A training method and system of a V2X ring traffic strategy-robot task graph linkage and a storage medium

Abstract

The present application belongs to the field of intelligent transportation and robot control technology, and discloses a kind of V2X in ring traffic strategy-robot task graph linkage training method, system and storage medium;Among them, the method includes according to task graph generation for heterogeneous robot execution sequence, in ring environment execution execution sequence, obtain multi-source sensing feedback and execution log, based on multi-source sensing feedback and execution log obtain operation key index, according to operation key index evaluation, obtain evaluation result, if evaluation result meets preset threshold value, then according to evaluation result update strategy weight and robot action parameter, generate replayable script and evidence chain, with replayable script as input, evidence chain as comparison benchmark, establish consistency compensation mapping, based on consistency compensation mapping output consistency score, complete V2X robot training;The present application can realize the structured association of V2X strategy and heterogeneous robot action, and cross-site rapid migration.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation and robot control technology, specifically to a training method, system, and storage medium for V2X-based traffic strategy-robot task graph linkage. Background Technology

[0002] Currently, intelligent transportation and robotic collaboration technologies for road operations mainly fall into two categories: one is V2X-based roadside traffic strategy application technology, which issues commands such as traffic control and diversion to vehicles through facilities such as RSUs, traffic signals, and guidance screens; the other is embodied intelligent robot single-machine control technology for tasks such as obstacle removal and detection. However, both have long been in a stage of independent development and fragmented application, without forming a unified collaborative system.

[0003] In existing technologies, there is a lack of unified representation and closed-loop linkage mechanism between V2X traffic strategies and heterogeneous robots. Roadside control and diversion instructions are only directed at vehicles, and robots still need to rely on manual scheduling to complete tasks such as placing traffic cones and clearing obstacles, resulting in emergency response times of 15-30 minutes, making minute-level automated collaboration impossible. At the same time, environmental parameters (such as slope, friction coefficient, and line of sight) vary significantly across different sites. The strategy parameters and robot action scripts in existing technologies cannot be automatically transferred and adapted. Cross-site deployment requires more than 48 hours of re-adjustment, and it is impossible to dynamically optimize strategy weights and action parameters based on operational KPIs such as traffic capacity recovery rate and hazard exposure duration.

[0004] Therefore, the existing technologies lack a unified representation and closed-loop linkage mechanism between V2X traffic strategies and heterogeneous robots, and cannot achieve rapid cross-site migration and KPI-driven dynamic optimization, which urgently need to be addressed. Summary of the Invention

[0005] The purpose of this invention is to provide a training method, system, and storage medium for V2X-based traffic strategy-robot task graph linkage, in order to overcome the problems existing in the prior art. This invention can not only realize the structured association between V2X strategy and heterogeneous robot actions, but also realize rapid migration across sites.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: In a first aspect, the present invention provides a training method for V2X-based traffic strategy-robot task graph linkage, comprising the following steps: S1, receive V2X policy frames and scene images of the corresponding site from the roadside unit, wherein the V2X policy frames include signal control policies, speed limit policies and guidance screen policies. S2, parse the V2X policy frame to obtain the policy primitive set and constraint vector, and construct a task graph including skill nodes, condition nodes and safety nodes based on the policy primitive set, constraint vector and scene profile; S3, Generate an execution sequence for heterogeneous robots based on the task graph, wherein the heterogeneous robots include at least two of the following: ground mobile robots, legged robots, and drones; S4, execute the execution sequence in the loop environment to obtain multi-source sensor feedback and execution log, wherein the multi-source sensor feedback includes robot body sensor data, environmental perception data, traffic flow and roadside state data; S5. Based on the multi-source sensor feedback and execution log, key operational indicators are obtained, and an evaluation is performed based on the key operational indicators to obtain an evaluation result. The key operational indicators include at least one of the following: lockdown completion time, traffic capacity recovery rate, road restoration time, and duration of hazardous exposure. S6. If the evaluation result meets the preset threshold, update the strategy weight and robot action parameters according to the evaluation result, generate a replayable script and evidence chain, and execute S7; if the evaluation result does not meet the preset threshold, execute S2 again. S7. Using the replayable script as input and the evidence chain as a comparison benchmark, establish a consistency compensation mapping, output a consistency score based on the consistency compensation mapping, and complete the V2X robot training.

[0007] In some embodiments, parsing the V2X policy frame to obtain a set of policy primitives and constraint vectors specifically includes: The signal control strategy is mapped to the control node in the task graph, the speed limiting strategy is mapped to the speed constraint node in the task graph, and the induction screen strategy is mapped to the guide node in the task graph. The set of policy primitives is generated based on the control node, velocity constraint node, and guiding node. Extract the spatiotemporal constraints and priority constraints implicit in the control nodes, velocity constraint nodes, and guidance nodes, and quantize the spatiotemporal constraints and priority constraints into constraint vectors.

[0008] In some embodiments, the construction of a task graph including skill nodes, condition nodes, and safety nodes based on the set of policy primitives, constraint vectors, and scenario profiles specifically includes: Based on the set of policy primitives, the robot actions corresponding to each policy primitive are extracted, and the robot actions are mapped to skill nodes in the task graph. Condition nodes are constructed based on constraint vectors, and safety nodes are extracted from the constraint vectors. Based on the scene profile, the skill nodes, condition nodes, and safety nodes are dynamically pruned and prioritized to generate the task graph.

[0009] In some embodiments, the in-loop environment includes at least one of software in-loop, hardware in-loop, and vehicle in-loop.

[0010] In some embodiments, the evaluation based on key operational indicators to obtain evaluation results specifically includes: The operational key indicators are weighted using a joint objective function to obtain the evaluation result, wherein the joint objective function includes a safety regularization term.

[0011] In some embodiments, if the evaluation result meets a preset threshold, updating the strategy weights and robot action parameters based on the evaluation result, and generating a replayable script and evidence chain, specifically includes: If the evaluation result meets the preset threshold, the weight vector in the joint objective function is adjusted to obtain the optimized strategy weights. At the same time, the robot motion parameters are optimized to obtain the optimized robot motion parameters. The robot motion parameters include adjusting the cone deployment spacing, the towing robot speed, the UAV flight altitude, and the broadcast power. The optimized strategy weights and the optimized robot motion parameters are solidified to generate a replayable script; The robot's sensor data, environmental perception data, traffic flow and roadside status data, and the V2X strategy frame are globally synchronized to obtain a global time synchronization result. The hash values ​​of the global time synchronization results are calculated at each level and then concatenated to form a chain structure, which is then bound to a replayable script for storage and used for safety auditing and process traceability.

