Multi-vehicle cooperative dynamic negotiation scheduling method and related device

Abstract

The application provides a multi-vehicle cooperative dynamic negotiation scheduling method and related device, the method comprises: obtaining the following behavior data of human-driven vehicles and lane-changing behavior data, cooperative request semantic data broadcast by networked vehicles through V2X, and cooperative vehicle knowledge graph; using an LLM model to infer the above data, extract the HDV driving intention probability distribution and ICV cooperative intention graph vector; determine the inherent right-of-way space of the target area accordingly; if the preset threshold is not met, carry out multi-vehicle cooperative game reasoning through LLM, generate a semantic decision set, and convert it into bottom layer control parameters to execute scheduling, which can accurately identify the complex interaction intention of HDV and ICV in a mixed traffic environment, dynamically build a reasonable right-of-way allocation model, and improve the vehicle scheduling efficiency and decision safety.

Description

Technical Field

[0001] This application relates to the field of data processing technology, specifically to a multi-vehicle collaborative dynamic negotiation scheduling method and related apparatus. Background Technology

[0002] With the rapid development of intelligent transportation and autonomous driving technologies, vehicle-to-everything (V2X) and artificial intelligence (AI) are gradually becoming key technological supports for promoting the intelligence of transportation systems. In actual road traffic scenarios, especially in areas such as complex urban intersections and multi-lane highways, there are many situations where human-driven vehicles (HDVs) and connected autonomous vehicles (ICVs) are mixed together.

[0003] However, most existing mainstream traffic cooperation methods rely on rule-based control models or finite state machine strategies, which lack the ability to understand complex interaction intentions and dynamic reasoning. Especially when facing highly dynamic interactions among multiple vehicles, they suffer from rigid decision-making and difficulty in adapting to dynamic changes. Summary of the Invention

[0004] This application provides a multi-vehicle collaborative dynamic negotiation scheduling method and related device, which can accurately identify the complex interaction intentions of HDV and ICV in mixed traffic environments, dynamically construct a reasonable right-of-way allocation model, and break through the limitations of traditional traffic cooperative systems in decision-making rigidity and difficulty in adapting to dynamic changes by using a multi-vehicle game negotiation method, thereby improving vehicle scheduling efficiency and decision-making safety.

[0005] A first aspect of this application provides a multi-vehicle cooperative dynamic negotiation scheduling method, the method comprising: Acquire longitudinal following behavior data, lateral lane-changing behavior data, semantic data of collaborative requests broadcast by connected vehicles via V2X, and a knowledge graph of collaborative vehicles; Based on the cooperative vehicle knowledge graph, the longitudinal following behavior data, lateral lane-changing behavior data, and cooperative request semantics are processed using an LLM model to obtain the HDV driving intention probability distribution and ICV cooperative intention vector. The inherent right-of-way space of the target area is determined based on the HDV driving intention probability distribution and the ICV cooperative intention vector. When the inherent right-of-way space of the target area does not meet the preset threshold, the LLM model is used to perform multi-vehicle cooperative negotiation processing on the human-driven vehicles and connected vehicles to obtain a multi-vehicle cooperative semantic decision set. Based on the multi-vehicle cooperative semantic decision set, determine the underlying vehicle control parameters; The underlying vehicle control parameters are executed to perform multi-vehicle collaborative scheduling of the human-driven vehicles and connected vehicles.

[0006] In one possible implementation, the step of processing the longitudinal following behavior data, lateral lane-changing behavior data, and cooperative request semantics using an LLM model based on the cooperative vehicle knowledge graph to obtain the HDV driving intention probability distribution and ICV cooperative intention vector includes: The longitudinal following behavior data and the lateral lane changing behavior data are converted into natural language descriptions to obtain traffic scene text fragment data. Based on the traffic scenario text fragment data and the collaborative vehicle knowledge graph, a rule matching engine is used to retrieve relevant traffic regulations and determine rule constraint prompts. Based on the traffic scene text fragment data, collaborative request semantics, and rule constraint prompts, LLM multimodal intent reasoning is performed to obtain the HDV driving intent probability distribution and ICV collaborative intent vector; The vehicle dynamics model is used to perform a physical feasibility check on the HDV driving intention probability distribution and ICV cooperative intention vector, and the HDV driving intention probability distribution and ICV cooperative intention vector that pass the physical feasibility check are output.

[0007] In one possible implementation, when the inherent right-of-way space of the target area does not meet a preset threshold, an LLM model is used to perform multi-vehicle cooperative negotiation processing on the human-driven vehicles and connected vehicles to obtain a multi-vehicle cooperative semantic decision set, including: Based on the inherent right-of-way space gap and the preset threshold, the amount of right-of-way to be released is calculated using a space demand quantification algorithm; Based on the amount of right-of-way to be released and the ICV cooperative intent vector, an LLM is used to generate a set of negotiable right-of-way parameters, which includes cooperative action type, space release amount and time window. Based on the negotiable right-of-way parameter set, a cooperation request is sent to the associated connected vehicles using the V2X communication protocol, and a cooperation commitment vector is received in response. Based on the collaborative commitment vector and the HDV driving intention probability distribution, LLM is used to perform multi-vehicle game equilibrium calculation to generate the multi-vehicle collaborative semantic decision set.

[0008] In one possible implementation, determining the vehicle's underlying control parameters based on the multi-vehicle cooperative semantic decision set includes: Based on the multi-vehicle collaborative semantic decision set, a semantic parsing engine is used to extract high-level control instructions, wherein the high-level control instructions include action type, action range and time constraints. Based on the high-level control instructions, the vehicle's kinematic trajectory is calculated using a bicycle model to determine the original set of control parameters; Based on the original control parameter set, a safety boundary verification algorithm is used to verify physical feasibility. When the verification passes, the underlying control parameters of the vehicle are output. When the verification fails, the LLM model is triggered to regenerate the multi-vehicle cooperative semantic decision set.

[0009] One possible implementation also includes: The steps of acquiring longitudinal following behavior data, lateral lane-changing behavior data of human-driven vehicles, semantic data of cooperative requests broadcast by connected vehicles via V2X, and cooperative vehicle knowledge graphs, and executing the underlying control parameters of the vehicles to perform multi-vehicle cooperative scheduling of human-driven vehicles and connected vehicles are evaluated to obtain the k-th scheduling evaluation result. If the information indicated by the k-th scheduling evaluation result is a collision, escalation of conflict, or other negative result, the k-th scheduling evaluation result is transformed into a high-dimensional vector result of the k-th scheduling evaluation. The high-dimensional vector result of the kth scheduling evaluation is entered into the memory bank; If the information indicated by the k-th scheduling evaluation result is safe, then this step ends.

[0010] This example proposes a multi-vehicle cooperative dynamic negotiation scheduling method. First, it acquires longitudinal following and lateral lane-changing behavior data of human-driven vehicles, as well as semantic data of cooperative requests broadcast by connected vehicles via V2X. Combined with a constructed cooperative vehicle knowledge graph, it uses an LLM model to perform multimodal semantic reasoning on traffic behavior data, accurately extracting the probability distribution of HDV driving intentions and ICV cooperative intention vectors. Based on right-of-way space calculation and intention game, it generates multi-vehicle cooperative semantic decisions, which are finally transformed into underlying control parameters to realize multi-vehicle scheduling and cooperative control. This method can accurately identify the complex interaction intentions of HDVs and ICVs in mixed traffic environments, dynamically construct a reasonable right-of-way allocation model, and overcome the limitations of traditional traffic cooperative systems, such as rigid decision-making and difficulty in adapting to dynamic changes, by using a multi-vehicle game negotiation approach, thereby improving vehicle scheduling efficiency and decision-making safety.

