A method, system, device, and readable storage medium for avoiding special vehicles.

Abstract

This invention discloses a method, system, device, and storage medium for special vehicle obstacle avoidance. The method includes: real-time acquisition of attack mode vectors and message credibility scores; generation of adversarial disturbance scenarios based on the attack mode vectors and a pre-set adversarial generative network; and two-level fusion credibility calculation of the message credibility scores to generate an uncertainty set boundary; and obtaining the final control command based on the adversarial disturbance scenario, the uncertainty set boundary, and a rolling optimization strategy to generate an active obstacle avoidance control command for controlling the vehicle to avoid a target vehicle. This invention uses a dual closed-loop feedback architecture to map credibility to the control uncertainty set boundary, employs a rolling hot-start single-step approximation strategy to solve robust model predictive control, reduces computation time, achieves real-time onboard operation, resists sensor spoofing and V2X attacks, and improves safety.

Description

Technical Field

[0001] This application relates to the field of vehicle control technology, specifically to a method, system, device, and readable storage medium for avoiding obstacles by special vehicles. Background Technology

[0002] With the development of Intelligent Transportation Systems (ITS), special vehicles (such as fire trucks, ambulances, and police cars) are increasingly relying on vehicle-to-everything (V2X) technology and multi-sensor fusion perception systems to achieve priority passage and proactive obstacle avoidance when performing emergency tasks. Existing cooperative obstacle avoidance systems are typically based on model predictive control (MPC) or rule-based algorithms, assuming that sensor data (cameras, radar) and V2X communication messages are absolutely reliable.

[0003] However, in complex real-world environments, special-purpose vehicles face severe safety challenges. On one hand, malicious attackers may interfere with cameras, deceive radar signals, or forge V2X messages, causing the avoidance system to make dangerous decisions based on erroneous information. On the other hand, to improve robustness, traditional robust model predictive control typically employs the Min-Max optimization method, which needs to consider worst-case disturbances. This method often involves multiple nested iterations, with computation times typically on the order of seconds, making it difficult to meet the millisecond-level real-time requirements of vehicle control. Furthermore, existing V2X message credibility assessments often use simple voting mechanisms, failing to consider the impact of vehicle distance on observation accuracy, and are prone to failure when the proportion of malicious nodes is high.

[0004] Existing control strategies rely mainly on human experience, lack analysis of sensor and communication attacks, and lack feedback on detection results, resulting in core problems such as low design efficiency and poor data consistency. Summary of the Invention

[0005] This application provides a method, system, device, and readable storage medium for avoiding obstacles to special vehicles, which can solve the technical problem of poor security of existing special vehicle avoidance systems when facing sensor errors and V2X communication attacks.

[0006] In a first aspect, embodiments of this application provide a method for avoiding special vehicles, the method comprising: The attack pattern vector and message credibility score are acquired in real time. The attack pattern vector is generated based on the multi-sensor data of the target vehicle, and the message credibility score is generated based on the V2X communication messages sent by the target vehicle. Based on the obtained attack pattern vectors and the pre-set adversarial generation network, adversarial perturbation scenarios are generated, and the credibility scores of the obtained messages are fused in two levels to calculate the credibility and generate the uncertainty set boundary. Based on the aforementioned adversarial disturbance scenario, the uncertainty set boundary, and the preset rolling optimization strategy, the final control command is obtained, generating an active avoidance control command for controlling the vehicle to avoid the target vehicle.

[0007] In conjunction with the first aspect, in one implementation, generating adversarial perturbation scenarios through an adversarial generative network based on the attack pattern vector includes: The attack pattern vector is input as a conditional input vector into the conditional Wasserstein generative adversarial network. The conditional Wasserstein generative adversarial network identifies the attack type and attack intensity in the attack pattern vector. Based on the attack type and attack intensity, an adversarial perturbation scenario is generated for the specific attack, thus obtaining the adversarial perturbation scenario.

[0008] In conjunction with the first aspect, in one implementation, the step of performing a two-level fusion credibility calculation on the obtained message credibility score to generate an uncertainty set boundary includes: The message credibility score is used as a local credibility score; The global credibility score is obtained by fusing and calculating multiple local credibility scores. The global credibility score is converted into an uncertain set boundary in the control constraints using a preset mapping function.