[0012] In some embodiments, the step of establishing a consistency compensation mapping using the replayable script as input and the evidence chain as a comparison benchmark, and outputting a consistency score based on the consistency compensation mapping, specifically includes: Using the replayable script as input, the optimized strategy weights and optimized robot motion parameters are loaded, and the entire process is reproduced and executed in the in-loop environment of the target site. Using the evidence chain as a comparison benchmark, the execution log of the target site is hash-compared with the evidence chain. Based on the comparison result, a consistency compensation mapping is established, and the optimized strategy weight and the optimized robot action parameters are dynamically adjusted to obtain the dynamic adjustment result. A consistency score is output based on the dynamically adjusted results.

[0013] Secondly, the present invention provides a training system for V2X-based traffic strategy-robot task graph linkage, comprising: The V2X interface module is used to receive V2X policy frames and scene images, wherein the V2X policy frames include signal control policies, speed limiting policies, and guidance screen policies. The task graph engine module is used to parse the V2X policy frame, obtain the policy primitive set and constraint vector, and construct a task graph including skill nodes, condition nodes and safety nodes based on the policy primitive set, constraint vector and scene profile. An execution controller module is used to generate an execution sequence for heterogeneous robots based on the task graph, and then schedule the heterogeneous robots, wherein the heterogeneous robots include at least two of the following: ground mobile robots, legged robots, and drones. In-loop environment module, used to execute the execution sequence in an in-loop environment to obtain multi-source sensor feedback and execution log, wherein the multi-source sensor feedback includes robot body sensor data, environmental perception data, traffic flow and roadside state data; The evaluation module is used to obtain key operational indicators based on the multi-source sensor feedback and execution logs, evaluate them, and obtain evaluation results. The key operational indicators include at least one of the following: lockdown completion time, traffic capacity recovery rate, road restoration time, and hazard exposure duration. The evaluation result judgment module is used to determine: if the evaluation result meets the preset threshold, the strategy weight and robot action parameters are updated according to the evaluation result, a replayable script and evidence chain are generated, and the consistency regression module is executed; if the evaluation result does not meet the preset threshold, the task graph engine module is executed again. The consistency regression module is used to establish a consistency compensation mapping by taking the replayable script as input and the evidence chain as a comparison benchmark, and output a consistency score based on the consistency compensation mapping to complete the V2X robot training.

[0014] Thirdly, the present invention provides a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method.

[0015] Fourthly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described method.

[0016] The above technical solution has the following advantages or beneficial effects: Firstly, this invention provides a training method for V2X-based traffic strategy-robot task graph linkage. Addressing the shortcomings of existing technologies, such as the lack of a unified representation and closed-loop linkage mechanism between V2X traffic strategies and heterogeneous robots, and the inability to achieve rapid cross-site migration and KPI-driven dynamic optimization, this invention constructs a unified mapping from strategy frames to task graphs, achieving a structured association between V2X strategies and heterogeneous robot actions. This reduces emergency collaborative response time from the traditional 15-30 minutes to less than 3 minutes. Furthermore, through replayable scripts and evidence chain mechanisms, the entire process is auditable and reproducible. Finally, through consistency compensation mapping and parallel regression verification, rapid cross-site migration is achieved, reducing regression testing time from 48 hours to less than 8 hours, with a consistency score exceeding 95%. This significantly reduces deployment costs and improves robustness in complex scenarios.

[0017] In some embodiments, by mapping V2X strategies such as signal control, speed limiting, and guidance screens to control nodes, speed constraint nodes, and guidance nodes in the task graph, respectively, and generating a set of policy primitives and constraint vectors accordingly, a unified structured expression of heterogeneous V2X instructions and robot actions is realized. This provides standardized input for subsequent task graph construction, timing arrangement, and collaborative scheduling, and improves the accuracy and interpretability of multi-source policy fusion.

[0018] In some embodiments, by supporting switching between at least one or more combinations of software-in-the-loop, hardware-in-the-loop, and vehicle-in-the-loop, this method can cover the entire process testing requirements from algorithm simulation to real vehicle verification, taking into account both development efficiency and verification authenticity, and providing a flexible, configurable, and virtual-real integrated in-the-loop training environment for V2X strategies and robot collaboration.

[0019] In some embodiments, by employing a joint objective function that includes a safety regularization term to perform weighted evaluation of key operational indicators such as lockdown completion time, traffic capacity recovery rate, and duration of hazard exposure, multi-objective quantitative optimization of strategy weights and action parameters is achieved. This improves collaborative efficiency while taking into account operational safety, making the training results more in line with real operational needs.

[0020] In some embodiments, by evaluating and solidifying the optimized strategy weights and robot motion parameters after achieving the target, a replayable script is generated, which realizes the standardized output and cross-scenario reuse of training results; at the same time, an immutable evidence chain is constructed based on global time synchronization and hash chain technology, which is bound to the script for storage, providing full-process reliable support for security auditing, process traceability and compliance certification.

[0021] In some embodiments, by optimizing key motion parameters such as the deployment spacing of traffic cones, the speed of the towing robot, the flight altitude of the drone, and the broadcast power, precise control over the control efficiency, operational safety, and environmental adaptability is achieved, enabling heterogeneous robots to adaptively adjust their execution strategies under different site conditions, and significantly improving the scenario generalization capability of collaborative operations.

[0022] Secondly, this invention provides a training system for V2X-based traffic strategy-robot task graph linkage. By using a replayable script as input and an evidence chain as a benchmark, the system reproduces the execution at the target site and performs hash comparison, establishes a consistency compensation mapping, and dynamically adjusts the strategy weights and action parameters. Finally, it outputs a quantitative consistency score, realizing cross-site quantitative evaluation and rapid adaptation of training results, significantly shortening the migration cycle, and ensuring the predictability and compliance of deployment effects.

[0023] Thirdly, the present invention provides a computer device that, through a processor executing a specific computer program, can efficiently implement the steps of the method of the present invention. When performing data processing tasks, the computer device can accurately perform numerical calculations and logical judgments, avoiding errors caused by human factors. At the same time, since the computer program has high stability and reliability, it can ensure the accuracy and consistency of the data processing results.