[0011] A second aspect of this application provides a multi-vehicle cooperative dynamic negotiation scheduling device, the device comprising: The acquisition unit is used to acquire longitudinal following behavior data of human-driven vehicles, lateral lane-changing behavior data, semantic data of collaborative requests broadcast by connected vehicles through V2X, and collaborative vehicle knowledge graph. The first processing unit is used to process the longitudinal following behavior data, lateral lane changing behavior data and collaborative request semantics based on the collaborative vehicle knowledge graph using an LLM model, to obtain the HDV driving intention probability distribution and ICV collaborative intention vector. The second processing unit is used to determine the inherent right-of-way space of the target area based on the HDV driving intention probability distribution and the ICV cooperative intention vector. The third processing unit is used to perform multi-vehicle cooperative negotiation processing on the human-driven vehicle and the connected vehicle using the LLM model when the inherent right-of-way space of the target area does not meet the preset threshold, so as to obtain a multi-vehicle cooperative semantic decision set. The fourth processing unit is used to determine the underlying control parameters of the vehicle based on the multi-vehicle cooperative semantic decision set. The scheduling unit is used to execute the underlying control parameters of the vehicle and to perform multi-vehicle collaborative scheduling of the human-driven vehicles and connected vehicles.

[0012] One possible implementation also includes: A memory unit is configured to convert the k-th scheduling evaluation result into a high-dimensional vector result of the k-th scheduling evaluation if the information indicated by the k-th scheduling evaluation result is a collision, escalation of conflict, or other negative result. The high-dimensional vector result of the kth scheduling evaluation is entered into the memory bank; If the information indicated by the k-th scheduling evaluation result is safe, then this step ends.

[0013] In one possible implementation, in the aspect of processing the longitudinal following behavior data, lateral lane-changing behavior data, and cooperative request semantics using an LLM model based on the cooperative vehicle knowledge graph to obtain the HDV driving intention probability distribution and ICV cooperative intention vector, the first processing unit is configured to: The longitudinal following behavior data and the lateral lane changing behavior data are converted into natural language descriptions to obtain traffic scene text fragment data. Based on the traffic scenario text fragment data and the collaborative vehicle knowledge graph, a rule matching engine is used to retrieve relevant traffic regulations and determine rule constraint prompts. Based on the traffic scene text fragment data, collaborative request semantics, and rule constraint prompts, LLM multimodal intent reasoning is performed to obtain the HDV driving intent probability distribution and ICV collaborative intent vector; The vehicle dynamics model is used to perform a physical feasibility check on the HDV driving intention probability distribution and ICV cooperative intention vector, and the HDV driving intention probability distribution and ICV cooperative intention vector that pass the physical feasibility check are output.

[0014] In one possible implementation, when the inherent right-of-way space of the target area does not meet a preset threshold, an LLM model is used to perform multi-vehicle cooperative negotiation processing on the human-driven vehicle and the connected vehicle to obtain aspects of the multi-vehicle cooperative semantic decision set. The third processing unit is used for: Based on the inherent right-of-way space gap and the preset threshold, the amount of right-of-way to be released is calculated using a space demand quantification algorithm; Based on the amount of right-of-way to be released and the ICV cooperative intent vector, an LLM is used to generate a set of negotiable right-of-way parameters, which includes cooperative action type, space release amount and time window. Based on the negotiable right-of-way parameter set, a cooperation request is sent to the associated connected vehicles using the V2X communication protocol, and a cooperation commitment vector is received in response. Based on the collaborative commitment vector and the HDV driving intention probability distribution, LLM is used to perform multi-vehicle game equilibrium calculation to generate the multi-vehicle collaborative semantic decision set.

[0015] In one possible implementation, in the aspect of determining the vehicle's underlying control parameters based on the multi-vehicle cooperative semantic decision set, the fourth processing unit is configured to: Based on the multi-vehicle collaborative semantic decision set, a semantic parsing engine is used to extract high-level control instructions, wherein the high-level control instructions include action type, action range and time constraints. Based on the high-level control instructions, the vehicle's kinematic trajectory is calculated using a bicycle model to determine the original set of control parameters; Based on the original control parameter set, a safety boundary verification algorithm is used to verify physical feasibility. When the verification passes, the underlying control parameters of the vehicle are output. When the verification fails, the LLM model is triggered to regenerate the multi-vehicle cooperative semantic decision set.

[0016] A third aspect of this application provides a terminal including a processor, an input device, an output device, and a memory, wherein the processor, input device, output device, and memory are interconnected, wherein the memory is used to store a computer program, the computer program including program instructions, and the processor is configured to invoke the program instructions to execute the steps of the multi-vehicle cooperative dynamic negotiation scheduling method as described in the first aspect of this application.

[0017] A fourth aspect of this application provides a computer-readable storage medium storing a computer program for electronic data interchange, wherein the computer program causes a computer to perform some or all of the steps described in the multi-vehicle cooperative dynamic negotiation scheduling method of the first aspect of this application.

[0018] A fifth aspect of this application provides a computer program product, comprising a non-transitory computer-readable storage medium storing a computer program operable to cause a computer to perform some or all of the steps described in the multi-vehicle cooperative dynamic negotiation scheduling method of the first aspect of this application. The computer program product may be a software installation package. Attached Figure Description

[0019] 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.

[0020] Figure 1 This application provides a schematic diagram of the overall process of a multi-vehicle cooperative dynamic negotiation scheduling method. Figure 2 This application provides a schematic diagram illustrating the process of determining the HDV driving intent probability distribution and ICV collaborative intent vector in a multi-vehicle cooperative dynamic negotiation scheduling method. Figure 3 This application provides a schematic diagram of the overall structure of a multi-vehicle cooperative dynamic negotiation scheduling device. Figure 4 This application provides a schematic diagram of the structure of a terminal. Figure label: Acquisition unit-1, first processing unit-2, second processing unit-3, third processing unit-4, fourth processing unit-5, scheduling unit-6, memory unit-7. Detailed Implementation

[0021] 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.

[0022] The terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0023] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application can be combined with other embodiments.

[0024] To better understand the multi-vehicle cooperative dynamic negotiation scheduling method provided in this application, a brief introduction to the scenarios in which this method is applied is given below. Current mainstream traffic cooperation methods largely rely on rule-based control models or finite state machine strategies, lacking the ability to understand complex interactive intentions and dynamic reasoning. Especially when facing highly dynamic multi-vehicle interactions, they suffer from rigid decision-making and difficulty adapting to dynamic changes. Meanwhile, some studies have introduced reinforcement learning or game theory methods, but these generally suffer from strong dependence on training samples and poor generalization ability, making it difficult to meet the high safety and robustness requirements of real-world traffic environments. Furthermore, traditional methods generally lack in-depth modeling of human driving behavior, failing to effectively identify and predict HDV intentions, leading to unreasonable right-of-way allocation and low cooperation efficiency in multi-vehicle interaction scenarios.