[0009] In conjunction with the first aspect, in one implementation, the step of fusing and calculating multiple local credibility scores to obtain a global credibility score includes: Obtain the relative distance between the target vehicle and the current vehicle that provides each of the aforementioned local confidence scores; The consensus weight is calculated using a distance-weighted Bayesian consensus algorithm based on the relative distance. The global credibility score is obtained by weighting and summing the multiple local credibility scores using the consensus weights.

[0010] In conjunction with the first aspect, in one implementation, the step of performing robust control solution based on the adversarial disturbance scenario and the uncertainty set boundary through a preset rolling optimization strategy to obtain the final control command includes: Construct a robust prediction model that includes the aforementioned adversarial perturbation scenario; The boundary of the uncertainty set is used as the constraint parameter of the robust prediction model; The robust prediction model is solved using a rolling hot-start single-step approximation strategy to generate the optimal control sequence at the current moment. The current control quantity in the optimal control sequence is used as the final control command.

[0011] In conjunction with the first aspect, in one implementation, the step of solving the robust prediction model using a rolling hot-start single-step approximation strategy to generate the optimal control sequence at the current moment includes: Obtain the optimal control sequence obtained from the previous control cycle; The value of the optimal control sequence obtained from the previous control cycle is assigned to the optimization variable of the current control cycle; Based on the assigned optimization variables, a single-step gradient update is performed on the robust prediction model to obtain the solution result of the current control cycle and generate the optimal control sequence at the current moment.

[0012] In conjunction with the first aspect, in one implementation, the attack pattern vector is generated based on multi-sensor data collected from the target vehicle, including: Consistency verification was performed on the acquired sensor data of the same type based on the preset factor graph model, and the verification results were obtained. If the verification result contains an error value, then an attack pattern vector is generated based on the error value.

[0013] Secondly, embodiments of this application provide a special vehicle avoidance system, the special vehicle avoidance system comprising: The data and evaluation module is used to acquire attack pattern vectors and message credibility scores in real time. The attack pattern vectors are generated based on the multi-sensor data of the target vehicle, and the message credibility scores are generated based on the V2X communication messages sent by the target vehicle. The scene generation module is used to generate adversarial perturbation scenes based on the attack mode vector through an adversarial generation network. The boundary mapping module is used to perform two-level fusion credibility calculation based on the message credibility score, and map the fused credibility to the uncertainty set boundary in the control constraints. The control solution module is used to perform robust control solution based on the adversarial disturbance scenario and the uncertainty set boundary through a preset rolling optimization strategy, obtain the final control command, and generate an active avoidance control command for controlling the vehicle to avoid the target vehicle.

[0014] Thirdly, embodiments of this application provide a special vehicle avoidance device, the special vehicle avoidance device including a processor, a memory, and a special vehicle avoidance program stored in the memory and executable by the processor, wherein when the special vehicle avoidance program is executed by the processor, it implements the steps of the special vehicle avoidance method as described in any one of claims 1 to 7.

[0015] Fourthly, embodiments of this application provide a computer-readable storage medium storing a special vehicle avoidance program, wherein when the special vehicle avoidance program is executed by a processor, it implements the steps of the special vehicle avoidance method as described in any one of claims 1 to 7.

[0016] The beneficial effects of the technical solutions provided in this application include: By acquiring attack pattern vectors and message credibility scores in real time; generating adversarial disturbance scenarios based on the acquired attack pattern vectors and a pre-set adversarial generation network; and performing two-level fusion credibility calculation on the acquired message credibility scores to generate an uncertain set boundary; and obtaining the final control command based on the adversarial disturbance scenario, the uncertain set boundary, and a preset rolling optimization strategy, generating an active avoidance control command for controlling the vehicle to avoid the target vehicle; this enables closed-loop collaborative defense of perception, communication, and control, reduces the collision rate in mixed attack scenarios, significantly reduces computation time, enables complex robust control algorithms to run in real time on the vehicle chip, prevents remote malicious nodes from manipulating consensus results, and ensures that the generated avoidance trajectory conforms to the vehicle's dynamic characteristics. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the first embodiment of the special vehicle avoidance method of the present invention; Figure 2 This is a system architecture block diagram of the first embodiment of the special vehicle avoidance system of the present invention; Figure 3 This is a schematic diagram of the approximate single-step process of the rolling hot start method for avoiding special vehicles in this invention; Figure 4 This is a schematic diagram of the vehicle-road cooperative message credibility process in the special vehicle avoidance method of the present invention. Detailed Implementation

[0018] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0020] Firstly, embodiments of this application provide a method for avoiding obstacles by special vehicles.