[0024] Fourthly, the present invention provides a computer-readable storage medium. By programming the steps of the method of the present invention into a computer program and storing it on the computer-readable storage medium, users can easily load these programs onto any compatible computer device and execute them without rewriting or converting the code, which greatly improves the convenience and flexibility of program execution. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of a training method for V2X-in-the-loop traffic strategy-robot task graph linkage according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a training system for V2X-in-the-loop traffic strategy-robot task graph linkage, as shown in an embodiment of the present invention. Figure 3 This is a schematic diagram of V2X policy frame mapping as shown in an embodiment of the present invention; Figure 4 This is a schematic diagram of the chain of evidence shown in an embodiment of the present invention; Figure 5 This is a schematic diagram of consistency compensation mapping as shown in an embodiment of the present invention; Figure 6 This is a schematic diagram of the structure of a computer device as shown in an embodiment of the present invention. Detailed Implementation

[0026] The present invention will be further described in detail below with reference to specific embodiments. These descriptions are for explanation purposes only and are not intended to limit the scope of the invention.

[0027] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0028] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0029] Historical technologies related to intelligent transportation and robot collaboration can be broadly categorized into the following core areas, all of which are currently in a "fragmented application" stage: **Standalone script-based embodied intelligent robot technology:** Robots (ground-towing robots, legged detection robots, and drones) in road operation scenarios (such as accident clearance and road closures) are controlled by independent scripts, only able to respond to preset local commands. For example, drones are only responsible for aerial photography, and towing robots only perform fixed-path clearance, lacking linkage logic with roadside traffic facilities. **Isolated V2X traffic strategy application technology:** While V2X infrastructure such as RSU roadside units, traffic signals, variable speed limit signs, and guidance screens have been deployed in scenarios like autonomous driving test sites and urban arterial roads, their output traffic strategies (such as closure commands and diversion guidance) are only directed to vehicles or traffic control centers, lacking a collaborative mechanism with robot actions. Strategy execution relies on manual scheduling (such as manually placing traffic cones and manually coordinating clearance robots with traffic signals). Single-dimensional in-the-loop testing technology: Existing in-the-loop testing (SIL software in the loop, HIL hardware in the loop, VIL vehicle in the loop) are mostly used separately for autonomous driving algorithm verification or robot single-machine performance testing, without forming an integrated in-the-loop training environment of "V2X policy-robot action-environment feedback".

[0030] The existing technologies lack a unified representation and closed-loop linkage mechanism between V2X traffic strategies and heterogeneous robots, and cannot achieve rapid cross-site migration and KPI-driven dynamic optimization, which urgently need to be solved.

[0031] This embodiment provides a training method for V2X-based traffic strategy-robot task graph linkage, see [link / reference]. Figure 1 This includes the following steps: Step 1: Receive V2X policy frames and corresponding scene images from roadside units, wherein the V2X policy frames include signal control policies, speed limit policies and guidance screen policies. Step 2: parse the V2X policy frame to obtain the policy primitive set and constraint vector, and construct a task graph including skill nodes, condition nodes and safety nodes based on the policy primitive set, constraint vector and scene profile; Step 3: Generate an execution sequence for heterogeneous robots based on the task graph, wherein the heterogeneous robots include at least two of the following: ground mobile robots, legged robots, and drones; Step 4: Execute the execution sequence in the loop environment to obtain multi-source sensor feedback and execution log, wherein the multi-source sensor feedback includes robot body sensor data, environmental perception data, traffic flow and roadside status data; Step 5: Based on the multi-source sensor feedback and execution log, obtain key operational indicators, evaluate them, and obtain evaluation results. The key operational indicators include at least one of the following: lockdown completion time, traffic capacity recovery rate, road restoration time, and hazard exposure duration. Step 6: If the evaluation result meets the preset threshold, update the strategy weight and robot action parameters according to the evaluation result, generate a replayable script and evidence chain, and execute Step 7; if the evaluation result does not meet the preset threshold, execute Step 2 again. Step 7: Using the replayable script as input and the evidence chain as a comparison benchmark, establish a consistency compensation mapping, output a consistency score based on the consistency compensation mapping, and complete the V2X robot training.

[0032] This invention constructs a unified mapping mechanism from V2X policy frames to task graphs, parsing multi-source policies such as signal control, speed limits, and guidance screens into structured policy primitives and constraint vectors. These are then integrated with robot skill nodes, condition nodes, and safety nodes to form an executable DAG task graph. This achieves unified representation and collaborative scheduling of V2X traffic policies and heterogeneous robot actions, effectively solving the problems of separation and reliance on manual scheduling in existing technologies. Emergency response time is reduced from 15-30 minutes to less than 3 minutes. Simultaneously, by using replayable scripts to solidify optimized policy weights and action parameters, and generating an immutable evidence chain based on hash chain technology, combined with a consistency compensation mapping mechanism, policy parameters and robot actions can be automatically adapted according to environmental differences such as slope, line of sight, and friction in new sites. This reduces cross-site regression testing time from 48 hours to less than 8 hours, achieving a consistency score of over 95%, and enabling rapid transfer and quantitative evaluation of training results.

[0033] Example: This embodiment provides a training method for V2X-based traffic strategy-robot task graph linkage. It parses the RSU / signal / guide screen policy frames into task graph nodes and constraints, and uses an event orchestrator to uniformly schedule heterogeneous robots (ground / legged / airborne) to execute auditable policy primitives. The evaluation results are used as training metrics to complete a closed loop of generation—execution—evaluation—replayback—update, including the following steps: Step 1: Receive V2X policy frames and corresponding scene images from roadside units, wherein the V2X policy frames include signal control policies, speed limit policies, and guidance screen policies.

[0034] In some embodiments, step 1 specifically includes: The roadside cooperative unit receives policy frames from the cloud or edge nodes; it acquires real-time data on the current road environment, traffic flow, weather conditions, and vehicle status through roadside detectors, traffic flow detectors, weather sensors, and positioning devices; it merges historical operation data with current real-time scene data, updates and generates the latest scene profile in real time.

[0035] Step 2: Parse the V2X policy frame to obtain the policy primitive set and constraint vector. Based on the policy primitive set, constraint vector and scene profile, construct a task graph including skill nodes, condition nodes and safety nodes.

[0036] In some embodiments, parsing the V2X policy frame to obtain a set of policy primitives and constraint vectors specifically includes: Step 2.1, see Figure 3 The signal control strategy is mapped to the control node in the task graph, the speed limiting strategy is mapped to the speed constraint node in the task graph, and the guidance screen strategy is mapped to the guidance node in the task graph.

[0037] Step 2.2: Generate the set of policy primitives based on the control node, velocity constraint node, and guiding node.

