guided + local constraint". The vehicle-side uses the cloud-based passage time window and suggested trajectory as hard constraints, combined with the latest local situation map, to perform constrained model predictive control replanning. A high-frequency reporting and cloud-based consistency monitoring mechanism is established. The cloud calculates system-level collaborative consistency indicators by comparing the actual and expected states of the vehicles, elevating the monitoring granularity from "individual vehicles" to the "system collaboration level". The vehicle-side undertakes the final trajectory tracking control, avoiding the communication delay uncertainty caused by "cloud remote operation" and ensuring the real-time performance and stability of control. Early risk detection: before small deviations accumulate into serious conflicts, the cloud can detect potential collaborative failure risks in advance through abnormal indicators. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a flowchart illustrating the implementation steps of the method described in this application.

[0047] Figure 2 This is a schematic diagram of the system structure connection of this application. Detailed Implementation

[0048] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0049] Please see Figure 1As shown, this application provides a vehicle-road-cloud integrated multi-vehicle cooperative simulation decision-making method in the first aspect, including: S1, the simulated vehicle generates a local fusion situation map based on local sensor and V2X communication data, and constitutes the local candidate intent spectrum of the simulated vehicle.

[0050] In a specific example, the simulated vehicle generates a local fusion situation map based on local sensor and V2X communication data, and constitutes the local candidate intent spectrum of the simulated vehicle, including: S201, performing a data fusion operation based on the multi-source sensor data and V2X communication data to generate a local fusion situation map for describing the vehicle's surrounding environment and dynamic target state.

[0051] S202, based on the local fusion situation map, perform a multi-strategy trajectory generation operation on the simulated vehicle to obtain a candidate trajectory set containing at least two candidate trajectories and their corresponding driving intention labels.

[0052] S203, based on a preset cost evaluation function, perform an initial preference weight calculation operation on each candidate trajectory in the candidate trajectory set to obtain the initial preference weight of each candidate trajectory.

[0053] S204. Based on the candidate trajectory set, the corresponding driving intention labels and the initial preference weights, perform an intention spectrum structured encapsulation operation to generate and upload the local candidate intention spectrum of the simulation vehicle.

[0054] It should be noted that data fusion refers to the spatiotemporal alignment and information complementarity integration of local environmental perception data acquired from the simulated vehicle's own sensors (such as LiDAR, cameras, and millimeter-wave radar) with dynamic information (such as position, speed, and heading) received through vehicle-to-the-world (V2X) communication from other vehicles or roadside facilities. This operation includes, but is not limited to, coordinate system unification, timestamp synchronization, target association, and state estimation, aiming to eliminate the uncertainty of a single data source and build a more comprehensive and reliable model of the vehicle's surrounding environment. Furthermore, multi-source sensor data and V2X communication data need to be calibrated and preprocessed before fusion to address issues such as data heterogeneity, inconsistent sampling rates, and communication latency, ensuring the basic accuracy of the fused data.

[0055] It should be noted that the local fusion situation map is a multi-dimensional data structure that represents the dynamic environment in which the vehicle is located at a specific point in time (or within a short time domain). It includes not only static road geometry information (lane lines, traffic signs, etc.) but also the real-time positions, speeds, accelerations, and predicted short-term motion trends of dynamic traffic participants (other vehicles, pedestrians, etc.). Its purpose is to provide an accurate and consistent input environment model for subsequent trajectory planning, serving as the foundation for the vehicle's autonomous and collaborative decision-making.

[0056] It should be noted that multi-strategy trajectory generation refers to the parallel instantiation of multiple planning strategies with different driving styles or cooperative tendencies under the same local fusion situational map input. These strategies can be pre-defined as aggressive, conservative, cooperative, etc. Each strategy operates independently, calculating one or more future short-term (e.g., future seconds) vehicle movement paths (i.e., candidate trajectories) starting from the vehicle's current position based on the same environmental information but different decision logics (such as different safety margins, traffic efficiency weights, and yielding rules). Each candidate trajectory is assigned a driving intent label representing its generation strategy during generation.

[0057] Furthermore, the parallel execution of multiple driving strategies is achieved by simultaneously activating multiple planning algorithm threads or logic units in the vehicle planning module. These threads or units share the input situation map but have independent cost functions and sets of constraints, thereby efficiently exploring the solution space.

[0058] It should be noted that the candidate trajectory set is a dataset containing K (K>=2) candidate trajectories and their respective driving intention labels. Each candidate trajectory physically represents a possible driving path of the vehicle over a future period of time (defined by a series of spatiotemporal point sequences), while the driving intention label is a discrete or low-dimensional continuous variable used to qualitatively or quantitatively describe the driving behavior tendencies reflected by the trajectory (such as "rapid overtaking," "smooth following," "active yielding," etc.). The purpose of this set is to cover multiple reasonable driving choices for the vehicle in a given environment, providing rich and semantically clear candidate solutions for subsequent cloud-based collaborative decision-making.

[0059] It should be noted that the initial preference weight calculation operation is based on a preset cost evaluation function, which quantifies and scores each candidate trajectory in the candidate trajectory set. The preset cost evaluation function is a mathematical function that integrates multiple indicators, and its expression is:

[0060] in, Indicates the first The initial preference weights of the candidate trajectories, This is represented by the number corresponding to the candidate trajectory. , This represents the total number of candidate trajectories; Represented as the first Candidate trajectories, Represented as the first The safety cost of each candidate trajectory, such as the negative mapping of indicators like minimum distance to obstacles and reciprocal collision time; Represented as the first The comfort cost of each candidate trajectory, such as acceleration and integral of jerk; Represented as the first The efficiency cost of candidate trajectories, such as estimated travel time or deviation from the desired speed. Function It is a function that performs a weighted summation or nonlinear fusion of these individual costs. The purpose of this operation is to assign a preliminary merit score to each candidate trajectory at the local level, based on the perspective of the individual vehicle.

[0061] Furthermore, the initial preference weight is explained as follows: The initial preference weight is a non-negative scalar value that represents the degree to which the candidate trajectory is recommended when considering only traditional driving indicators such as safety, comfort, and efficiency of the individual vehicle. A higher weight indicates that, from the perspective of the individual vehicle, the trajectory has better overall performance. Its purpose is to provide a reference benchmark reflecting the local preferences of individual vehicles during subsequent multi-vehicle collaborative game in the cloud, helping to balance the interests of individuals and the group during the collaborative process.

[0062] It should be noted that the intent spectrum structured encapsulation operation refers to encapsulating a set of candidate trajectories describing the same simulated vehicle. The corresponding set of driving intention tags and the calculated initial preference weight set The data is packaged according to a predefined protocol format to form the vehicle's local candidate intent spectrum. This spectrum is a complete data packet containing all possible driving options for the vehicle and their local evaluations. Subsequently, this data packet is uploaded to the cloud server via the vehicle-to-cloud communication link. Furthermore, structured encapsulation typically employs lightweight data serialization formats such as JSON and Protobuf to ensure transmission efficiency and cross-platform compatibility.

[0063] It's important to note that the local candidate intent spectrum is a core data structure. It represents a set of feasible action plans, each with clear semantics (intent labels) and preliminary evaluations (initial weights), that vehicles "report" to the cloud-based collaborative decision-making layer. It not only provides spatiotemporal information about the trajectory but, more importantly, assigns a behavioral interpretation to each trajectory through intent labels. Its role is to provide appropriately granular and information-rich decision input for cloud-based multi-vehicle intent coordination and conflict resolution, making it crucial for advancing from "trajectory coordination" to "intent coordination." Uploading to the cloud is for aggregating the intent spectra of all traffic participants, enabling collaborative optimization at a global level.