[0021] In one embodiment, reference is made to Figure 1 , Figure 1 This is a schematic flowchart illustrating an embodiment of the special vehicle avoidance method of this application. Figure 1 As shown, the special vehicle avoidance method includes: real-time acquisition of attack pattern vectors and message credibility scores. The attack pattern vectors are generated based on multi-sensor data of the target vehicle, and the message credibility scores are generated based on vehicle-road cooperative communication messages received from the target vehicle; based on the acquired attack pattern vectors and a pre-set adversarial generation network, an adversarial disturbance scenario is generated, and a two-level fusion credibility calculation is performed on the acquired message credibility scores to generate an uncertainty set boundary; based on the adversarial disturbance scenario, the uncertainty set boundary, and a preset rolling optimization strategy, the final control command is obtained, generating an active avoidance control command for controlling the vehicle to avoid the target vehicle.

[0022] In this embodiment, during vehicle operation, the on-board controller collects multi-sensor data and V2X messages (vehicle-to-everything communication messages) in real time. The multi-sensor data includes camera images, LiDAR point clouds, and millimeter-wave radar data. The V2X messages include basic security messages sent by surrounding vehicles. In response to the completion of data collection, an attack pattern vector and a message credibility score are generated. If there is an error value in the verification results between sensor data, an attack pattern vector is generated based on the error value. The message credibility score is generated based on the received vehicle-to-everything communication messages.

[0023] Furthermore, in one embodiment, obtaining the attack pattern vector includes: performing consistency verification on the obtained similar sensor data according to a preset factor graph model to obtain a verification result; if there is an error value in the verification result, then generating an attack pattern vector based on the error value.

[0024] In this embodiment, the vehicle controller identifies the data consistency of the current sensors by constructing a factor graph model. As the vehicle moves, the sensors collect environmental data. Each sensor is associated with a corresponding observation factor. The vehicle controller communicates with the sensors to obtain the observation data associated with the sensor. The observation data includes one or more of the following: camera ranging and radar ranging. Camera ranging is defined as the obstacle distance observation value output by the vision algorithm, and radar ranging is defined as the distance observation value calculated by radar signal reflection.

[0025] Based on the verification results, the attack mode vector of the current sensor is determined. For each sensor, the corresponding verification result can be obtained. The vehicle controller calculates the residual between the observations of similar sensors. The input feature of the algorithm is the error value. By comparing the preset threshold, the attack mode vector is determined. The vehicle controller calculates the matching degree between the error value of the current sensor and the preset standard. The type with the highest matching degree is used as the attack mode vector of the current sensor.

[0026] Furthermore, in one embodiment, generating an adversarial perturbation scenario includes: inputting an attack pattern vector as a conditional input vector into a conditional Wasserstein generative adversarial network; identifying the attack type and attack intensity in the attack pattern vector through the conditional Wasserstein generative adversarial network; and generating an adversarial perturbation scenario against a specific attack based on the attack type and attack intensity, thereby obtaining the adversarial perturbation scenario.

[0027] In this embodiment, the vehicle controller obtains the adversarial disturbance scenario corresponding to the current attack mode vector through a pre-set generative model. The conditional Wasserstein generative adversarial network (GAN) is a neural network model stored inside the vehicle controller to input the attack mode vector into the pre-trained GAN. The GAN is a deep generative model whose structure includes an input layer, a generator, and a discriminator, and incorporates a gradient penalty term. The input layer receives the condition vector and random noise. The generator and discriminator contain multi-layer neural network structures. The model training process includes: collecting a large amount of sample data of attack scenarios, each sample containing input conditions and labels; training using the Wasserstein distance loss function and gradient penalty term; stopping training when the quality of the validation set reaches a preset standard; the vehicle controller preprocesses the acquired current attack mode vector and inputs it into the pre-trained GAN for forward computation. The GAN outputs the adversarial disturbance scenario, and the vehicle controller selects this scenario as the disturbance term for the current control input.