[0038] In some embodiments, control command primitives are extracted based on the control node, speed constraint primitives are extracted based on the speed constraint node, and guidance information primitives are extracted based on the guidance node. The set of strategy primitives is generated by combining the control command primitives, speed constraint primitives, and guidance information primitives.

[0039] In some embodiments, the set of policy primitives may further include blockade primitives, array transformation primitives, drag-and-drop primitives, containment primitives, broadcast primitives, plateau primitives, and stick-to-the-ground primitives.

[0040] Step 2.3: Extract the spatiotemporal constraints and priority constraints implicit in the control node, velocity constraint node and guidance node, and quantize the spatiotemporal constraints and priority constraints into constraint vectors.

[0041] Step 2.4: Extract robot actions corresponding to each policy primitive based on the policy primitive set, map robot actions to skill nodes in the task graph, construct condition nodes based on constraint vectors, extract safety nodes from constraint vectors, and dynamically prune and prioritize skill nodes, condition nodes, and safety nodes based on scene profiles to generate the task graph.

[0042] In some embodiments, step 2.4 specifically includes: The robot actions corresponding to each primitive are extracted from the parsed set of policy primitives. These actions are then mapped to skill nodes in the task graph, where each skill node represents an executable atomic action (such as drone broadcasting, cone deployment, or obstacle removal). Based on the temporal, dependency, and priority constraints in the constraint vector, condition nodes are constructed to connect the triggering relationships between skill nodes. For example, "drone broadcasting completed" serves as a condition node to trigger the "cone deployment" skill node. Safety parameters such as collision threshold, emergency stop domain, and yield rules are extracted from the constraint vector to construct safety nodes, which are then associated with all relevant skill nodes in the form of global edge constraints. Based on the road type, environmental parameters, and event level in the scene profile, the skill nodes, condition nodes, and safety nodes are dynamically pruned and prioritized to form a directed acyclic graph (DAG) task graph, where edges represent temporal dependencies, mutual exclusion, or parallel relationships between nodes.

[0043] In some embodiments, the task graph may further include control nodes, velocity constraint nodes, and guiding nodes.

[0044] In some embodiments, the safety node includes at least one of an emergency stop domain, soft and hard limits, a collision threshold, and a pedestrian yielding priority.

[0045] Step 3: Generate an execution sequence for heterogeneous robots based on the task graph, wherein the heterogeneous robots include at least two of the following: ground mobile robots, legged robots, and drones.

[0046] In some embodiments, the heterogeneous robot may further include drone broadcasting, traffic cone array transformation, towing / blocking, and foot-based detection.

[0047] Step 4: Execute the execution sequence in the in-loop environment to obtain multi-source sensor feedback and execution log, wherein the multi-source sensor feedback includes robot body sensor data, environmental perception data, traffic flow and roadside state data.

[0048] In some embodiments, the in-loop environment includes at least one of Software-in-the-Loop (SIL), Hardware-in-the-Loop (HIL), and Vehicle-in-the-Loop (VIL), and the in-loop environments are switchable.

[0049] In some embodiments, a task graph is distributed to heterogeneous robots using an auditable policy primitive library, and the execution sequence is executed in an in-loop environment.

[0050] In some embodiments, this step provides all the measured data required for closed-loop training, enabling the execution-evaluation-optimization-feedback process to form a complete closed loop, ensuring that the training process is reproducible, verifiable, and auditable.

[0051] Step 5: Based on the multi-source sensor feedback and execution log, obtain the key operational indicators (KPIs), evaluate them, and obtain the evaluation results. The key operational indicators include at least one of the following: lockdown completion time, traffic capacity recovery rate, road restoration time, and hazard exposure duration.

[0052] In some embodiments, the evaluation based on key operational indicators to obtain evaluation results specifically includes: The operational key indicators are weighted using a joint objective function to obtain the evaluation result, wherein the joint objective function includes a safety regularization term.

[0053] In some embodiments, the weighting of the key operational indicators using a joint objective function includes the following formula: ; In the formula, This represents the comprehensive optimization objective function; Indicates the weight of the lockdown completion time; Indicators representing the evaluation criteria for the completion time of lockdown; Indicates the weight of road restoration time; Indicators representing road restoration efficiency; Indicates the weight of improved traffic capacity; Indicators representing the degree of traffic recovery; Indicates the weight of the duration of hazardous exposure; This indicates an assessment of the speed at which risk is eliminated; Indicates the weight of the security item; This indicates compliance with safety constraints.

[0054] Step 6: If the evaluation result meets the preset threshold, update the strategy weight and robot action parameters according to the evaluation result, generate a replayable script and evidence chain, and execute Step 7; if the evaluation result does not meet the preset threshold, execute Step 2 again.

[0055] In some embodiments, if the evaluation result meets a preset threshold, updating the strategy weights and robot action parameters based on the evaluation result, and generating a replayable script and evidence chain, specifically includes: Step 6.1: If the evaluation result meets the preset threshold, adjust the weight vector in the joint objective function to obtain the optimized strategy weights, and at the same time optimize the robot motion parameters to obtain the optimized robot motion parameters.

[0056] Step 6.2: Solidify the optimized strategy weights and the optimized robot motion parameters to generate a replayable script. The replayable script is used to generate a certification report and for external evaluation services.

[0057] In some embodiments, the replayable script solidifies the trained and optimized V2X policy, task graph structure, timing logic, robot motion parameters, and policy weights. It can be directly loaded, repeatedly executed, and migrated across environments in SIL / HIL / VIL, representing a standardized executable output of this training method. The evidence chain binds V2X policy frames, execution sequence, motion trajectory, sensor feedback, and evaluation results using time-synchronized hashing, forming an immutable execution record for security auditing, compliance certification, and process traceability. The replayable script and evidence chain are continuously used in subsequent processes: for cross-site consistency regression and parallel verification, serving as standard inputs for regression testing to calculate consistency scores; for scenario storage and reuse, serving as standard scenario units for the traffic operation event library; for security certification and auditing, meeting the traceability, reproducibility, and verifiability requirements of traffic operation testing; and for fault reproduction and iterative training, serving as the basis for subsequent optimization and incremental learning.

[0058] Step 6.3: Perform global time synchronization on the robot body sensing data, environmental perception data, traffic flow and roadside status data, and the V2X strategy frame to obtain the global time synchronization result. Calculate the hash value of the global time synchronization result level by level and then concatenate them to form a chain structure. Bind and store it with the replayable script for security auditing and process traceability.