[0064] S2. Based on the local candidate intent spectrum of each vehicle, analyze and derive a set of feasible intent combinations.

[0065] In a specific instance, the step of analyzing and obtaining a set of feasible intent combinations based on the local candidate intent spectrum of each vehicle includes: S301, performing intent tag extraction operation based on the local candidate intent spectrum of each vehicle to obtain a global intent tag set containing each candidate driving intent of each vehicle.

[0066] S302, Based on the global intent tag set and intent conflict cost matrix, run a fast collaborative filtering algorithm to obtain the global intent combination space.

[0067] S303, based on the intent conflict cost matrix and the global intent combination space, perform a total conflict cost enumeration calculation operation to obtain the total conflict cost corresponding to each possible global intent combination.

[0068] S304, based on a preset conflict cost threshold, perform a cost filtering operation on each total conflict cost to select a set of feasible intent combinations.

[0069] It should be noted that the intent label extraction operation refers to the cloud server parsing and extracting the driving intent labels contained in the local candidate intent spectrum data structure of each simulated vehicle. This operation traverses the intent spectrum of each vehicle, collects the discrete labels, and forms a set representing all possible behavioral patterns of each traffic participant in the current simulation scenario.

[0070] It should be noted that the global intent label set is a data structure whose elements are the union of deduplicated driving intent labels from each vehicle. It defines the complete set of semantically meaningful behavioral patterns available to each vehicle in the system at the moment of collaborative decision-making, serving as the input basis for subsequent global intent combination enumeration. Its role is to provide explicit, semantically meaningful decision variables for constructing the multi-vehicle collaborative decision space.

[0071] It's important to note that the fast collaborative filtering algorithm is an optimization algorithm based on graph search and pruning. This algorithm takes a global intent label set and an intent conflict cost matrix as input. First, it constructs a search tree, where each level represents a vehicle's decision (choosing an intent label from its intent spectrum), and the branches represent different intent choices for that vehicle. Through constraint propagation, the algorithm uses the intent conflict cost matrix early in the search to predict the potential lower bound of conflicts for some paths (i.e., combinations where some vehicles have already determined their intents). If the lower bound exceeds a threshold T_c, the path is pruned, and its subsequent branches are not expanded. This operation aims to efficiently reduce the number of combinations that need to be fully computed.

[0072] Furthermore, constraint propagation refers to using the determined intentions of some vehicles, combined with the paired intention conflict information recorded in the conflict cost matrix, to deduce the minimum additional conflict cost that the remaining undetermined vehicles will inevitably cause when choosing certain intentions, thereby achieving an optimistic estimate of the total conflict cost of the current search path.

[0073] It should be noted that the global intent combination space is the set of all possible global intent combinations that remain after pruning by the fast collaborative filtering algorithm and need to be fully evaluated. Each combination is an N-tuple (N is the total number of vehicles), where the i-th element represents the specific driving intent label selected by the i-th vehicle. This space physically represents the set of global behavioral schemes that, theoretically, can achieve low-conflict collaboration after initial conflict filtering.

[0074] It should be noted that the cost filtering operation compares the calculated total conflict cost of each global intent combination with a preset conflict cost threshold. This operation iterates through all combinations with calculated total conflict costs, retaining only global intent combinations whose total conflict cost is less than the preset conflict cost threshold. The preset conflict cost threshold is a parameter that is either preset by the system or dynamically adjusted according to the scenario, representing the upper limit of the acceptable level of global collaborative conflict in the cloud system.

[0075] In a specific instance, the step of performing the total conflict cost enumeration calculation operation to obtain the total conflict cost corresponding to each possible combination of global intentions includes: calculating the total conflict cost using a formula. Determine the total cost of conflict ,in and This is represented as a vehicle index. This represents the total number of vehicles. Represented as the first The driving intent label selected by the vehicle in the current global intent combination. Represented as the first The driving intent label selected by the vehicle in the current global intent combination. Indicated as driving intent label and driving intention label The fundamental conflict between them represents value. This is represented as an optional weighting coefficient used to adjust the first... Vehicles and the first The importance of conflicts between vehicles.

[0076] It should be noted that the default value for the optional weighting coefficient is 1.

[0077] It should be noted that the total conflict cost is a non-negative scalar value used to quantify the severity of multi-vehicle collaborative conflicts that may result from a specific combination of global intentions (i.e., each vehicle choosing its own driving intention). A lower cost indicates better compatibility between the vehicle intentions in that combination and lower potential risks in collaborative execution. Its role is to serve as a core quantitative indicator for screening low-conflict collaborative solutions.

[0078] This application introduces a fast collaborative filtering algorithm based on a conflict cost matrix. Utilizing constraint propagation and pruning techniques, it enumerates and filters all feasible intent combinations with low conflict costs from the possibilities of combinatorial explosion. This solves the combinatorial explosion problem, eliminating obviously high-conflict combinations in a very short time, ensuring millisecond-level response in cloud computing and meeting real-time requirements. Instead of directly providing a unique solution, it outputs a "set of feasible intent combinations," preserving redundancy and providing a broader decision space for subsequent optimization and scheduling. Through a predefined or learned conflict cost matrix, it achieves a quantitative evaluation of multi-vehicle intent conflicts in complex interaction scenarios, offering a more global perspective than traditional pairwise collision detection.

[0079] S3. Based on each feasible intention combination, analyze and obtain the precise spatiotemporal resource set occupied by each vehicle in the combination, and generate the corresponding resource scheduling package.

[0080] In a specific instance, the step of analyzing and obtaining the precise spatiotemporal resource set occupied by each vehicle in the combination based on each feasible intent combination, and generating the corresponding resource scheduling package, includes: S501, performing a vehicle-intent-trajectory matching operation for each feasible intent combination in the feasible intent combination set, so as to determine a representative candidate trajectory corresponding to its specified intent label for each vehicle in the feasible intent combination from the local candidate intent spectrum of the corresponding vehicle.

[0081] S502, perform a spatiotemporal occupancy analysis on the representative candidate trajectory to quantify the space occupied by the vehicle during its movement along the trajectory in a discretized spatiotemporal dimension, and obtain the precise set of spatiotemporal resources requested by the vehicle under the feasible intention combination.

[0082] S503, based on the precise spatiotemporal resource set of all vehicles under the same feasible intention combination, perform a resource request packaging operation to generate a resource scheduling package that uniquely corresponds to the feasible intention combination.

[0083] The package typically contains a unique identifier for the feasible intent combination, the IDs of each vehicle within the combination, and the precise spatiotemporal resource set corresponding to each vehicle.

[0084] It's important to note that the vehicle-intent-trajectory matching operation refers to the process where, for a specific feasible combination of intents received from the cloud (i.e., an N-tuple where N is the total number of vehicles, and each element represents a driving intent label selected by a vehicle), the cloud server queries and locates a specific candidate trajectory labeled with that intent label from the vehicle's previously uploaded local candidate intent spectrum data structure, based on the vehicle ID and the corresponding driving intent label. This operation establishes a unique correspondence between "vehicle - selected intent - specific trajectory," which is the foundation for subsequent precise resource analysis. This operation is typically performed efficiently through cloud-based index lookup or hash mapping.