[0025] The multi-vehicle collaborative dynamic negotiation scheduling method is applied to multi-vehicle collaborative dynamic negotiation scheduling devices. Figure 1 A schematic diagram of the overall process of a multi-vehicle cooperative dynamic negotiation scheduling method is shown. Figure 1 As shown, it includes: S1. Acquire longitudinal following behavior data, lateral lane-changing behavior data, semantic data of collaborative requests broadcast by connected vehicles via V2X, and collaborative vehicle knowledge graph.

[0026] The behavioral data of human-driven vehicles comes from the environmental dynamics simulation submodule. The longitudinal behavior of HDVs is modeled using the Intelligent Driver Model (IDM), while the lateral lane-changing behavior is described using the MOBIL model. The IDM model captures the vehicle's acceleration and deceleration behavior, while the MOBIL model evaluates the rationality of lane changes through lane-changing excitation functions and safety constraints.

[0027] For example, lane-changing intentions are determined by the following indicators: safe distance, speed gain, comfort, and their weighted combination.

[0028] At the same time, ICVs broadcast semantic request information such as their location, speed, and expected behavior through the V2X protocol; in addition, the cooperative vehicle knowledge graph pre-encodes traffic rules and right-of-way priorities for various typical scenarios (such as lane changes, intersections, and merging of main and auxiliary roads).

[0029] S2. Based on the cooperative vehicle knowledge graph, the longitudinal following behavior data, lateral lane-changing behavior data, and cooperative request semantics are processed using an LLM model to obtain the HDV driving intention probability distribution and ICV cooperative intention vector.

[0030] The inference module converts the collected longitudinal following data, lateral lane-changing data, and semantic request information into text fragments. These fragments are then fused with an LLM model and a knowledge graph to perform multimodal semantic inference. This process not only identifies the behavioral intentions of HDVs (such as going straight, changing lanes, and slowing down), but also, by combining traffic regulations and road topology, infers the cooperative behavioral intentions of ICVs, further outputting the probability distribution of HDV driving intentions and the ICV cooperative intention vector.

[0031] S3. Determine the inherent right-of-way space of the target area based on the HDV driving intention probability distribution and ICV cooperative intention vector.

[0032] The calculation of the inherent right-of-way space in the target area is based on physical modeling of the current state of the HDV and ICV. The inherent right-of-way space refers to the safe space that each vehicle needs to maintain under the current operation, including the distance between itself and other vehicles, lane change space, etc. The system introduces a vehicle dynamics model, and combines parameters such as the speed, acceleration, and response delay of the HDV and ICV to calculate the total available space in a unit area and determine whether it meets the preset cooperative scheduling threshold.

[0033] S4. When the inherent right-of-way space of the target area does not meet the preset threshold, the LLM model is used to perform multi-vehicle cooperative negotiation processing on the human-driven vehicle and the connected vehicle to obtain a multi-vehicle cooperative semantic decision set.

[0034] When the available space is insufficient, step S4 is executed, employing an LLM model to conduct multi-vehicle cooperative negotiation between the HDV and ICV. The concept of a "negotiable right-of-way parameter set" is introduced, including the expected amount of space to be released, the types of cooperative actions allowed (such as acceleration to yield, lane changing to yield), and their time windows. Based on the LLM language's generative capabilities and game equilibrium reasoning, the interactive negotiation process between multiple vehicles can be simulated to generate a multi-vehicle cooperative semantic decision set, ensuring that the decision simultaneously satisfies road rule constraints and physical feasibility.

[0035] S5. Determine the underlying control parameters of the vehicle based on the multi-vehicle cooperative semantic decision set.

[0036] This process maps multi-vehicle semantic decision commands to low-level vehicle control parameters. A semantic parsing engine can be used to extract high-level action intentions (such as left turn, acceleration), and a proportional controller can be used to calculate acceleration and front wheel angle. Subsequently, kinematic calculations are performed using a bicycle model to output a complete set of control parameters.

[0037] S6. Execute the underlying vehicle control parameters to perform multi-vehicle collaborative scheduling of the human-driven vehicles and connected vehicles.

[0038] The system sends control parameters to the vehicle controller, where the HDV assisted system or ICV autonomous driving system executes the corresponding operations, enabling dynamic collaborative scheduling of multiple vehicles in complex traffic scenarios. Through LLM semantic reasoning and V2X communication, it breaks through the rigid limitations of traditional traffic control, improving scheduling efficiency and the flexibility and safety of traffic decisions.

[0039] This example proposes a multi-vehicle cooperative dynamic negotiation scheduling method. First, it acquires longitudinal following and lateral lane-changing behavior data of human-driven vehicles, as well as semantic data of cooperative requests broadcast by connected vehicles via V2X. Combined with a constructed cooperative vehicle knowledge graph, it uses an LLM model to perform multimodal semantic reasoning on traffic behavior data, accurately extracting the probability distribution of HDV driving intentions and ICV cooperative intention vectors. Based on right-of-way space calculation and intention game, it generates multi-vehicle cooperative semantic decisions, which are finally transformed into underlying control parameters to realize multi-vehicle scheduling and cooperative control. This method can accurately identify the complex interaction intentions of HDVs and ICVs in mixed traffic environments, dynamically construct a reasonable right-of-way allocation model, and overcome the limitations of traditional traffic cooperative systems, such as rigid decision-making and difficulty in adapting to dynamic changes, by using a multi-vehicle game negotiation approach, thereby improving vehicle scheduling efficiency and decision-making safety.

[0040] In one possible implementation, such as Figure 2 As shown, Figure 2 This paper illustrates a flowchart of the process for determining the HDV driving intention probability distribution and ICV collaborative intention vector in a multi-vehicle cooperative dynamic negotiation scheduling method. The step involves processing the longitudinal following behavior data, lateral lane-changing behavior data, and cooperative request semantics using an LLM model based on the cooperative vehicle knowledge graph to obtain the HDV driving intention probability distribution and ICV collaborative intention vector, including: S201. Perform natural language description conversion on the longitudinal following behavior data and the lateral lane changing behavior data to obtain traffic scene text fragment data.

[0041] S202. Based on the traffic scenario text fragment data and the collaborative vehicle knowledge graph, a rule matching engine is used to retrieve relevant traffic regulations and determine rule constraint prompts. S203. Based on the traffic scene text fragment data, collaborative request semantics, and rule constraint prompts, perform LLM multimodal intent reasoning to obtain the HDV driving intent probability distribution and ICV collaborative intent vector.

[0042] S204. Use a vehicle dynamics model to perform a physical feasibility check on the HDV driving intention probability distribution and ICV cooperative intention vector, and output the HDV driving intention probability distribution and ICV cooperative intention vector that have passed the physical feasibility check.

[0043] Specifically, HDV's longitudinal and lateral behavioral data within a certain time window (e.g., the past 3 seconds) can be converted into traffic description text with linguistic semantics. The conversion process combines the vehicle's current speed, acceleration, pose information, and relative relationship with surrounding vehicles to generate scene fragments, such as: "A vehicle traveling at a medium speed is changing lanes to the left in the main lane, and there is a slow-moving vehicle ahead." This conversion module is implemented through feature template matching or deep learning language modeling techniques.