[0059] In some embodiments, the robot motion parameters include adjusting the cone deployment spacing, the robot towing speed, the drone flight altitude, and the broadcast power.

[0060] In some embodiments, the evidence chain is constructed as follows: first, the V2X policy frame, robot motion trajectory and key sensor snapshots are globally synchronized in time so that the three types of data are unified to the same time base; then, based on the time-synchronized data, the hash value is calculated level by level in the execution order to form a chain hash structure (hash chain).

[0061] In some embodiments, the hash chain serves to: prevent tampering of data throughout the entire process, ensure traceability of the execution process, and enable reproducibility of test results, thereby meeting the requirements of traffic operation safety audits, third-party certification, and compliance verification. The hash chain is continuously used in subsequent steps: bound to replayable scripts for benchmark comparison in cross-site consistency regression verification; incorporated into certification reports as proof of compliance; used as a unique and trusted identifier for event entry; and used for training process replay, fault location, and accountability tracing.

[0062] In some embodiments, heterogeneous robots collaborate through a task graph engine, supporting failure rollback and timeout bypass strategies.

[0063] Step 7: Using the replayable script as input and the evidence chain as a comparison benchmark, establish a consistency compensation mapping, output a consistency score based on the consistency compensation mapping, and complete the V2X robot training.

[0064] In some embodiments, the step of establishing a consistency compensation mapping using the replayable script as input and the evidence chain as a comparison benchmark, and outputting a consistency score based on the consistency compensation mapping, specifically includes: Step 7.1: Using the replayable script as input, load the optimized strategy weights and optimized robot motion parameters, and reproduce the entire process in the in-loop environment of the target site.

[0065] Step 7.2, see Figure 5 Using the evidence chain as a comparison benchmark, the execution log of the target site is hash-compared with the evidence chain. Based on the comparison result, a consistency compensation mapping is established, and the optimized strategy weight and optimized robot action parameters are dynamically adjusted (parallel regression is performed) to obtain the dynamic adjustment result.

[0066] In some embodiments, the consistency compensation mapping is further established based on the slope, sight distance, surface friction, and wind field parameters of each site.

[0067] Step 7.3: Output a consistency score based on the dynamic adjustment results.

[0068] In some embodiments, timestamps, policy frames, action trajectories, sensor data, and execution logs are hashed and bound to form an immutable chain of evidence throughout the entire process; at the same time, a replayable script and authentication report are generated based on the collected data, and a consistency score is calculated during cross-site regression.

[0069] In some embodiments, the consistency score is used to quantitatively evaluate the adaptability and execution consistency of the trained strategy, task graph, and action parameters when deployed across sites, and is the ultimate basis for determining whether the training has met the deployment requirements.

[0070] The core innovation of this invention lies in constructing a full-link closed-loop system encompassing V2X policy frames, task graphs, robot collaboration, in-loop training, and cross-site adaptation. This innovation can be summarized by three main mechanisms. First, through a unified mapping mechanism between policy and task graphs, policy frames output from RSUs, traffic lights, and guidance screens are parsed into standardized policy primitives. Combined with scene profiling, a DAG task graph containing skill nodes, condition nodes, safety nodes, control nodes, speed constraint nodes, and guidance nodes is constructed, achieving a structured association between traffic policies and robot actions. Second, the in-loop closed-loop training system for heterogeneous robot collaboration adopts a switchable in-loop environment integrating SIL / HIL / VIL. It maps the task graph into a temporal execution sequence of multiple robot types and automatically adjusts the weight allocation in the joint objective function based on scene characteristics such as road type, event level, and environmental conditions. This enables the system to dynamically optimize multiple objectives, including operational safety, traffic efficiency, and detection accuracy, in diverse scenarios such as highways, urban roads, mountain roads, cross-sea bridges, and emergency situations, forming a closed-loop iteration of generation, execution, evaluation, and update. Finally, the auditable evidence chain and cross-site consistency compensation mechanism bind policy frames, robot motion trajectories and key sensor snapshots into an immutable evidence chain through time-synchronized hash chain technology, covering the entire process from policy issuance to environmental feedback. It also uses site profiles to establish a consistency compensation mapping model and combines parallel regression scheduling to achieve automatic adaptation of policy parameters, effectively shortening the cross-site migration cycle and ensuring the stable and efficient operation of the system in different deployment environments.

[0071] This invention demonstrates significant advantages. First, by unifying the scheduling of heterogeneous robots through a task graph, minute-level emergency coordination can be achieved, reducing emergency lockdown and diversion response time from the traditional 15-30 minutes to ≤3 minutes, and increasing traffic capacity recovery rate by over 40%. Second, the end-to-end evidence chain (hash chain + timestamp) meets audit and authentication requirements, ensuring full traceability of the strategy-action-feedback process, and the generated authentication report directly complies with traffic operation safety audit standards. Regarding cross-site deployment, consistency compensation mapping and parallel regression mechanisms achieve over 95% consistency in cross-site scripts, reducing regression testing time from 48 hours to ≤8 hours and migration costs by 70%. The system exhibits strong robustness in complex scenarios: closed-loop optimization can dynamically adjust strategy weights and robot parameters (e.g., automatically reducing the spacing between road cones and decreasing the speed of towing robots in rainy weather), adapting to various environments such as mountainous areas, cities, and rain / fog, and reducing the multi-agent collaborative collision rate to 0. Meanwhile, multi-dimensional quantitative optimization, with operational KPIs (lockdown time, road reopening time, duration of hazard exposure, etc.) as core indicators, replaces the traditional binary feedback, making the optimization of strategies and robot actions more aligned with actual operational needs. Furthermore, the system is compatible with collaboration among multiple types of intelligent agents, including ground-based, legged, and aerial agents, further enhancing its overall adaptability in complex scenarios.

[0072] In one embodiment of the present invention, a training method for V2X-in-the-loop traffic strategy-robot task graph linkage is provided, comprising the following steps: Step 1: At K3+200 on the expressway, the sensor detects a truck overturning (occupying two lanes). The RSU issues a closure policy frame, and the roadside sensing device uploads a scene profile (including accident location, traffic flow, and weather).

[0073] Step 2: Parse the policy frame to obtain the primitive set (broadcast, blockade, probe, drag) and constraint vector (blockade time limit ≤ 3 min, collision threshold ≤ 0.5 m); construct the task graph based on the primitive set, constraint vector and scene profile (skill nodes: drone broadcast, V-shaped deployment of traffic cones, etc.; condition nodes: traffic cones are deployed after the drone completes the broadcast; safety nodes: emergency stop domain covers the core area of ​​the accident).