[0085] It should be noted that the representative candidate trajectory refers to the trajectory uniquely determined in the vehicle-intent-trajectory matching operation and used for subsequent precise spatiotemporal resource analysis. Given a combination of feasible intents, it is the selected spatiotemporal path plan representing a vehicle executing its specified driving intent (such as "turn left to overtake" or "go straight through"). It is the ideal driving path of the vehicle under the selected driving intent over a future period, consisting of a series of spatiotemporal points (containing information such as position, attitude, and speed) arranged in chronological order.

[0086] It's important to note that spatiotemporal occupancy analysis is a process of precisely quantifying the spatiotemporal occupancy implicit in candidate trajectories. This operation discretizes the continuous future spatiotemporal domain. First, the future time period covered by the trajectory is divided into several discrete time slices. Then, for each time slice, the spatial region occupied by the vehicle's contour or envelope at that moment (which can be based on the vehicle's length, width, and other geometric models) is calculated. Finally, this spatial region is mapped onto a three-dimensional (two-dimensional space + one-dimensional time) discrete grid (spatiotemporal grid). The set of all basic units (voxels) occupied by the vehicle's spatial region in each time slice on this grid represents the precise set of spatiotemporal resources requested by the vehicle. This operation aims to transform the abstract concept of a "trajectory" into a quantifiable and conflict-detectable model of "spatiotemporal resource occupancy."

[0087] Furthermore, the mathematical expression for spatiotemporal occupancy analysis can be described as follows: for a given representative candidate trajectory, the precise set of spatiotemporal resources it occupies. It can be calculated in the following ways: ,in Represented as discrete time points, This represents the time period covered by the traversed trajectory. For the vehicle at the time point The geometric shape envelope (e.g., a rectangle) is calculated based on the trajectory. For a predefined three-dimensional spatiotemporal voxel grid, it discretizes space into a two-dimensional raster and time into multiple levels; function Its function is to map the geometric envelope to the corresponding level (corresponding time point) of the spatiotemporal grid and return the set of all spatial grid cells (i.e., spatiotemporal voxels) covered by the geometric envelope.

[0088] It should be noted that the precise spatiotemporal resource set is a data structure containing unique identifiers or coordinates of all spatiotemporal voxels that a vehicle will occupy within a future timeframe under a specific intent and trajectory. Each spatiotemporal voxel is the smallest indivisible unit in a spatiotemporal grid, representing the right to occupy a small spatial region within a specific time slice. It represents the vehicle's "reservation" or "request" for that portion of spatiotemporal resources. Its function is to transform the trajectory conflict detection problem into an intersection detection problem of spatiotemporal voxel sets, providing precise and computable input for subsequent conflict resolution.

[0089] It should be noted that the resource request packaging operation refers to the process where, for a given feasible intent combination, the cloud encapsulates the precise spatiotemporal resource sets of all vehicles within the combination according to a preset data structure, forming a complete resource scheduling package. This package typically contains a unique identifier for the feasible intent combination, the IDs of each vehicle within the combination, and the precise spatiotemporal resource sets corresponding to each vehicle. Its function is to bind a complete, global, low-conflict driving intent scheme together with its required precise spatiotemporal resource occupancy information, forming a "candidate scheduling scheme" data unit that can be directly evaluated and processed by subsequent scheduling algorithms.

[0090] It's important to note that the resource scheduling package is a data structure bound to a unique combination of feasible intents. This package not only records "how each vehicle intends to drive" (intent combination), but also precisely records "which spatiotemporal resources each vehicle needs to occupy to achieve this driving method" (resource set). It is the key data interface connecting upper-layer intent coordination and lower-layer resource scheduling, enabling the cloud to perform final, unambiguous conflict resolution and optimization of all feasible intent combinations on the unified and precise dimension of "spatiotemporal resources."

[0091] S4, based on the resource scheduling package, analyzes and derives the optimal collaborative scheduling scheme in the cloud.

[0092] In a specific instance, the step of analyzing and deriving the optimal collaborative scheduling scheme based on the resource scheduling package includes: S601, combining the resource scheduling package and the corresponding feasible intent as a candidate scheduling scheme.

[0093] S602, based on the candidate scheduling scheme, the cloud performs a comprehensive collaborative weight calculation operation on the vehicle utility and system benefits of each vehicle in the scheme to obtain a comprehensive collaborative weight that characterizes the overall collaborative merit of the candidate scheduling scheme.

[0094] S603, based on the comprehensive collaborative weight of all candidate scheduling schemes, the cloud performs the first round of weight sorting operation to generate a candidate scheme sorting list arranged in descending order of comprehensive collaborative weight.

[0095] S604, a cloud-based conflict resolution algorithm based on collaborative weighted bidding: the first round sorts by weight; the second round performs conflict detection and constructs a set of non-conflicting preferred solutions; the third round selects the one with the highest weight from the preferred solution set as the optimal collaborative scheduling solution; the optimal collaborative scheduling solution determines the passage time window and suggested reference trajectory for each vehicle.

[0096] It should be noted that the comprehensive collaborative weight calculation operation refers to the cloud calculating a scalar value for each candidate scheduling scheme (i.e., a resource scheduling package and its bound feasible intent combination) to comprehensively quantify the overall collaborative benefits of the scheme. This operation first extracts the initial preference weights representing candidate trajectories under the selected intent of each vehicle in the scheme from the local candidate intent spectrum of each vehicle constituting the scheme. The sum of these initial preference weights constitutes the "total vehicle utility" of the scheme, reflecting the degree to which the scheme meets the needs of individual vehicles. Simultaneously, the cloud evaluates the potential global benefits of implementing the scheme (such as improved overall traffic efficiency, increased average speed, and reduced overall conflict risk), quantifying them into a "system benefit" value. The comprehensive collaborative weight is obtained by weighted fusion of the total vehicle utility and the system benefit, with the specific calculation formula as follows: ,in Indicates the overall collaborative weight. This represents the sum of the initial preference weights for all vehicles under this scheme. This represents the calculated system benefit value. It is in Adjustable weighting coefficients within the interval are used to balance the importance of individual utility and overall benefit. This operation places different candidate solutions within a unified, comparable quantitative framework, providing a basis for subsequent bidding and ranking.

[0097] Furthermore, the total vehicle utility is a dimensionless scalar value that summarizes the degree of preference each vehicle has for the assigned intention and trajectory from its own perspective (based on local costs such as safety, comfort, and efficiency) under the current candidate scheduling scheme.

[0098] Furthermore, system benefit is a dimensionless scalar value used to measure the degree to which a candidate scheduling scheme improves the operational status of the entire traffic system (such as intersections and road segments). For example, the reduction in conflict risk can be approximated by calculating the estimated reduction in total system travel time or total delay after the scheme is implemented, or by taking the reciprocal of the total conflict cost of the scheme. Its role is to evaluate the merits of the scheme at the macro-system level.

[0099] It should be noted that: 1. The first round is sorted by weight, and the specific process is as follows: Based on the candidate scheme ranking list, the cloud starts from the candidate scheduling scheme ranked first, and sequentially performs spatiotemporal voxel conflict detection on the precise spatiotemporal resource set of each vehicle in the candidate scheduling scheme to determine whether there is an internal resource conflict; 2. Conflict detection is performed in the second round to construct a set of mutually non-conflicting preferred schemes, and the specific process is as follows: Based on the results of the spatiotemporal voxel conflict detection operation, the cloud performs a second round of preferred scheme construction operation: adding candidate scheduling schemes without internal resource conflicts to a set of mutually non-conflicting "preferred schemes", and skipping all other candidate scheduling schemes that have resource conflicts with any scheme in the preferred scheme, so as to traverse the entire candidate scheme ranking list and complete the construction of the preferred scheme set; 3. The third round selects the one with the highest weight from the preferred scheme as the optimal collaborative scheduling scheme, and the specific process is as follows: Based on the preferred scheme set, the cloud performs a third round of optimal scheme selection operation: selecting the candidate scheduling scheme with the highest comprehensive collaborative weight from the preferred scheme set to determine it as the optimal collaborative scheduling scheme. This scheme determines the final passage time window and suggested reference trajectory for each simulated vehicle in the scenario.