[0044] Secondly, semantic parsing of scene text can be performed based on the constructed collaborative vehicle knowledge graph. The knowledge graph encompasses traffic rule information, conflict priorities, and common behavioral restrictions under different traffic structures. For example, in the "collaborative lane changing" scenario, the system will associate the "lateral conflict" rule in the graph, namely, "when there is insufficient safety clearance, lane-changing vehicles should yield to vehicles in the target lane." After retrieving clauses that semantically match the current scene through the rule matching engine, relevant prompts are extracted as inference constraints, such as "yield to vehicles in the main lane" and "maintain minimum safe distance."

[0045] Next, three types of data—"traffic scene text fragments," "V2X cooperative request semantics," and "rule constraint prompts"—are input into the LLM model. LLM, through natural language processing and cross-modal semantic fusion capabilities, jointly analyzes the possible driving intentions of the HDV in the current context and expresses them as a probability distribution (e.g., 50% chance of maintaining the current lane, 30% chance of changing lanes to the left, and 20% chance of changing lanes to the right). Simultaneously, it infers the cooperative strategy expression vectors of the ICV (e.g., accelerating to yield, actively merging lanes, etc.), thus constructing the foundation for multi-vehicle interaction semantics.

[0046] To ensure the physical feasibility of the semantic reasoning results, the system introduces a vehicle dynamics model to constrain and verify these results. This model considers parameters such as acceleration boundaries, maximum steering wheel angle, and distance between the vehicle and other vehicles, and verifies the practical feasibility of each intent through simulation. Only when acceleration, lane change trajectory, and other conditions meet safety constraints are the corresponding intent probabilities or vectors retained and used for subsequent scheduling.

[0047] The final output of the HDV driving intent probability distribution and ICV collaborative intent vector not only has semantic understanding capabilities, but also meets traffic rules and dynamic safety constraints, providing a solid data foundation for subsequent right-of-way arbitration and multi-vehicle collaborative scheduling.

[0048] In one possible implementation, when the inherent right-of-way space of the target area does not meet a preset threshold, an LLM model is used to perform multi-vehicle cooperative negotiation processing on the human-driven vehicles and connected vehicles to obtain a multi-vehicle cooperative semantic decision set, including: S401. Based on the inherent right-of-way space gap and the preset threshold, the amount of right-of-way to be released is calculated using a space demand quantification algorithm.

[0049] S402. Based on the amount of right-of-way to be released and the ICV cooperative intent vector, LLM is used to generate a set of negotiable right-of-way parameters, which includes cooperative action type, space release amount and time window.

[0050] S403. Based on the negotiable right-of-way parameter set, send a cooperation request to the associated connected vehicles using the V2X communication protocol, and receive the feedback cooperation commitment vector.

[0051] S404. Based on the collaborative commitment vector and the HDV driving intention probability distribution, LLM is used to perform multi-vehicle game equilibrium calculation to generate the multi-vehicle collaborative semantic decision set.

[0052] Specifically, the system first identifies the "inherent right-of-way space" of each vehicle within the current traffic target area through vehicle dynamics calculations and perception fusion, which is the minimum safe control area required by each vehicle in its current state. If the system determines that the remaining space in this area is insufficient to meet the lane-changing, turning, or yielding needs of the target vehicle, the "multi-vehicle cooperative negotiation" module can be triggered.

[0053] Secondly, the amount of space to be released, ΔS, can be determined using a space demand quantification algorithm based on the gap (defined as the required right-of-way space minus the actual remaining space). This quantification takes into account factors such as the demand of main vehicles, conflict types, and rule priorities.

[0054] Next, the system constructs a "negotiable right-of-way parameter set" by combining ICV cooperative intent vectors (e.g., intention to actively decelerate, change lanes to the right to avoid obstacles, etc.). This parameter set describes cooperative options in natural language or vector representation, including action type (e.g., acceleration, deceleration, lane change), the amount of space to be released (e.g., 0.5 vehicle length), and acceptable response time window (e.g., completing the operation within 3 seconds).

[0055] This parameter set is then broadcast as a "cooperation request" to surrounding ICVs via the V2X protocol. The receiving ICV responds based on its own state (such as acceleration capability and road boundaries) and generates an information vector containing whether it accepts cooperation and the release space and time it can provide, namely the "cooperation commitment vector".

[0056] Based on the collected collaborative commitment vectors and the probability distribution of HDV driving intentions, the system constructs a multi-vehicle game model. After considering the benefits (such as traffic speed and lane-changing safety) and costs (such as additional braking and path deviation) of each party, it uses LLM for semantic reasoning to derive a set of optimal equilibrium strategies. These strategies are expressed in the form of semantic decision sets, such as "ICV-A right lane change to avoid the obstacle" and "HDV maintains its current lane and waits 3 seconds before changing lanes".

[0057] In one possible implementation, determining the vehicle's underlying control parameters based on the multi-vehicle cooperative semantic decision set includes: S501. Based on the multi-vehicle collaborative semantic decision set, a semantic parsing engine is used to extract high-level control instructions, wherein the high-level control instructions include action type, action range, and timeliness constraints.

[0058] S502. Based on the high-level control instructions, calculate the vehicle's kinematic trajectory using a bicycle model to determine the original control parameter set.

[0059] S503. Based on the original control parameter set, a safety boundary check algorithm is used to verify physical feasibility. When the verification passes, the vehicle's underlying control parameters are output; when the verification fails, the LLM model is triggered to regenerate the multi-vehicle cooperative semantic decision set.

[0060] The semantic parsing engine can parse natural language descriptions in the "multi-vehicle collaborative semantic decision set", such as "the main vehicle maintains a constant speed in the current lane for 3 seconds and then changes lanes to the left to avoid the following vehicle", into structured instruction information, including: the action type is "change lanes to the left", the action range is "1.2 times the lane width", and the time constraint is "complete within 3 seconds".

[0061] Subsequently, based on this structured high-level control command, the vehicle's trajectory and state variable updates can be derived using a bicycle model. This model can calculate the vehicle's pose update using the following formula: By combining the amplitude of the motion with the control time window, a set of original control parameters for a time series can be obtained (such as the desired acceleration sequence, steering wheel angle sequence, etc.).

[0062] After trajectory planning is completed, the physical feasibility of the original control parameter set can be verified. This verification algorithm can detect whether the control signal exceeds the vehicle's physical capability boundaries, such as whether the acceleration exceeds the maximum acceleration / deceleration threshold, whether the front wheel steering angle exceeds the mechanical limit, and whether the trajectory violates road boundaries or conflicts with the trajectories of surrounding vehicles.

[0063] Once the control parameters pass the aforementioned safety verification, they can be used as the vehicle's underlying control parameters and directly executed by the vehicle. If infeasibility or potential conflicts are found during the physical verification phase, the LLM model can be called again, and the semantic decision set can be regenerated by combining the latest traffic state information. The above steps can be repeated until a set of underlying control parameters that meet safety, real-time performance, and physical constraints is obtained.

[0064] One possible implementation also includes: S7. Evaluate the step of acquiring longitudinal following behavior data, lateral lane-changing behavior data, cooperative request semantic data broadcast by connected vehicles via V2X, and cooperative vehicle knowledge graph to execute the underlying control parameters of the vehicle and perform multi-vehicle cooperative scheduling of the human-driven vehicle and connected vehicles, and obtain the k-th scheduling evaluation result.

[0065] S8A. If the information indicated by the k-th scheduling evaluation result is a collision, increased conflict, or other negative result, the k-th scheduling evaluation result is transformed into a high-dimensional vector result of the k-th scheduling evaluation.