[0074] Step 3: The task graph engine maps nodes to a temporal sequence. After receiving a task graph containing skill nodes, condition nodes, and safety nodes, the task graph engine performs topological sorting and temporal constraint parsing on the task graph. Following the traffic operation safety logic of first guiding, then blocking, then probing, and finally handling, the task graph nodes are mapped to a strict multi-agent execution temporal sequence. The specific mapping steps are as follows: First time sequence phase: During the initial period t0 to t30s after the strategy is issued, the task map engine first schedules drones to execute broadcast skill nodes, sending detour prompts to vehicles upstream of the accident area, guiding traffic to change lanes and avoid the accident in advance, creating safe conditions for subsequent control deployment.

[0075] The second time sequence phase: After the drone broadcast is completed and the "broadcast completed" condition is met, the road cone group blockade deployment enters the time period t30s to t2min. The task map engine schedules the road cone group to execute the blockade deployment skill node, and automatically completes the blockade and isolation of the accident area in a V-shaped formation to ensure that the blockade range, spacing and position meet the safety constraint node requirements.

[0076] The third time sequence phase: After the deployment of the road cone group for hazard source detection is completed, the legged robot enters the time period from t2min to t2min30s. The task graph engine schedules the legged robot to enter the closed area to perform the hazard source detection skill node, conduct close inspection of the accident site, identify hazards such as fuel leaks, structural damage, and secondary collision risks, and output the condition node result of "no secondary hazard source".

[0077] Fourth time sequence phase: Obstacle clearing and disposal by towing robot After the legged robot completes the detection and confirms that there are no secondary hazards, the time period from t2min30s to t40min begins. The task map engine schedules the towing robot to execute the obstacle clearing skill node, to steadily tow away, remove and clean up the accident vehicle until the road is ready to be reopened to traffic.

[0078] The above time sequence is uniformly arranged by the task graph engine to ensure the operation of drones, traffic cone groups, legged robots, and towing machines, as shown in Table 1: Table 1 Time Series List

[0079] Step 4: In the VIL environment, the real test vehicle changes lanes in advance after receiving a broadcast from the drone; in the SIL environment, the traffic cone group automatically moves to the target position (error ≤ 0.3 m), the legged robot avoids accident debris and completes the detection, and outputs the execution log (including the action timestamps of each device) and sensor feedback (drone aerial video, traffic cone GPS data, and legged robot infrared detection map).

[0080] Step 5, calculate the KPI vector: lockdown time 2 min 30 s (≤ 3 min), traffic capacity recovery rate 85% (85% of the original traffic capacity after clearing obstacles), road recovery time 40 min, hazard exposure duration 5 min; generate the evidence chain (including t0 policy frame hash, t30s road cone deployment hash, t40 min traffic recovery hash), and output the KPI vector [2.5, 85%, 40, 5]; evidence chain Z (time synchronization hash chain file).

[0081] Step 6, calculate the joint objective function: w1=0.3, w2=0.2, w3=0.3, w4=0.1, w5=0.1, substitute to get J=89 points (threshold 80 points), no need to adjust the policy weights, only optimize the speed parameters of the towing robot (increase from 5 km / h to 8 km / h), output the updated policy weights (no change) and the robot motion parameter table.

[0082] Step 7: KPIs are met (lockdown time ≤ 3 min, J ≥ 80 points), solidify the replayable script (supports repeated execution of the scenario), and the certification report (including KPI curve and evidence chain summary). Output the replayable script (exe format) and the audit certification report (PDF).

[0083] Step 8: If relocation to suburban roads (surface friction 0.6, wind field level 3), establish a site profile (slope 2°, visibility 1000 meters), generate a compensation mapping (road cone deployment spacing reduced from 2 m to 1.5 m, drone flight altitude reduced by 5 m), conduct parallel regression testing, and output a consistency score of 96% and a regression test report.

[0084] In one embodiment of the present invention, a training method for V2X-in-the-loop traffic strategy-robot task graph linkage is provided for V2X robot training in accident-induced lane closure scenarios. This embodiment uses a truck rollover accident on an urban expressway as a scenario to verify the lane closure coordination effect of the method, including the following steps: Step 1: The RSU issues an accident lane closure strategy frame, which includes signal control strategy (turn on the upstream red light), speed limit strategy (speed limit 30km / h) and guidance screen strategy (display "Accident ahead, please detour"), and integrates roadside perception data to generate a scene profile (expressway, sunny day, traffic flow 300 vehicles / hour).

[0085] Step 2: Parse the policy frame to obtain the blocking primitive, broadcast primitive, drag primitive, and detection primitive as a policy primitive set, and extract the constraint vector (blocking time limit ≤ 3 minutes, collision threshold ≤ 0.5 meters); construct the task graph based on the policy primitive set, constraint vector, and scene profile, including skill nodes (drone broadcast, V-shaped deployment of traffic cones, foot detection, drag clearing), condition nodes (broadcast completed, no secondary hazard source), and safety nodes (emergency stop domain).

[0086] Step 3, generate the execution sequence: t0-30 seconds, drone broadcasts detour instructions; t30 seconds-2 minutes, road cones are deployed to form a V-shaped control zone; t2-2.5 minutes, legged robots detect hazards; t2.5-40 minutes, towing robots clear obstacles.

[0087] Step 4: Execute in a VIL+SIL hybrid in-loop environment to collect sensor feedback such as robot trajectory and traffic flow status.

[0088] Step 5, calculate the operational KPIs: lockdown completion time 2 minutes 30 seconds, hazardous exposure duration 5 minutes, road restoration time 40 minutes, and traffic capacity recovery rate 85%.

[0089] Step 6: If the evaluation results meet the threshold, solidify and generate a replayable script and evidence chain to achieve the goal of a lockdown time of ≤3 minutes.

[0090] Step 7: Using the replayable script as input, load the optimized strategy weights and action parameters, and reproduce the execution in the in-loop environment of the target site on mountain roads; using the evidence chain as the comparison benchmark, perform hash comparison between the execution log of the target site and the original evidence chain, establish a consistency compensation mapping (reduce the road cone deployment spacing from 2 m to 1.5 m, and reduce the drone flight altitude by 5 m), output a consistency score of 96%, and complete the training.