[0100] It should be noted that the first round of weight sorting involves the cloud sorting all candidate scheduling schemes in descending order based on their calculated comprehensive collaborative weight values. The purpose of this operation is to prioritize schemes with theoretically higher comprehensive collaborative benefits, establishing processing priorities for subsequent conflict resolution, improving algorithm efficiency, and avoiding wasting time on low-quality schemes. Furthermore, the candidate scheme ranking list is a data structure where elements are candidate scheduling schemes, arranged strictly in descending order of their comprehensive collaborative weights; this list defines the order in which the cloud processes each scheme during subsequent conflict resolution algorithms.

[0101] It's important to note that the spatiotemporal voxel conflict detection operation is a crucial step in determining the feasibility of a candidate scheduling scheme. This operation, for a given candidate scheduling scheme, iterates through the precise spatiotemporal resource sets of all vehicles within the scheme. The detection logic is as follows: determine if there is an intersection between the precise spatiotemporal resource sets of all vehicles within the scheme; that is, whether at least one spatiotemporal voxel is simultaneously requested by two or more vehicles. If an intersection exists, it is determined that these two or more vehicles have a spatiotemporal resource conflict under this scheme, thus determining that the candidate scheduling scheme has internal resource conflicts and is therefore infeasible. This detection is implemented using efficient set intersection operations (e.g., using hash tables or bitmaps).

[0102] It's important to note that the second round of constructing the preferred set is a greedy conflict resolution process. The cloud starts from the top of the sorted list (the highest-weighted solution) and checks the current solution. If the solution passes the spatiotemporal voxel conflict detection (i.e., has no internal conflicts), it is added to the "preferred solution set." Furthermore, once a solution is added to the preferred set, the cloud quickly skips all subsequent candidate scheduling solutions in the sorted list that conflict with that solution based on its resource usage (without performing a full internal conflict check on the skipped solutions), because they are mutually exclusive with the preferred solution in terms of spatiotemporal resources and cannot be executed simultaneously. The cloud continues to check the next unskipped solution in the list, repeating this process until the entire sorted list has been traversed. The core of this operation is to construct a set of several non-conflicting candidate scheduling solutions (i.e., their spatiotemporal resource requests are pairwise non-overlapping).

[0103] It should be noted that the preferred scheme set is a data structure containing a collection of candidate scheduling schemes that are free from spatiotemporal resource conflicts, obtained through the second round of conflict resolution. All schemes in this set are feasible (without internal conflicts) and mutually exclusive (without overlapping resources), providing a high-quality, conflict-free candidate pool for the final decision.

[0104] It should be noted that the third round of optimal solution selection is the final decision-making step. In the pre-constructed set of preferred solutions, the cloud-based system directly selects the candidate scheduling scheme with the highest overall collaborative weight, determining it as the optimal output of the entire collaborative decision-making process—the optimal collaborative scheduling scheme. Since the schemes within the preferred solution set are guaranteed to be conflict-free, this selection only requires a simple comparison of the maximum values.

[0105] It should be noted that the optimal collaborative scheduling scheme is a data structure that includes the finalized global intent combination, representative candidate trajectories for each vehicle (as suggested reference trajectories), and a unique and non-overlapping passage time window allocated to each vehicle by these trajectories (the time window can be determined by analyzing the timestamps of vehicle trajectories in key conflict areas). This scheme will be distributed to each vehicle to guide its subsequent precise trajectory planning and collaborative execution.

[0106] This application designs a collaborative weighted bidding mechanism; it treats candidate scheduling schemes as bidding targets, and selects the optimal collaborative scheduling scheme that comprehensively considers individual utility and system benefits through a three-round mechanism of "ranking-conflict detection-optimization"; it achieves a balance between fairness and efficiency: through the design of comprehensive collaborative weights, it prevents "aggressive" vehicles from always occupying resources (considering individual utility), and avoids a significant decrease in overall traffic efficiency due to favoring weaker vehicles (considering system benefits); it ensures that the finally selected scheme is globally conflict-free and has the best comprehensive benefits within a given candidate set.

[0107] S5. Based on the travel time window and suggested reference trajectory allocated to the vehicle in the optimal collaborative scheduling scheme, and combined with the latest local environment information, the vehicle end performs constrained trajectory replanning and verification, and generates and prepares to execute an accurate and feasible trajectory.

[0108] In a specific instance, the process of constrained trajectory replanning and verification, generating and preparing to execute a precise feasible trajectory, includes: S701, based on the received cloud-deployed passage time window and suggested reference trajectory, and combined with the latest local fusion situation map, performing a constrained trajectory replanning operation to generate a precise feasible trajectory that satisfies vehicle dynamics constraints, control limits, and strictly conforms to the passage time window.

[0109] S702, Perform a forward-looking safety verification operation on the precise feasible trajectory, including collision detection with the latest perceived obstacle trajectory and verification of whether it meets the passage time window constraint.

[0110] It should be noted that constrained trajectory replanning refers to the process where, after receiving the collaborative decision-making results from the cloud (i.e., the travel time window and the suggested reference trajectory), the vehicle no longer performs unconstrained free planning. Instead, it uses the suggested reference trajectory from the cloud as a guiding reference, the travel time window as a hard constraint, and combines the latest local fusion situation map updated in real time by local sensors. The process then utilizes the Model Predictive Control (MPC) framework to solve the trajectory problem on local computing resources. The core of this operation is to transform the trajectory planning problem into a constrained optimization problem within a finite time domain, solving for a series of control inputs that ensure the planned trajectory (represented by the predicted state sequence) satisfies the vehicle's own physical limitations (such as maximum acceleration, steering angular velocity, etc.) while its passage time through key spatial locations (such as intersection entrances and merging points) strictly falls within the travel time window allocated by the cloud.

[0111] Furthermore, using the suggested reference trajectory as guidance means using the suggested reference trajectory issued from the cloud as a reference trajectory term in the cost function (or objective function) of the MPC optimization problem. This guides the locally planned trajectory to closely approximate the path suggested by the global collaborative solution, while satisfying hard constraints, in order to maintain the collaborative intent. Using the travel time window as a hard constraint means adding the arrival time of vehicles at specific spatial points (defined by the suggested reference trajectory or critical path points) as a state inequality constraint to the MPC optimization problem, forcing the solver to find a solution that meets the time requirement.

[0112] It should be noted that a precise feasible trajectory refers to a spatiotemporal path obtained by solving a constrained MPC optimization problem. It consists of a series of vehicle states (such as position, velocity, and heading angle) at discrete time points and corresponding control inputs (such as acceleration and front wheel steering angle). The trajectory is "precise" because its states and control variables are numerically executable; and "feasible" because it strictly satisfies three types of constraints: first, constraints from the vehicle's own dynamic equations (kinematics or dynamic model); second, constraints from physical control limits for actuators and safety (such as maximum acceleration and steering angle limits); and third, constraints from the collaborative time window issued by the cloud. Its role is to serve as the final driving plan that the vehicle will execute, conforming to both the local real-time environment and global collaborative agreements.