[0066] S9. Record the high-dimensional vector result of the k-th scheduling evaluation into the memory bank.

[0067] S8B. If the information indicated by the k-th scheduling evaluation result is safe, then this step ends.

[0068] In this example, after executing the vehicle's underlying control parameters and performing multi-vehicle collaborative scheduling of human-driven vehicles and connected vehicles, the entire scheduling process can be evaluated and analyzed to improve the system's safety and self-optimization capabilities. This process includes: evaluating the acquired longitudinal following behavior data and lateral lane-changing behavior data of human-driven vehicles, the semantic data of collaborative requests broadcast by connected vehicles via V2X, and the collaborative vehicle knowledge graph, along with the semantic decision set, control parameters, and actual execution feedback generated during the scheduling process, as a complete scheduling trajectory to obtain the k-th scheduling evaluation result; if the evaluation result indicates a collision, escalation of conflict, or other negative outcomes, the evaluation result can be encoded as a high-dimensional vector representation; subsequently, this high-dimensional vector result can be stored in memory; if the evaluation result indicates safety, there is no need to update the memory.

[0069] In practical implementation, the k-th scheduling evaluation can be based on the status data and traffic results collected during the actual execution phase, and use a conflict detection algorithm to determine whether there are negative events such as collisions, sudden braking, or dangerous approach. For example, anomalies can be determined by judging whether the minimum safe distance threshold has been exceeded, whether the longitudinal deceleration has exceeded the critical value, or whether there is a sudden turning behavior.

[0070] If the aforementioned negative events are detected, a "scheduling evaluation description" can be constructed, which includes: scene description (such as lane structure, relative vehicle position, traffic rules), behavioral decisions (such as lane changing intention, acceleration and deceleration decisions), interaction results (such as whether the vehicle is given way or whether it successfully cuts in line), and negative consequences (such as collision or interference). This description is then converted into a dense semantic vector using a text embedding model (such as BERT) to form the high-dimensional vector result of the k-th scheduling evaluation.

[0071] Next, the vector result can be stored in the memory bank as an experience storage unit for semantic retrieval in similar scenarios in the future. The memory bank is organized using a key-value structure, supporting the query of Top-k similar historical experiences through the current scenario vector, and serving as a prompt input for the LLM semantic reasoning process, prompting the model "In similar situations, a decision of ×× led to ×× consequences, please avoid it".

[0072] If the k-th scheduling evaluation result is safe, that is, no conflict occurs during the scheduling process, the behavior is stable, and the intentions of each vehicle are achieved without interference, then there is no need to perform a memory update operation to avoid redundancy in the system memory and a decrease in efficiency.

[0073] Through the above steps, negative experience can be accumulated and reused based on the post-scheduling evaluation mechanism, making the method not only have real-time decision-making capabilities, but also long-term learning and dynamic optimization capabilities, thus enhancing its stability and adaptability in complex traffic environments.

[0074] For those consistent with the above, please refer to Figure 3 , Figure 3 This application provides a schematic diagram of the structure of a multi-vehicle cooperative dynamic negotiation scheduling device. For example... Figure 3 As shown, the device includes: The acquisition unit is used to acquire longitudinal following behavior data of human-driven vehicles, lateral lane-changing behavior data, semantic data of collaborative requests broadcast by connected vehicles through V2X, and collaborative vehicle knowledge graph. The first processing unit is used to process the longitudinal following behavior data, lateral lane changing behavior data and collaborative request semantics based on the collaborative vehicle knowledge graph using an LLM model, to obtain the HDV driving intention probability distribution and ICV collaborative intention vector. The second processing unit is used to determine the inherent right-of-way space of the target area based on the HDV driving intention probability distribution and the ICV cooperative intention vector. The third processing unit is used to perform multi-vehicle cooperative negotiation processing on the human-driven vehicle and the connected vehicle using the LLM model when the inherent right-of-way space of the target area does not meet the preset threshold, so as to obtain a multi-vehicle cooperative semantic decision set. The fourth processing unit is used to determine the underlying control parameters of the vehicle based on the multi-vehicle cooperative semantic decision set. The scheduling unit is used to execute the underlying control parameters of the vehicle and to perform multi-vehicle collaborative scheduling of the human-driven vehicles and connected vehicles.

[0075] One possible implementation also includes: A memory unit is configured to convert the k-th scheduling evaluation result into a high-dimensional vector result of the k-th scheduling evaluation if the information indicated by the k-th scheduling evaluation result is a collision, escalation of conflict, or other negative result. The high-dimensional vector result of the kth scheduling evaluation is entered into the memory bank; If the information indicated by the k-th scheduling evaluation result is safe, then this step ends.

[0076] In one possible implementation, in the aspect of processing the longitudinal following behavior data, lateral lane-changing behavior data, and cooperative request semantics using an LLM model based on the cooperative vehicle knowledge graph to obtain the HDV driving intention probability distribution and ICV cooperative intention vector, the first processing unit is configured to: The longitudinal following behavior data and the lateral lane changing behavior data are converted into natural language descriptions to obtain traffic scene text fragment data. Based on the traffic scenario text fragment data and the collaborative vehicle knowledge graph, a rule matching engine is used to retrieve relevant traffic regulations and determine rule constraint prompts. Based on the traffic scene text fragment data, collaborative request semantics, and rule constraint prompts, LLM multimodal intent reasoning is performed to obtain the HDV driving intent probability distribution and ICV collaborative intent vector; The vehicle dynamics model is used to perform a physical feasibility check on the HDV driving intention probability distribution and ICV cooperative intention vector, and the HDV driving intention probability distribution and ICV cooperative intention vector that pass the physical feasibility check are output.

[0077] In one possible implementation, when the inherent right-of-way space of the target area does not meet a preset threshold, an LLM model is used to perform multi-vehicle cooperative negotiation processing on the human-driven vehicle and the connected vehicle to obtain aspects of the multi-vehicle cooperative semantic decision set. The third processing unit is used for: Based on the inherent right-of-way space gap and the preset threshold, the amount of right-of-way to be released is calculated using a space demand quantification algorithm; Based on the amount of right-of-way to be released and the ICV cooperative intent vector, an LLM is used to generate a set of negotiable right-of-way parameters, which includes cooperative action type, space release amount and time window. Based on the negotiable right-of-way parameter set, a cooperation request is sent to the associated connected vehicles using the V2X communication protocol, and a cooperation commitment vector is received in response. Based on the collaborative commitment vector and the HDV driving intention probability distribution, LLM is used to perform multi-vehicle game equilibrium calculation to generate the multi-vehicle collaborative semantic decision set.

[0078] In one possible implementation, in the aspect of determining the vehicle's underlying control parameters based on the multi-vehicle cooperative semantic decision set, the fourth processing unit is configured to: Based on the multi-vehicle collaborative semantic decision set, a semantic parsing engine is used to extract high-level control instructions, wherein the high-level control instructions include action type, action range and time constraints. Based on the high-level control instructions, the vehicle's kinematic trajectory is calculated using a bicycle model to determine the original set of control parameters; Based on the original control parameter set, a safety boundary verification algorithm is used to verify physical feasibility. When the verification passes, the underlying control parameters of the vehicle are output. When the verification fails, the LLM model is triggered to regenerate the multi-vehicle cooperative semantic decision set.