[0091] In one embodiment of the present invention, a training method for V2X-based traffic strategy-robot task graph linkage is provided for cross-site regression. This embodiment uses the migration from an urban expressway (site 1) to a mountain road (site 2) as a scenario to verify the cross-site consistency compensation and regression effect of the method, including the following steps: Step 1: Site 1 has completed training and generated a replayable script and evidence chain. When migrating to Site 2, the RSU reissues the same accident road closure policy frame and constructs a scene profile for Site 2 (slope 5°, visibility 800m, surface friction 0.6, wind field level 3).

[0092] Step 2: Load the replayable script in the SIL+HIL hybrid ring environment at site 2 and execute the entire process of lockdown-clearing-recovery. Initial execution revealed: the enhanced wind field caused incomplete drone broadcast coverage (lockdown timed out at 4 minutes), and the reduced ground friction caused the towing robot to slip.

[0093] Step 3: If the evaluation result does not meet the preset threshold, repeat Step 2. Based on the profile characteristics of Site 2, adjust the weights of the joint objective function (increase the weight of the security item). ), optimize robot motion parameters (drone flight altitude reduced from 50 m to 30 m, broadcast power increased from 10 W to 15 W; towing robot speed reduced from 8 km / h to 5 km / h), and generate a new replayable script and evidence chain adapted to site 2.

[0094] Step 4: Using the new replayable script as input and the evidence chain of site 1 as the comparison benchmark, establish a consistency compensation mapping (generate a parameter correspondence table between site 1 and site 2). Perform parallel regression verification (running both the compensation mapping group and the non-compensation mapping group simultaneously). The compensation mapping group achieved a lockdown time of 2 minutes and 50 seconds, a throughput recovery rate of 82%, and a consistency score of 96%. The regression test duration was 7.5 hours (≤8 hours), and the script consistency reached 96% (≥95%), completing cross-site training.

[0095] In one embodiment of the present invention, a training method for V2X-based traffic strategy-robot task graph linkage is provided. This embodiment takes an accident-induced road closure scenario as an example to verify the compliance audit capability of the method, including the following steps: After executing steps 1 to 6, the system generates a replayable script and a chain of evidence. The chain of evidence adopts a hash chain structure, which binds the V2X policy frame hash (t0), the drone broadcast trajectory hash (t30s), the road cone deployment location hash (t2min), the foot detection data hash (t2.5min), and the road recovery status hash (t40min) in chronological order, forming an immutable record of the entire execution process.

[0096] After step 7 is completed, the assessment and playback module automatically calls the preset audit template and generates an authentication report based on the evidence chain. The report contains three parts: first, the KPI curve (lockdown completion time 2 minutes 30 seconds, traffic capacity recovery rate 85%, road restoration time 40 minutes, hazard exposure duration 5 minutes); second, the evidence chain summary (including hash fingerprints and timestamps of each link); and third, bypass records (including abnormal handling logs such as timed bypass and fault rollback). The above authentication report can be directly submitted to the traffic operation management department, meeting the requirements for safety audit and compliance verification, and realizing a traceable and tamper-proof audit record of the entire process of "policy issuance - action execution - environmental feedback".

[0097] This invention receives policy frames and real-time scene data, integrates historical operational data and current environmental data, and updates the scene profile in real time. Based on the policy frames and scene profile, it constructs a globally unified task graph. The task graph undergoes logic, timing, dependency, and scene consistency checks. Based on the verified task graph, it executes global collaborative control between the roadside and vehicle-mounted systems. The execution results are then fed back to update the scene profile, forming a closed-loop iteration. Finally, based on consistency checks, the collaborative control results are output. This invention achieves unified policy, ordered tasks, reliable collaboration, strong scene adaptability, full traceability, auditability, and verifiability, making it suitable for vehicle-road cooperative systems in complex road environments.

[0098] In one embodiment of the present invention, a training system for V2X-in-the-loop traffic strategy-robot task graph linkage is provided, see [link to relevant documentation]. Figure 2 ,include: The V2X interface module is used to receive V2X policy frames and scene images, wherein the V2X policy frames include signal control policies, speed limiting policies, and guidance screen policies. The event orchestrator module is used to break down the core objectives of a task; The task graph engine module is used to parse the V2X policy frame, obtain the policy primitive set and constraint vector, and construct a task graph including skill nodes, condition nodes and safety nodes based on the policy primitive set, constraint vector and scene profile. An execution controller module is used to generate an execution sequence for heterogeneous robots based on the task graph, and then schedule the heterogeneous robots, wherein the heterogeneous robots include at least two of the following: ground mobile robots, legged robots, and drones. In-loop environment module, used to execute the execution sequence in an in-loop environment to obtain multi-source sensor feedback and execution log, wherein the multi-source sensor feedback includes robot body sensor data, environmental perception data, traffic flow and roadside state data; The evaluation module is used to obtain key operational indicators based on the multi-source sensor feedback and execution logs, evaluate them, and obtain evaluation results. The key operational indicators include at least one of the following: lockdown completion time, traffic capacity recovery rate, road restoration time, and hazard exposure duration. The evaluation result judgment module is used to determine: if the evaluation result meets the preset threshold, the strategy weight and robot action parameters are updated according to the evaluation result, a replayable script and evidence chain are generated, and the consistency regression module is executed; if the evaluation result does not meet the preset threshold, the task graph engine module is executed again. The consistency regression module is used to establish a consistency compensation mapping by taking the replayable script as input and the evidence chain as a comparison benchmark, and output a consistency score based on the consistency compensation mapping to complete the V2X robot training.

[0099] In some embodiments, the consistency regression module can also perform parallel regression verification.

[0100] In some embodiments, the task graph engine module selects primitives from a policy primitive library and applies security constraints.

[0101] In some embodiments, the evaluation module and the evaluation result judgment module generate an authentication report according to a preset template.

[0102] See Figure 6 This embodiment provides a computer device, which includes a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal and is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions in the computer storage medium to realize the corresponding method flow or corresponding function. The processor described in this embodiment can be used to execute related operations of a V2X-in-the-loop traffic strategy-robot task graph linkage training method.

[0103] This embodiment provides a computer-readable storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space, which stores the terminal's operating system; and, in this storage space, it also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the training method for a V2X-in-the-loop traffic strategy-robot task map linkage in this embodiment.