[0113] Furthermore, the constrained trajectory replanning operation is typically implemented by solving a model predictive control (MPC) optimization problem of the following form, mathematically expressed as:

[0114]

[0115]

[0116]

[0117]

[0118] in The optimization objective is to minimize the cost function, and the optimization variable (i.e. the quantity to be solved) is the control input sequence from the current time (k=0) to the moment before the end of the prediction time domain (k=N-1). The cost function is a scalar value used to comprehensively evaluate the deviation between the predicted trajectory and the desired target. The smaller the value, the closer the planned trajectory is to the reference trajectory, and the smoother the control action. Represented as the prediction time domain length, it is a positive integer representing the number of discrete time steps forward in the optimization problem, which determines the future time range considered in the planning. Represented as a time step index, it is an integer with a value ranging from 0 (current time) to N-1 (terminal time), used to identify each discrete time point in the prediction time domain; Represented as at the prediction time step The vehicle's predicted state vector at the predicted time step is a multi-dimensional column vector that typically contains physical quantities such as the vehicle's position coordinates, heading angle, and longitudinal velocity in a two-dimensional plane. Its significance lies in representing the vehicle's state at the predicted time step. The expected state of motion at that time; Represented as at the prediction time step The reference state vector at time has the same dimension as Similarly, the suggested reference trajectory directly derived from the cloud is based on the status points at the corresponding time, providing ideal path and status guidance for local planning; This represents the square of the weighted Euclidean norm; specifically, , , ,in , and This is the weight matrix; This is expressed as (usually a positive semi-definite matrix) the penalized state tracking error. This is represented as (usually a positive definite matrix) the penalty for the deviation between the control input and the reference input, or simply to suppress excessive control input. This is represented as (usually a positive semi-definite matrix) the penalty for terminal state error. The diagonal elements of these matrices correspond to the weights of each component in the state vector or control vector, used to adjust the degree of emphasis placed on different objectives during the optimization process; Represented as at the prediction time step The predictive control input vector at time; typically includes the vehicle's longitudinal acceleration and front wheel steering angle, etc., and is the actuator command required to make the vehicle reach the predicted state; Represented as at the prediction time step The reference control input vector at that time can usually be set as a zero vector or derived from the dynamics of the reference trajectory, and is used to guide the smoothing of the control input; Represented as constraints in the vehicle dynamics model; It is a discrete-time state transition function that describes the motion of a vehicle; the equation constraint requires that the predicted state sequence must conform to the physical motion of the vehicle. , , , , which are the upper and lower limit constraint vectors of the state variables and the upper and lower limit constraint vectors of the control variables, respectively. Ensure that the predicted vehicle status (such as speed and position) is within a safe and reasonable physical range; Ensure that the calculated control commands (such as acceleration and steering angle) do not exceed the physical limits (control limits) of the vehicle actuators; Used to calculate the time it takes for a vehicle to reach a specific key spatial point (e.g., the entrance to a conflict zone at an intersection) based on a predicted trajectory; and These represent the start and end times of the access time window issued by the cloud, respectively. This is represented as a travel time window constraint; it is a hard inequality constraint about time that forces the predicted arrival time of a vehicle at a specific key point to fall exactly within the time window specified in the cloud. It is the core constraint for ensuring the consistency of multi-vehicle collaboration.

[0119] It's important to note that the forward-looking safety verification operation refers to a comprehensive safety check performed on the planned, precisely feasible trajectory before execution. This operation comprises two core components: first, based on the latest local perception data (i.e., the predicted trajectories of dynamic obstacles in the local fusion situation map), full-time collision detection is performed on the precisely feasible trajectory to determine whether the vehicle will spatially interfere with any obstacles while traveling along the trajectory; second, the trajectory is re-verified to ensure it strictly adheres to the passage time window issued by the cloud, ensuring that its temporal characteristics have not deviated due to numerical errors or constraint relaxation in the local planning. This operation serves as the final safety check before execution, ensuring that the trajectory to be executed is not only feasible at the planning level but also safe at the safety level, and consistent with cloud-based collaborative instructions. It is a crucial step in ensuring vehicle safety and system collaborative robustness.

[0120] Furthermore, collision detection typically involves calculating the occupied area (i.e., the vehicle envelope) at each moment as the vehicle moves along a precise feasible trajectory and then performing a geometric intersection assessment with the predicted occupied areas of all obstacles at the same moment. If there is no intersection throughout the entire trajectory time domain, the collision detection is passed.

[0121] It should be noted that the trajectory confirmed through the forward-looking safety verification operation is the final approved trajectory for execution. It inherits all the characteristics of the "precisely feasible trajectory" and undergoes additional dual verification for safety and collaborative consistency, ensuring the reliability and safety of the vehicle when executing cloud-based collaborative commands in complex dynamic environments. If verification fails, the vehicle must trigger replanning or report the anomaly to the cloud, thereby activating the aforementioned dynamic feedback and online iterative optimization mechanism.

[0122] S6. Based on the vehicle's precise and feasible trajectory, the cloud and the vehicle work together to monitor the execution status and analyze the results to obtain a collaborative consistency index.

[0123] In a specific example, the cloud and vehicle work together to monitor the execution status based on the vehicle's precise feasible trajectory and analyze and obtain the collaborative consistency index, including: S801, the vehicle performs collaborative trajectory execution operation based on the verified precise feasible trajectory and simultaneously performs high-frequency actual status reporting operation to obtain the actual trajectory status of the vehicle.

[0124] S802, the cloud performs state comparison and index calculation operations based on the actual trajectory status and the optimal collaborative scheduling scheme to obtain a collaborative consistency index for evaluating the collaborative effect of multiple vehicles.

[0125] It should be noted that collaborative trajectory execution refers to the process by which each simulated vehicle strictly follows a precise and feasible trajectory that has been verified for safety and has been prospectively proven. This trajectory is the final driving plan that has been planned in real time and verified for safety, under the constraints of the time window and suggested trajectory issued by the cloud, combined with the latest local environment. Executing this operation means that the vehicle's control system (such as drive-by-wire, steering, and braking systems) will output a series of control commands (such as acceleration and front wheel angle) defined by the trajectory sequence, driving the vehicle to travel along the predetermined path in the simulation or real vehicle environment. Its function is to realize the traffic arrangement formulated by global collaborative decision-making (optimal collaborative scheduling scheme), and it is a key link in the process of multi-vehicle collaboration from "planning" to "implementation".

[0126] It should be noted that high-frequency real-state reporting refers to vehicles executing cooperative trajectories periodically sending their current actual motion status to the cloud monitor via the vehicle-to-cloud (V2X) communication link at a frequency far lower than the trajectory planning and control cycle (e.g., every 50-100 milliseconds). The data packets reported in this operation typically include, but are not limited to, vehicle ID, timestamp, real-time position in the global or local coordinate system, real-time speed, real-time heading angle, and possible control feedback. Its purpose is to provide the cloud with a real and continuous feedback data stream of the execution status of each vehicle's cooperative scheme, providing data input for consistent monitoring in the cloud.

[0127] It should be noted that the actual trajectory state is a data sequence or data stream that represents a series of spatiotemporal motion state points of the vehicle during actual execution. Physically, it is the result of fusing measurements from vehicle sensors (such as GNSS, IMU, and wheel speedometers) with the output of the vehicle state estimator, reflecting the vehicle's true position, attitude, and kinematic information at a specific moment. Its function is to compare it with the "ideal" state expected in the cloud-based collaborative solution, thereby exposing deviations in the execution process and serving as a direct basis for evaluating the consistency between "planning" and "execution."