[0079] For examples consistent with the above embodiments, please refer to... Figure 4 , Figure 4A schematic diagram of a terminal structure provided in an embodiment of this application is shown in the figure. It includes a processor, an input device, an output device, and a memory. The processor, input device, output device, and memory are interconnected. The memory is used to store a computer program, which includes program instructions. The processor is configured to call the program instructions. The program includes instructions for performing the following steps. Acquire longitudinal following behavior data, lateral lane-changing behavior data, semantic data of collaborative requests broadcast by connected vehicles via V2X, and a knowledge graph of collaborative vehicles; Based on the cooperative vehicle knowledge graph, the longitudinal following behavior data, lateral lane-changing behavior data, and cooperative request semantics are processed using an LLM model to obtain the HDV driving intention probability distribution and ICV cooperative intention vector. The inherent right-of-way space of the target area is determined based on the HDV driving intention probability distribution and the ICV cooperative intention vector. When the inherent right-of-way space of the target area does not meet the preset threshold, the LLM model is used to perform multi-vehicle cooperative negotiation processing on the human-driven vehicles and connected vehicles to obtain a multi-vehicle cooperative semantic decision set. Based on the multi-vehicle cooperative semantic decision set, determine the underlying vehicle control parameters; The underlying vehicle control parameters are executed to perform multi-vehicle collaborative scheduling of the human-driven vehicles and connected vehicles.

[0080] This example proposes a multi-vehicle cooperative dynamic negotiation scheduling method. First, it acquires longitudinal following and lateral lane-changing behavior data of human-driven vehicles, as well as semantic data of cooperative requests broadcast by connected vehicles via V2X. Combined with a constructed cooperative vehicle knowledge graph, it uses an LLM model to perform multimodal semantic reasoning on traffic behavior data, accurately extracting the probability distribution of HDV driving intentions and ICV cooperative intention vectors. Based on right-of-way space calculation and intention game, it generates multi-vehicle cooperative semantic decisions, which are finally transformed into underlying control parameters to realize multi-vehicle scheduling and cooperative control. This method can accurately identify the complex interaction intentions of HDVs and ICVs in mixed traffic environments, dynamically construct a reasonable right-of-way allocation model, and overcome the limitations of traditional traffic cooperative systems, such as rigid decision-making and difficulty in adapting to dynamic changes, by using a multi-vehicle game negotiation approach, thereby improving vehicle scheduling efficiency and decision-making safety.

[0081] In one possible implementation, a multi-vehicle collaborative dynamic negotiation scheduling system is provided, comprising three modules: a scenario management module, an inference module, and a memory module. Scene Management Module The scenario management module implements cooperative driving simulation and control through two sub-modules: the environmental dynamics simulation sub-module uses an Intelligent Driver Model (IDM) and a Minimized Lane Change Induced Braking Model (MOBIL) to simulate the longitudinal following and lateral lane-changing behaviors of human-driven vehicles (HDVs), respectively, constructing a realistic mixed traffic scenario; the model-based control execution sub-module transforms the semantic decision commands output by the LLM into low-level control parameters and uses a bicycle model to accurately calculate the vehicle's motion state, ensuring accurate execution of decisions. This module, by decoupling high-level decision-making from low-level control, retains the semantic reasoning advantages of the LLM while ensuring the accuracy and safety of vehicle control.

[0082] Reasoning module: This module uses state interaction, intent sharing, and recognition to arrive at a final decision through right-of-way arbitration and negotiation.

[0083] The reasoning module first obtains vehicle and lane information from the environment and roadside control unit. It then arbitrates and negotiates right-of-way in conflict areas by sharing intents of ICVs and recognizing intents of HDVs. Finally, it makes a multi-vehicle collaborative decision to arrive at a safe and efficient decision.

[0084] Memory module: This module enhances the system's continuous learning capabilities through a dual mechanism of memory storage and memory retrieval. Drawing inspiration from the mechanism by which human drivers accumulate experience and optimize decisions through practice, the memory module endows autonomous vehicles with human-like experience learning capabilities: it transforms historical interaction data into a searchable knowledge base and enables cross-scenario experience reuse through Retrieval Enhanced Generation (RAG) technology, allowing the vehicle to optimize current interaction strategies based on historical experience.

[0085] Scene Management Module (1) Environmental Dynamics Simulation Submodule To simulate background traffic flow in real traffic environments (especially human-driven HDVs) and provide realistic interactive scenarios.

[0086] Furthermore, considering that the combination of Intelligent Driver Model (IDM) and Minimize Lane Change Induced Braking Model (MOBIL) is widely used to characterize human driving behavior and has been proven effective in various scenarios such as intersections, roundabouts, and merging zones, we use IDM and MOBIL models respectively to simulate the longitudinal and lateral behavior of human-driven vehicles (HDVs).

[0087] Longitudinal motion (IDM model): The formula for controlling the acceleration and deceleration of HDV is as follows: in For the desired speed, Minimum parking distance, The desired headway.

[0088] Lateral motion (MOBIL model): The decision-making process for HDV lane-changing behavior includes lane-changing incentives and safety checks. in As a politeness factor, This is the safety braking threshold.

[0089] (2) Model-based control execution submodule The semantic decisions output by the LLM (such as {decelerate, cruise, accelerate, change lanes left, change lanes right}) are transformed into vehicle-level control signals (acceleration). Front wheel steering angle ): Control logic: A proportional controller is used to map higher-level commands to lower-level signals. in For reference acceleration, For the desired heading angle, , To control the gain.

[0090] Kinematics Update: Calculate the vehicle state at the next moment using a bicycle model: State Interaction The interaction between vehicle and lane information is fundamental to making the entire decision. Lane information is crucial for determining traffic rules or subsequent right-of-way arbitration and negotiation. Different road structures have different arbitration rules. Road information is obtained as follows: Road topology: same-lane following scenario, cooperative lane changing scenario, adjacent lane connection scenario, and intersection scenario. Road signs: Go straight, turn left, and turn right. Types of road collisions: rear-end collisions, side-impact collisions, head-on collisions Speed ​​limits and traffic volume on the road: , The vehicle's driving status reflects its dynamic information in a timely manner. In addition to its own vehicle information, acquiring basic driving information of surrounding vehicles is indispensable. This allows for more accurate trajectory planning and vehicle control. Current lane status Target lane status Current vehicle coordinates The current speed of the vehicle is The current acceleration of the vehicle is Current vehicle width Current vehicle length The vehicle's yaw angle is The distance between the vehicle and the vehicle in front in the current lane is The distance between the vehicle and the vehicle behind in the current lane is 2) Intent recognition and sharing During driving, ICVs can obtain the vehicle's driving intentions through intelligent network sharing, and conduct right-of-way arbitration and negotiation in advance to obtain the final driving decision. However, HDVs need to identify their driving intentions, determine the HDV's intention to go straight, turn left or turn right through the road return sign information, predict its subsequent trajectory through its 3-second historical trajectory, and analyze the HDV's lane-changing intentions using three indicators: distance, speed and comfort.

[0091] (1) Safety distance index In the formula: These are the lane and vehicle number, respectively. For lane vehicle The required safe distance, in meters; Vehicle reaction time, in seconds; It is half the reciprocal of the vehicle's maximum deceleration, in m / s² 2 ; The vehicle length is in meters (m). For lane vehicle At any moment The speed, m / s; For lane Get on the vehicle and The distance between them, in meters (m).