[0104] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0105] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0106] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0107] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0108] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A training method for V2X-based traffic strategy-robot task graph linkage, characterized in that, Includes the following steps: S1, receive V2X policy frames and scene images of the corresponding site from the roadside unit, wherein the V2X policy frames include signal control policies, speed limit policies and guidance screen policies. S2, parse the V2X policy frame to obtain the policy primitive set and constraint vector, and construct a task graph including skill nodes, condition nodes and safety nodes based on the policy primitive set, constraint vector and scene profile; S3, Generate an execution sequence for heterogeneous robots based on the task graph, wherein the heterogeneous robots include at least two of the following: ground mobile robots, legged robots, and drones; S4, execute the execution sequence in the loop environment to obtain multi-source sensor feedback and execution log, wherein the multi-source sensor feedback includes robot body sensor data, environmental perception data, traffic flow and roadside state data; S5. Based on the multi-source sensor feedback and execution log, key operational indicators are obtained, and an evaluation is performed based on the key operational indicators to obtain an evaluation result. The key operational indicators include at least one of the following: lockdown completion time, traffic capacity recovery rate, road restoration time, and duration of hazardous exposure. S6. If the evaluation result meets the preset threshold, update the strategy weight and robot action parameters according to the evaluation result, generate a replayable script and evidence chain, and execute S7; if the evaluation result does not meet the preset threshold, execute S2 again. S7. Using the replayable script as input and the evidence chain as a comparison benchmark, establish a consistency compensation mapping, output a consistency score based on the consistency compensation mapping, and complete the V2X robot training.

2. The training method for V2X-in-the-loop traffic strategy-robot task graph linkage according to claim 1, characterized in that, Parsing the V2X policy frame yields the set of policy primitives and constraint vectors, specifically including: The signal control strategy is mapped to the control node in the task graph, the speed limiting strategy is mapped to the speed constraint node in the task graph, and the induction screen strategy is mapped to the guide node in the task graph. The set of policy primitives is generated based on the control node, velocity constraint node, and guiding node. Extract the spatiotemporal constraints and priority constraints implicit in the control nodes, velocity constraint nodes, and guidance nodes, and quantize the spatiotemporal constraints and priority constraints into constraint vectors.

3. The training method for V2X-in-the-loop traffic strategy-robot task graph linkage according to claim 1, characterized in that, The construction of a task graph based on the set of policy primitives, constraint vectors, and scenario profiles, including skill nodes, condition nodes, and safety nodes, specifically includes: Based on the set of policy primitives, the robot actions corresponding to each policy primitive are extracted, and the robot actions are mapped to skill nodes in the task graph. Condition nodes are constructed based on constraint vectors, and safety nodes are extracted from the constraint vectors. Based on the scene profile, the skill nodes, condition nodes, and safety nodes are dynamically pruned and prioritized to generate the task graph.

4. The training method for V2X-in-the-loop traffic strategy-robot task graph linkage according to claim 1, characterized in that, The in-loop environment includes at least one of software in the loop, hardware in the loop, and vehicle in the loop.

5. The training method for V2X-in-the-loop traffic strategy-robot task graph linkage according to claim 1, characterized in that, The evaluation based on key operational indicators, and the resulting evaluation results, specifically include: The operational key indicators are weighted using a joint objective function to obtain the evaluation result, wherein the joint objective function includes a safety regularization term.

6. The training method for V2X-in-the-loop traffic strategy-robot task graph linkage according to claim 5, characterized in that, If the evaluation result meets the preset threshold, the strategy weights and robot action parameters are updated based on the evaluation result, and a replayable script and evidence chain are generated, specifically including: If the evaluation result meets the preset threshold, the weight vector in the joint objective function is adjusted to obtain the optimized strategy weights. At the same time, the robot motion parameters are optimized to obtain the optimized robot motion parameters. The robot motion parameters include adjusting the cone deployment spacing, the towing robot speed, the UAV flight altitude, and the broadcast power. The optimized strategy weights and the optimized robot motion parameters are solidified to generate a replayable script; The robot's sensor data, environmental perception data, traffic flow and roadside status data, and the V2X strategy frame are globally synchronized to obtain a global time synchronization result. The hash values ​​of the global time synchronization results are calculated at each level and then concatenated to form a chain structure, which is then bound to a replayable script for storage and used for safety auditing and process traceability.

7. The training method for V2X-in-the-loop traffic strategy-robot task graph linkage according to claim 6, characterized in that, The process of establishing a consistency compensation mapping using the replayable script as input and the evidence chain as a comparison benchmark, and outputting a consistency score based on the consistency compensation mapping, specifically includes: Using the replayable script as input, the optimized strategy weights and optimized robot motion parameters are loaded, and the entire process is reproduced and executed in the in-loop environment of the target site. Using the evidence chain as a comparison benchmark, the execution log of the target site is hash-compared with the evidence chain. Based on the comparison result, a consistency compensation mapping is established, and the optimized strategy weight and the optimized robot action parameters are dynamically adjusted to obtain the dynamic adjustment result. A consistency score is output based on the dynamically adjusted results.

8. A training system for V2X-based traffic strategy-robot task graph linkage, characterized in that, include: The V2X interface module is used to receive V2X policy frames and scene images, wherein the V2X policy frames include signal control policies, speed limiting policies, and guidance screen policies. The task graph engine module is used to parse the V2X policy frame, obtain the policy primitive set and constraint vector, and construct a task graph including skill nodes, condition nodes and safety nodes based on the policy primitive set, constraint vector and scene profile. An execution controller module is used to generate an execution sequence for heterogeneous robots based on the task graph, and then schedule the heterogeneous robots, wherein the heterogeneous robots include at least two of the following: ground mobile robots, legged robots, and drones. In-loop environment module, used to execute the execution sequence in an in-loop environment to obtain multi-source sensor feedback and execution log, wherein the multi-source sensor feedback includes robot body sensor data, environmental perception data, traffic flow and roadside state data; The evaluation module is used to obtain key operational indicators based on the multi-source sensor feedback and execution logs, evaluate them, and obtain evaluation results. The key operational indicators include at least one of the following: lockdown completion time, traffic capacity recovery rate, road restoration time, and hazard exposure duration. The evaluation result judgment module is used to determine: if the evaluation result meets the preset threshold, the strategy weight and robot action parameters are updated according to the evaluation result, a replayable script and evidence chain are generated, and the consistency regression module is executed; if the evaluation result does not meet the preset threshold, the task graph engine module is executed again. The consistency regression module is used to establish a consistency compensation mapping by taking the replayable script as input and the evidence chain as a comparison benchmark, and output a consistency score based on the consistency compensation mapping to complete the V2X robot training.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1-7.

Images