[0128] It should be noted that the state comparison and index calculation operations are performed by the cloud monitor. This operation first receives the actual trajectory states reported by each vehicle. Then, the cloud extracts the "expected state" sequence planned for each vehicle from the stored optimal collaborative scheduling scheme. This expected state sequence originates from the suggested reference trajectory in the scheme or the expected state at key time points derived from it (such as the expected arrival time and location at the entrance of the intersection conflict zone). The core of the comparison operation is to compare the actual state of the same vehicle at the same time (or the same spatial location) with its expected state one by one, calculating the deviation between the two in key dimensions.

[0129] Furthermore, based on the state deviation data of each vehicle at each time point, a system-level collaborative consistency index is calculated in the cloud. This index is a comprehensive scalar value used to quantitatively evaluate the overall fidelity or consistency level of the entire multi-vehicle system in executing the optimal collaborative scheduling scheme. The specific calculation process is as follows: ,in This is represented as a coordination consistency index, a non-negative scalar value. The smaller the value, the closer the actual execution state of all vehicles is to the expected state of the global coordination plan, and the higher the system coordination consistency; the larger the value, the greater the overall execution deviation and the worse the coordination effect. Represented as the total number of vehicles participating in the coordination, it is a positive integer representing the number of traffic participants being monitored; This represents the vehicle's corresponding number. , which is an integer used to identify different vehicles. It represents the total number of discrete state points to be evaluated (or the number of evaluation time slices), and is a positive integer representing the number of sampling points for state comparison within the evaluation time window; This is represented by the integer number corresponding to the discrete state point being evaluated. , used to identify key points at different evaluation moments or on the trajectory. Represented as the first The car in An evaluation discrete state point state deviation vector (the degree of deviation between the vehicle's performance and the collaborative plan at that point), which may contain one or more components such as position deviation (e.g., Euclidean distance), speed deviation, and time deviation (the deviation between the actual transit time and the expected time window). Represented as a deviation metric function, it is a function that maps a state deviation vector to a non-negative scalar deviation value. For example, it can be the norm of the deviation vector (such as the L2 norm), or a weighted sum of squares of the deviations of each component. Its function is to synthesize multidimensional deviations into a single, summable deviation value. Represented as the first The car in The weighting coefficient for each discrete state point is a non-negative scalar used to adjust the importance of deviations in the overall assessment for different vehicles and different key points (e.g., points in conflict areas are more important than points in non-conflict areas). It can be set to 1 by default.

[0130] It should be noted that the system-level coordination consistency index is a core evaluation metric (characterizing the overall fidelity or robustness of the multi-vehicle coordinated scheduling scheme in the actual execution phase). This index has a dual function: on the one hand, as the output of online monitoring, it provides system operators or upper-level applications with a visualized and quantitative evaluation of the real-time coordination effect; on the other hand, when this index is lower than a certain preset threshold (indicating poor coordination consistency) or when the state deviation of a single vehicle is continuously too large, it will serve as a key criterion for triggering S7 (Dynamic Feedback and Online Iterative Optimization), thereby driving the system to make adaptive adjustments to maintain or restore efficient coordinated operation.

[0131] This application adopts a hierarchical planning model of "cloud-guided + local constraint". The vehicle-side uses the cloud-based passage time window and suggested trajectory as hard constraints, combined with the latest local situation map, to perform constrained model predictive control replanning. A high-frequency reporting and cloud-based consistency monitoring mechanism is established. The cloud calculates system-level collaborative consistency indicators by comparing the actual and expected states of the vehicles, elevating the monitoring granularity from "individual vehicles" to the "system collaboration level". The vehicle-side undertakes the final trajectory tracking control, avoiding the communication delay uncertainty caused by "cloud remote operation" and ensuring the real-time performance and stability of control. Early risk detection: before small deviations accumulate into serious conflicts, the cloud can detect potential collaborative failure risks in advance through abnormal indicators.

[0132] S7. Analyze the collaborative consistency indicators and trigger the dynamic optimization process.

[0133] In a specific example, the analysis of the coordination consistency index and the triggering of the dynamic optimization process includes: S901, based on the system-level coordination consistency index and the actual trajectory status of a single vehicle, performing a dynamic optimization triggering judgment operation to identify vehicles whose actual trajectory status is less than a preset first threshold and whose expected state deviation of the optimal coordination scheduling scheme is greater than a preset second threshold.

[0134] S902, based on the latest local fusion situation map of the deviation vehicle, perform a new local candidate intent spectrum generation operation to reflect the current situation, so as to generate and upload the new local candidate intent spectrum.

[0135] S903, the cloud runs a fast scheduling fine-tuning algorithm based on the new local candidate intent spectrum and the original optimal collaborative scheduling scheme to generate a new collaborative scheduling scheme after fine-tuning.

[0136] S904, the fine-tuned new collaborative scheduling scheme is sent to the relevant vehicles to achieve online dynamic adjustment.

[0137] It should be noted that dynamic optimization trigger judgment refers to the real-time evaluation of the system's operating status by the cloud monitor or the vehicle's local decision-making unit. This operation is based on the coordination consistency index and the deviation (such as position deviation and time deviation) between the actual trajectory status reported by each vehicle and the corresponding expected status in the optimal coordination scheduling scheme. When the coordination consistency index is less than a preset first threshold, it indicates that the collaborative execution effect of the entire multi-vehicle system no longer meets the minimum requirements and system-level optimization is needed. When the state deviation of a vehicle (e.g., the deviation between its time to pass through a key point and the allocated time window, or the deviation between its position and the suggested reference trajectory) continuously exceeds a preset second threshold, the vehicle is determined to be a "deviation vehicle," whose behavior has significantly deviated from the coordination plan and may disrupt global coordination, requiring rapid local adjustments for that vehicle. The purpose of this operation is to accurately identify the scenarios and target vehicles that require online optimization.

[0138] It should be noted that the new local candidate intent spectrum generation operation, reflecting the current situation, refers to the simulation vehicle identified as a deviation vehicle re-executing S1 with its latest local fusion situation map as input. Specifically, the vehicle re-plans its multi-strategy trajectory based on the latest environmental perception data (which may include dynamic changes such as the actual position and speed of other vehicles), generating a set of candidate trajectories that reflect the "current" environment and the vehicle's "current" state (not the state at the initial planning time). The initial preference weights of each trajectory are then recalculated, and finally, this is encapsulated into a new local candidate intent spectrum. This new intent spectrum reflects the set of feasible driving options that the vehicle can take based on the current situation after deviating from the cooperative plan.

[0139] Furthermore, the new local candidate intent spectrum is identical in data content structure to the original local candidate intent spectrum initially uploaded by the vehicle. However, the input it generates (i.e., the local fusion situation map) is based on the latest, real-time environmental and status information. Therefore, the candidate trajectories it contains and their evaluations are more in line with the real situation currently faced by the vehicle, providing accurate and timely decision input for rapid adjustments in the cloud.

[0140] It should be noted that the fast scheduling fine-tuning algorithm is a lightweight, low-latency scheduling optimization algorithm executed after the cloud receives the new local candidate intent spectrum from the deviating vehicles. This algorithm uses the original optimal cooperative scheduling scheme (i.e., the scheme determined in step S4) as a "baseline" scheme. The core operation of the algorithm is to replace the intentions and resource requests of the deviating vehicles in the original scheme (i.e., their representative candidate trajectories and corresponding precise spatiotemporal resource sets) with the intentions reselected based on the new local candidate intent spectrum and the newly calculated spatiotemporal resource requests; simultaneously, the intentions and resource requests of other non-deviating vehicles remain unchanged. Subsequently, the algorithm only needs to perform rapid conflict detection and resolution on the limited, local potential new conflicts arising from the replacement (mainly occurring between the new resource requests of the deviating vehicles and the original resource requests of other vehicles), rather than enumerating and calculating the global intention combinations of all vehicles as in the initial coordination. This strategy of performing local corrections on the "baseline" greatly reduces computational complexity and meets the real-time requirements of online dynamic adjustment.