[0092] (2) Speed ​​index HV speed gains For vehicles Speed ​​after changing lanes to the target lane, m / s; The time for changing lanes is s.

[0093] In the formula, For the maximum turning angle, This refers to the vehicle's acceleration.

[0094] PFV speed gain of the vehicle behind in the target lane In the formula: For a safe headway, s; for Braking deceleration of vehicles changing lanes at any time, m / s 2 ; The maximum braking deceleration of the vehicle following in the target lane, in m / s² 2 ; The safe speed for vehicles changing lanes behind in the target lane, in m / s.

[0095] The PFV speed gain of the vehicle behind in the target lane is thus obtained. : (3) Comfort index In the formula: For lane Vehicle space occupancy rate. 0 in, Let be the total length of the observed road segment, in meters. For the first The length of the vehicle, in meters (m); To observe the number of vehicles within the road segment.

[0096] (4) Comprehensive lane-changing intention judgment index In the formula: For the penalty function; The desired speed of the vehicle is given in m / s. , It is a penalty parameter; The current acceleration is in m / s². 2 .

[0097] Using real-time vehicle operating status data as input, the system calculates safety, speed, and comfort indicators for both the vehicle and surrounding vehicles, and then uses a comprehensive lane-changing intention determination index to make a comprehensive decision on the HDV's driving lane and lane-changing intention.

[0098] 3) Multi-vehicle collaborative right-of-way arbitration To better ensure that every vehicle is protected by traffic regulations and has the right to maintain normal driving conditions, we combine right-of-way arbitration and negotiation with the LLM model to digitize road safety regulations.

[0099] The road topology information is obtained through road status interaction, and the corresponding traffic regulations for the scenario are used to construct a knowledge graph using an LLM model, as shown in the table below: When multiple vehicles intend to enter the same area, they are coordinated by priority. The higher priority vehicle is allocated the area first. HDVs always have higher priority than ICVs. When ICVs compete for the area, the priority is as shown in the table above. When all ICVs are changing lanes, forced lane changing has higher priority than free lane changing.

[0100] While LLM models possess strong reasoning capabilities, the lack of physical model constraints still presents potential safety risks. Therefore, to better guide multi-vehicle collaborative decision-making in intelligent connected vehicle interaction conflict scenarios, this paper integrates the reasoning capabilities of LLM with cyber-physical models for multi-vehicle collaborative right-of-way arbitration and negotiation. Specifically, it combines right-of-way calculations to arbitrate the driving rights of each vehicle. Each vehicle traveling on the road has its inherent safety space, which we call inherent right-of-way. During the lane-changing process of the primary vehicle (HV), the vehicles primarily affected include the vehicle in front in its own lane, the vehicle in front in the target lane, and the vehicle behind in the target lane. (1) The vehicle in front of this vehicle has the inherent right-of-way. In the formula: The speed of the main vehicle (hv) ), Let lag time be 0 seconds. The reaction time of an autonomous vehicle can generally be taken as 0 seconds. deceleration of the main vehicle (m / s) 2 ); The maximum deceleration of the vehicle (m / s) 2 ), The minimum deceleration of the vehicle (m / s²) 2 ), The maximum speed limit for this lane is ( ).

[0101] (2) Right of way of the vehicle on the left rear In the formula: The speed of the vehicle to the left rear ( ), This represents the human driver's reaction time lag (in seconds), typically taken as 2 seconds. Other parameters are the same as above.

[0102] (3) Right of way of the vehicle on the left rear Negotiable right of way (4) Right of way for vehicles on the left front Multi-vehicle collaborative right-of-way scheduling In the above situations, priority decisions for different vehicles are made through right-of-way arbitration. However, when the safety clearance between the target vehicle and the front and rear vehicles are both ICVs, right-of-way negotiation can be carried out to allow the front and rear vehicles to make operations such as acceleration and deceleration that do not affect themselves, thereby freeing up more free right-of-way, giving lane-changing vehicles more right-of-way, making lane-changing safer, and realizing dynamic scheduling of right-of-way in multi-vehicle collaboration.

[0103] It also includes: memory modules.

[0104] This allows vehicles to "learn" from past experiences to avoid repeating mistakes, achieving continuous learning and self-evolution. It achieves this through the following two main sub-functions: Memory Storage: After each vehicle completes an action, the system immediately evaluates its effect. If a collision, escalation of conflict, or other negative consequences occur, the system encapsulates the following sequence into an experience record: "Scene description (including lane, surrounding vehicle status, and intent) — Conflict description (conflict response to the conflict coordinator's proposed passage order) — Selected action — Final result." This record is then converted into a high-dimensional vector using a text embedding model and stored in the memory bank. If the action is safe, it does not need to be recorded. This avoids massive redundancy and ensures that every record in the memory bank represents a "worrying" negative example, providing valuable reference for subsequent retrievals.

[0105] Memory Retrieval: Before the next round of decision-making, the system vectorizes the current scenario and conflict description, and uses cosine similarity to retrieve the top-k most similar experiences (typically k=2) from the memory bank. These experiences are then concatenated into the LLM's prompts in natural language, explicitly stating, "Taking action ×× in a similar scenario last time led to ×× consequences; please avoid repeating the same mistake." The retrieval process uses dynamic thresholding to filter low-relevance memories, preventing irrelevant information from diluting the context.

[0106] The above mainly describes the solutions of the embodiments of this application from the perspective of the method execution process. It is understood that, in order to achieve the above functions, the terminal includes the corresponding hardware structure and / or software modules for executing each function. Those skilled in the art should readily recognize that, in conjunction with the units and algorithm steps of the various examples described in the embodiments provided herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0107] This application embodiment can divide the terminal into functional units according to the above method example. For example, each function can be divided into a separate functional unit, or two or more functions can be integrated into one processing unit. The integrated unit can be implemented in hardware or as a software functional unit. It should be noted that the unit division in this application embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods.

[0108] This application also provides a computer storage medium storing a computer program for electronic data interchange, which causes a computer to perform some or all of the steps of any of the multi-vehicle cooperative dynamic negotiation scheduling methods described in the above method embodiments.

[0109] This application also provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program that causes a computer to perform some or all of the steps of any of the multi-vehicle cooperative dynamic negotiation scheduling methods described in the above method embodiments.

[0110] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0111] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0112] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical or other forms.

[0113] The units described as separate components may or may not be physically separate. The components shown as units 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 units can be selected to achieve the purpose of this embodiment according to actual needs.

[0114] Furthermore, the functional units in the various embodiments of the application 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 as a software program module.

[0115] If the integrated unit is implemented as a software program module and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0116] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage device, which may include: a flash drive, a read-only memory, a random access memory, a magnetic disk, or an optical disk, etc.

[0117] The embodiments of this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A multi-vehicle collaborative dynamic negotiation scheduling method, characterized in that, include: Acquire longitudinal following behavior data, lateral lane-changing behavior data, semantic data of collaborative requests broadcast by connected vehicles via V2X, and a knowledge graph of collaborative vehicles; Based on the cooperative vehicle knowledge graph, the longitudinal following behavior data, lateral lane-changing behavior data, and cooperative request semantics are processed using an LLM model to obtain the HDV driving intention probability distribution and ICV cooperative intention vector. The inherent right-of-way space of the target area is determined based on the HDV driving intention probability distribution and the ICV cooperative intention vector. When the inherent right-of-way space of the target area does not meet the preset threshold, the LLM model is used to perform multi-vehicle cooperative negotiation processing on the human-driven vehicles and connected vehicles to obtain a multi-vehicle cooperative semantic decision set. Based on the multi-vehicle cooperative semantic decision set, determine the underlying vehicle control parameters; The underlying vehicle control parameters are executed to perform multi-vehicle collaborative scheduling of the human-driven vehicles and connected vehicles.