[0141] Furthermore, the specific execution process of the fast scheduling fine-tuning algorithm can be described as follows: First, from the new local candidate intent spectrum of the deviating vehicle, a replacement intent and its corresponding candidate trajectory are selected according to preset rules (such as selecting the intent with the highest initial preference weight, or selecting the intent with the lowest conflict cost with the original intent); then, spatiotemporal occupancy analysis is performed on the replacement trajectory to obtain a new precise spatiotemporal resource set; then, the new resource set is compared with the resource sets of other vehicles in the original scheme, and fast conflict detection is performed; if a conflict is detected, other intents are tried to be selected from the new intent spectrum of the deviating vehicle, or the resource occupancy of other affected vehicles in the original scheme is adjusted within a very small range of spatiotemporal scheduling (such as fine-tuning the start or end time of their passage time window), until a conflict-free and feasible new intent and resource allocation combination is found; finally, this locally adjusted scheme is packaged into a fine-tuned new cooperative scheduling scheme.

[0142] It should be noted that the fine-tuned new collaborative scheduling scheme is a data structure. While retaining most of the content of the original optimal collaborative scheduling scheme, it adjusts the driving intentions, suggested reference trajectories, and passage time windows of deviating vehicles (and possibly a few other vehicles affected by them). This scheme aims to correct the collaborative inaccuracies caused by environmental changes or execution deviations, bringing the system back to a highly efficient collaborative state.

[0143] It should be noted that distributing the fine-tuned new collaborative scheduling scheme to relevant vehicles refers to the cloud sending a data packet containing the adjusted intent, trajectory, and time window information to the deviating vehicle and other relevant vehicles involved in the adjustment through the vehicle-cloud communication link. Upon receiving the new scheme, the vehicle will immediately use it as the basis to re-execute S5 (vehicle-side constrained trajectory replanning and verification) and subsequent steps. This closed-loop operation enables online, dynamic, and adaptive adjustment of the collaborative strategy, improving the system's robustness and continuous collaborative capability in dynamic and uncertain environments.

[0144] Please see Figure 2 As shown, in its second aspect, this application provides a system for a multi-vehicle collaborative simulation decision-making method integrating vehicle, road, and cloud.

[0145] The system 100 of the vehicle-road-cloud integrated multi-vehicle cooperative simulation decision-making method of the present invention can be installed in an electronic device. Depending on the functions implemented, the system 100 may include a local candidate intent spectrum generation module 101, a feasible intent combination set generation module 102, a precise spatiotemporal resource set and resource scheduling package generation module 103, an optimal cooperative scheduling scheme cloud generation module 104, a module for generating and preparing precise feasible trajectories for execution 105, a cooperative consistency index generation module 106, and a dynamic optimization module. The module described in this invention can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, stored in the memory of the electronic device.

[0146] In this embodiment, the functions of each module / unit are as follows: The local candidate intent spectrum generation module is used to generate a local fusion situation map based on local sensor and V2X communication data of the simulated vehicle, and to form the local candidate intent spectrum of the simulated vehicle.

[0147] The feasible intent combination set generation module analyzes and derives a feasible intent combination set based on the local candidate intent spectrum of each vehicle.

[0148] The precise spatiotemporal resource set and resource scheduling package generation module analyzes and derives the precise spatiotemporal resource set occupied by each vehicle in the combination based on each feasible intention combination, and generates the corresponding resource scheduling package.

[0149] The optimal collaborative scheduling scheme generation module in the cloud analyzes and derives the optimal collaborative scheduling scheme based on resource scheduling packages.

[0150] The module for generating and preparing to execute a precise feasible trajectory is used by the vehicle to perform constrained trajectory replanning and verification based on the travel time window and suggested reference trajectory allocated to the vehicle in the optimal cooperative scheduling scheme, combined with the latest local environment information, and to generate and prepare to execute a precise feasible trajectory.

[0151] The collaborative consistency index generation module, based on the vehicle's precise feasible trajectory, coordinates with the cloud and the vehicle to monitor the execution status and analyze it to derive collaborative consistency indices.

[0152] The dynamic optimization module is used to analyze the collaborative consistency index and trigger the dynamic optimization process.

[0153] The vehicle-road-cloud integrated multi-vehicle cooperative simulation decision-making method and system provided in this application adopts a fast filtering algorithm based on intent spectrum and conflict cost matrix to compress the combinatorial explosion problem into the feasible solution space, significantly improving the cloud-based solution efficiency. It introduces a cooperative weight bidding mechanism to resolve conflicts while considering system benefits and vehicle preferences, ensuring the fairness and optimality of the scheduling scheme. Vehicles perform local replanning and safety verification based on cloud-based suggestions, adhering to cooperative intent while maintaining local real-time response capabilities. The system's cooperative status is monitored through a consistency index; when a deviation is triggered, only the deviating vehicle needs to provide a new intent spectrum, allowing for rapid cloud-based optimization and simplified scheduling, avoiding large-scale global replanning. This achieves low-latency online dynamic adjustment of the cooperative strategy, significantly improving the system's long-term stable operation capability in complex traffic scenarios.

[0154] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0155] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0156] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0157] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0158] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0159] 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 technical solutions of the present invention.

Claims

1. A multi-vehicle collaborative simulation decision-making method integrating vehicle, road, and cloud, characterized in that: include: S1. The simulated vehicle generates a local fusion situation map based on local sensor and V2X communication data, and constitutes the local candidate intent spectrum of the simulated vehicle. S2. Based on the local candidate intent spectrum of each vehicle, analyze and derive a set of feasible intent combinations; S3. Based on each feasible intention combination, analyze and obtain the precise spatiotemporal resource set occupied by each vehicle in the combination, and generate the corresponding resource scheduling package. S4. Based on resource scheduling packages, the cloud analyzes and derives the optimal collaborative scheduling scheme. S5. Based on the travel time window and suggested reference trajectory allocated to the vehicle in the optimal collaborative scheduling scheme, and combined with the latest local environment information, the vehicle end performs constrained trajectory replanning and verification, and generates and prepares to execute an accurate and feasible trajectory. S6. Based on the vehicle's precise and feasible trajectory, the cloud and the vehicle work together to monitor the execution status and analyze and derive collaborative consistency indicators. S7. Analyze the collaborative consistency indicators and trigger the dynamic optimization process.

2. The multi-vehicle collaborative simulation decision-making method integrating vehicle, road, and cloud as described in claim 1, characterized in that, The simulated vehicle generates a local fusion situation map based on local sensor and V2X communication data, and constitutes the simulated vehicle's local candidate intent spectrum, including: S201, Based on the multi-source sensor data and V2X communication data, perform data fusion operation to generate a local fusion situation map describing the vehicle's surrounding environment and dynamic target status; S202, Based on the local fusion situation map, perform a multi-strategy trajectory generation operation on the simulated vehicle to obtain a candidate trajectory set containing at least two candidate trajectories and their corresponding driving intention labels; S203, based on a preset cost evaluation function, perform an initial preference weight calculation operation on each candidate trajectory in the candidate trajectory set to obtain the initial preference weight of each candidate trajectory; S204. Based on the candidate trajectory set, the corresponding driving intention labels and the initial preference weights, perform an intention spectrum structured encapsulation operation to generate and upload the local candidate intention spectrum of the simulation vehicle.