2. The multi-vehicle cooperative dynamic negotiation scheduling method according to claim 1, characterized in that, The process involves using an LLM model to process the longitudinal following behavior data, lateral lane-changing behavior data, and cooperative request semantics based on the cooperative vehicle knowledge graph, resulting in an HDV driving intention probability distribution and an ICV cooperative intention vector, including: The longitudinal following behavior data and the lateral lane changing behavior data are converted into natural language descriptions to obtain traffic scene text fragment data. Based on the traffic scenario text fragment data and the collaborative vehicle knowledge graph, a rule matching engine is used to retrieve relevant traffic regulations and determine rule constraint prompts. Based on the traffic scene text fragment data, collaborative request semantics, and rule constraint prompts, LLM multimodal intent reasoning is performed to obtain the HDV driving intent probability distribution and ICV collaborative intent vector; The vehicle dynamics model is used to perform a physical feasibility check on the HDV driving intention probability distribution and ICV cooperative intention vector, and the HDV driving intention probability distribution and ICV cooperative intention vector that pass the physical feasibility check are output.

3. The multi-vehicle cooperative dynamic negotiation scheduling method according to claim 1, characterized in that, When the inherent right-of-way space of the target area does not meet a preset threshold, an LLM model is used to perform multi-vehicle cooperative negotiation processing on the human-driven vehicles and connected vehicles to obtain a multi-vehicle cooperative semantic decision set, including: Based on the inherent right-of-way space gap and the preset threshold, the amount of right-of-way to be released is calculated using a space demand quantification algorithm; Based on the amount of right-of-way to be released and the ICV cooperative intent vector, an LLM is used to generate a set of negotiable right-of-way parameters, which includes cooperative action type, space release amount and time window. Based on the negotiable right-of-way parameter set, a cooperation request is sent to the associated connected vehicles using the V2X communication protocol, and a cooperation commitment vector is received in response. Based on the collaborative commitment vector and the HDV driving intention probability distribution, LLM is used to perform multi-vehicle game equilibrium calculation to generate the multi-vehicle collaborative semantic decision set.

4. The multi-vehicle cooperative dynamic negotiation scheduling method according to claim 1, characterized in that, The step of determining the underlying vehicle control parameters based on the multi-vehicle cooperative semantic decision set includes: Based on the multi-vehicle collaborative semantic decision set, a semantic parsing engine is used to extract high-level control instructions, wherein the high-level control instructions include action type, action range and time constraints. Based on the high-level control instructions, the vehicle's kinematic trajectory is calculated using a bicycle model to determine the original set of control parameters; Based on the original control parameter set, a safety boundary verification algorithm is used to verify physical feasibility. When the verification passes, the underlying control parameters of the vehicle are output. When the verification fails, the LLM model is triggered to regenerate the multi-vehicle cooperative semantic decision set.

5. The multi-vehicle cooperative dynamic negotiation scheduling method according to claim 1, characterized in that, Also includes: The steps of acquiring longitudinal following behavior data, lateral lane-changing behavior data of human-driven vehicles, semantic data of cooperative requests broadcast by connected vehicles via V2X, and cooperative vehicle knowledge graphs, and executing the underlying control parameters of the vehicles to perform multi-vehicle cooperative scheduling of human-driven vehicles and connected vehicles are evaluated to obtain the k-th scheduling evaluation result. If the information indicated by the k-th scheduling evaluation result is that a collision has occurred or the conflict has intensified, the k-th scheduling evaluation result is transformed into a high-dimensional vector result of the k-th scheduling evaluation. The high-dimensional vector result of the kth scheduling evaluation is entered into the memory bank; If the information indicated by the k-th scheduling evaluation result is safe, then this step ends.

6. A multi-vehicle collaborative dynamic negotiation scheduling device, characterized in that, include: The acquisition unit is used to acquire longitudinal following behavior data of human-driven vehicles, lateral lane-changing behavior data, semantic data of collaborative requests broadcast by connected vehicles through V2X, and collaborative vehicle knowledge graph. The first processing unit is used to process the longitudinal following behavior data, lateral lane changing behavior data and collaborative request semantics based on the collaborative vehicle knowledge graph using an LLM model, to obtain the HDV driving intention probability distribution and ICV collaborative intention vector. The second processing unit is used to determine the inherent right-of-way space of the target area based on the HDV driving intention probability distribution and the ICV cooperative intention vector. The third processing unit is used to perform multi-vehicle cooperative negotiation processing on the human-driven vehicle and the connected vehicle using the LLM model when the inherent right-of-way space of the target area does not meet the preset threshold, so as to obtain a multi-vehicle cooperative semantic decision set. The fourth processing unit is used to determine the underlying control parameters of the vehicle based on the multi-vehicle cooperative semantic decision set. The scheduling unit is used to execute the underlying control parameters of the vehicle and to perform multi-vehicle collaborative scheduling of the human-driven vehicles and connected vehicles.

7. The multi-vehicle cooperative dynamic negotiation scheduling device according to claim 6, characterized in that, Also includes: A memory unit is used to convert the k-th scheduling evaluation result into a high-dimensional vector result if the information indicated by the k-th scheduling evaluation result is a collision or an escalation of conflict. The high-dimensional vector result of the kth scheduling evaluation is entered into the memory bank; If the information indicated by the k-th scheduling evaluation result is safe, then this step ends.

8. The multi-vehicle collaborative dynamic negotiation scheduling device according to claim 6, characterized in that, In the aspect of processing the longitudinal following behavior data, lateral lane-changing behavior data, and cooperative request semantics using an LLM model based on the cooperative vehicle knowledge graph to obtain the HDV driving intention probability distribution and ICV cooperative intention vector, the first processing unit is used to: The longitudinal following behavior data and the lateral lane changing behavior data are converted into natural language descriptions to obtain traffic scene text fragment data. Based on the traffic scenario text fragment data and the collaborative vehicle knowledge graph, a rule matching engine is used to retrieve relevant traffic regulations and determine rule constraint prompts. Based on the traffic scene text fragment data, collaborative request semantics, and rule constraint prompts, LLM multimodal intent reasoning is performed to obtain the HDV driving intent probability distribution and ICV collaborative intent vector; The vehicle dynamics model is used to perform a physical feasibility check on the HDV driving intention probability distribution and ICV cooperative intention vector, and the HDV driving intention probability distribution and ICV cooperative intention vector that pass the physical feasibility check are output.

9. A terminal, characterized in that, The system includes a processor, an input device, an output device, and a memory, which are interconnected. The memory stores a computer program, which includes program instructions. The processor is configured to invoke the program instructions to execute the multi-vehicle cooperative dynamic negotiation scheduling method as described in any one of claims 1-5.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, the computer program including program instructions, which, when executed by a processor, cause the processor to perform the multi-vehicle cooperative dynamic negotiation scheduling method as described in any one of claims 1-5.

Images