3. The multi-vehicle collaborative simulation decision-making method integrating vehicle, road, and cloud as described in claim 1, characterized in that, The set of feasible intent combinations, derived from the local candidate intent spectrum of each vehicle, includes: S301, based on the local candidate intent spectrum of each vehicle, perform intent tag extraction operation to obtain a global intent tag set containing each candidate driving intent of each vehicle; S302, Based on the global intent tag set and intent conflict cost matrix, run a fast collaborative filtering algorithm to obtain the global intent combination space; S303, based on the intent conflict cost matrix and the global intent combination space, perform a total conflict cost enumeration calculation operation to obtain the total conflict cost corresponding to each possible global intent combination; S304, based on a preset conflict cost threshold, perform a cost filtering operation on each total conflict cost to select a set of feasible intent combinations.

4. The multi-vehicle collaborative simulation decision-making method integrating vehicle, road, and cloud as described in claim 3, is characterized in that, The step of performing a total conflict cost enumeration calculation to obtain the total conflict cost corresponding to each possible combination of global intentions includes: Through calculation formula Determine the total cost of conflict ,in and This is represented as a vehicle index. This represents the total number of vehicles. Represented as the first The driving intent label selected by the vehicle in the current global intent combination. Represented as the first The driving intent label selected by the vehicle in the current global intent combination. Indicated as driving intent label and driving intention label The fundamental conflict between them represents value. This is represented as an optional weighting coefficient used to adjust the first... Vehicles and the first The importance of conflicts between vehicles.

5. The multi-vehicle collaborative simulation decision-making method integrating vehicle, road, and cloud as described in claim 1, characterized in that, The process involves analyzing and determining the precise spatiotemporal resource set occupied by each vehicle's request within a combination of feasible intentions, and generating a corresponding resource scheduling package, including: S501, For each feasible intent combination in the feasible intent combination set, perform a vehicle-intent-trajectory matching operation to determine a representative candidate trajectory corresponding to its specified intent label for each vehicle in the feasible intent combination from the local candidate intent spectrum of the corresponding vehicle. S502, perform spatiotemporal occupancy analysis on the representative candidate trajectory to quantify the space occupied by the vehicle during its movement along the trajectory in a discretized spatiotemporal dimension, and obtain the precise set of spatiotemporal resources requested by the vehicle under the feasible intention combination. S503, based on the precise spatiotemporal resource set of all vehicles under the same feasible intention combination, perform a resource request packaging operation to generate a resource scheduling package that uniquely corresponds to the feasible intention combination.

6. The multi-vehicle collaborative simulation decision-making method integrating vehicle, road, and cloud as described in claim 1, characterized in that, The process of analyzing and deriving the optimal collaborative scheduling scheme based on resource scheduling packets includes: S601 combines the resource scheduling package and the corresponding feasible intent as a candidate scheduling scheme; S602, Based on the candidate scheduling scheme, the cloud performs a comprehensive collaborative weight calculation operation on the vehicle utility and system benefits of each vehicle in the scheme to obtain a comprehensive collaborative weight that characterizes the global collaborative advantages and disadvantages of the candidate scheduling scheme. S603, Based on the comprehensive collaborative weight of all candidate scheduling schemes, the cloud performs the first round of weight sorting operation to generate a candidate scheme sorting list arranged in descending order of comprehensive collaborative weight; S604, a cloud-based conflict resolution algorithm based on collaborative weighted bidding: the first round sorts by weight; the second round performs conflict detection and constructs a set of non-conflicting preferred solutions; the third round selects the one with the highest weight from the preferred solution set as the optimal collaborative scheduling solution; the optimal collaborative scheduling solution determines the passage time window and suggested reference trajectory for each vehicle.

7. The multi-vehicle collaborative simulation decision-making method integrating vehicle, road, and cloud as described in claim 1, characterized in that, The process of constrained trajectory replanning and verification, generating and preparing for execution of an accurate and feasible trajectory, includes: S701, based on the received passage time window and suggested reference trajectory sent from the cloud, and combined with the latest local fusion situation map, performs a constrained trajectory replanning operation to generate an accurate and feasible trajectory that meets vehicle dynamics constraints, control limits and strictly conforms to the passage time window; S702, Perform a forward-looking safety verification operation on the precise feasible trajectory, including collision detection with the latest perceived obstacle trajectory and verification of whether it meets the passage time window constraint.

8. The multi-vehicle collaborative simulation decision-making method integrating vehicle, road, and cloud as described in claim 1, characterized in that, The process involves monitoring the execution status of a vehicle based on a precise and feasible trajectory, with the cloud and vehicle working together to analyze and derive collaborative consistency indicators, including: S801, the vehicle side performs collaborative trajectory execution operations based on the verified accurate and feasible trajectory, and simultaneously performs high-frequency real-state reporting operations to obtain the actual trajectory status of the vehicle. S802, the cloud performs state comparison and index calculation operations based on the actual trajectory status and the optimal collaborative scheduling scheme to obtain a collaborative consistency index for evaluating the collaborative effect of multiple vehicles.

9. The multi-vehicle collaborative simulation decision-making method integrating vehicle, road, and cloud as described in claim 1, characterized in that, The analysis of the collaborative consistency index and the triggering of the dynamic optimization process include: S901 performs dynamic optimization trigger judgment operations based on system-level coordination consistency index and actual trajectory status of a single vehicle to identify vehicles with actual trajectory status where the coordination consistency index is less than a preset first threshold and the expected state deviation of the optimal coordination scheduling scheme is greater than a preset second threshold. S902, based on the latest local fusion situation map of the deviation vehicle, perform a new local candidate intent spectrum generation operation to reflect the current situation, so as to generate and upload the new local candidate intent spectrum; S903, the cloud runs a fast scheduling fine-tuning algorithm based on the new local candidate intent spectrum and the original optimal collaborative scheduling scheme to generate a new collaborative scheduling scheme after fine-tuning. S904, the fine-tuned new collaborative scheduling scheme is sent to the relevant vehicles to achieve online dynamic adjustment.

10. A system for implementing the multi-vehicle cooperative simulation decision-making method integrating vehicle, road, and cloud as described in any one of claims 1-9, characterized in that, include: The local candidate intent spectrum generation module is used to generate a local fusion situation map based on local sensor and V2X communication data of the simulated vehicle, and to form the local candidate intent spectrum of the simulated vehicle. The feasible intent combination set generation module analyzes and derives the feasible intent combination set based on the local candidate intent spectrum of each vehicle. The precise spatiotemporal resource set and resource scheduling package generation module analyzes and derives the precise spatiotemporal resource set occupied by each vehicle in the combination based on each feasible intention combination, and generates the corresponding resource scheduling package. The optimal collaborative scheduling scheme generation module in the cloud analyzes and derives the optimal collaborative scheduling scheme based on resource scheduling packages. The module for generating and preparing to execute a precise feasible trajectory is used by the vehicle to perform constrained trajectory replanning and verification based on the travel time window and suggested reference trajectory allocated to the vehicle in the optimal cooperative scheduling scheme, combined with the latest local environment information, and to generate and prepare to execute a precise feasible trajectory. The collaborative consistency index generation module, based on the vehicle's precise feasible trajectory, coordinates with the cloud and the vehicle to monitor the execution status and analyze and derive collaborative consistency indices. The dynamic optimization module is used to analyze the collaborative consistency index and trigger the dynamic optimization process.

Images