[0028] It should be noted that the aforementioned multi-sensor data and vehicle-road cooperative communication messages were all obtained with authorization.

[0029] Furthermore, in one embodiment, the message credibility score is generated based on the vehicle-road cooperative communication message received from the target vehicle. The message credibility score adopts a reputation assessment mechanism, and the on-board controller reads the message credibility score signal.

[0030] The boundary of the uncertain set is collected through a boundary mapping module, which is a credibility assessment engine installed inside the vehicle controller. The vehicle controller detects the message credibility score in each sampling period. To calculate the boundary of the uncertain set, the vehicle controller adopts a two-level fusion credibility calculation method: a fusion window is set, the local credibility scores of multiple neighboring vehicles within the window are recorded, and then a weighted sum is performed to obtain the current global credibility score.

[0031] The uncertainty set boundary is acquired through a mapping function, which can be either a linear or nonlinear function. The vehicle controller receives the signal from the boundary mapping module to obtain the uncertainty set boundary value. Under attack conditions, the uncertainty set boundary corresponds to the attack intensity, and the accuracy of the boundary mapping function meets the control stability requirements.

[0032] Furthermore, in one embodiment, reference is made to Figure 4 , Figure 4 This is a schematic diagram of the vehicle-road cooperative message credibility process in the special vehicle avoidance method of this application. The generation of the uncertainty set boundary includes: taking the message credibility score as a local credibility score; fusing and calculating multiple local credibility scores to obtain a global credibility score; and converting the global credibility score into the uncertainty set boundary in the control constraints through a preset mapping function.

[0033] It should be noted that the preset mapping function is used to transform the global credibility score into the uncertainty set boundary. Those skilled in the art will understand that the specific form of the mapping function is set according to actual control requirements, and its purpose is to ensure that the size of the uncertainty set boundary reflects the credibility of the message.

[0034] In this embodiment, the vehicle controller obtains the relative distance between the target vehicle providing each local credibility score and the vehicle itself. Based on the relative distance, a distance-weighted Bayesian consensus algorithm is used to calculate the consensus weight. The consensus weight is then used to perform a weighted summation of multiple local credibility scores to obtain the global credibility score.

[0035] Specifically, the second level employs distance-weighted Bayesian consensus and uses a two-level fusion credibility calculation. The first level calculates the initial local credibility score of the target vehicle. The second-level calculation is based on a credibility score derived from multi-vehicle consensus. For example, the second level uses distance-weighted Bayesian consensus, with six vehicles within the communication range providing auxiliary observations. The onboard controller determines the relative distance between each target vehicle and its own vehicle. Calculate weights And combined with the observation results of each vehicle and logarithmic probability Perform Bayesian consensus calculation:

[0036] for (The news is true):

[0037]

[0038]

[0039] for (The information is false):

[0040]

[0041]

[0042] Normalization reliability:

[0043] Overall credibility :

[0044] determination: The V2X message from V1 is a suspicious message; in, The initial local confidence score of the target vehicle obtained from the first-level fusion. , This is the credibility threshold.

[0045] In summary, once the onboard controller determines the attack mode vector and message credibility score of the current vehicle, it can query the matching adjustment strategy within the robust control strategy based on the combination of adversarial disturbance scenarios and uncertainty set boundaries, and generate corresponding control parameters. These parameters are then sent to the vehicle control system for execution via the bus. The robust control strategy is a multi-dimensional mapping table, with the attack mode and credibility level as input dimensions and control parameters as output dimensions. For example, when the attack mode is normal and the credibility is high, the onboard controller calls the normal control mode, setting the uncertainty set boundary to the minimum value. When the attack mode is in an attack state and the credibility is low, the onboard controller calls the strong robust control mode, setting the uncertainty set boundary to the maximum value to maximize the safety margin.

[0046] In this embodiment, by acquiring the attack pattern vector and message credibility score, and executing a matching avoidance control strategy, the robustness and real-time adaptive adjustment of the avoidance control are achieved. Through this embodiment, the dynamic changes of sensor attacks and the changes in the credibility of communication messages can be taken into account simultaneously, so that the control system can provide the most suitable control output characteristics in different scenarios, thereby improving the overall balance between safety, smoothness and responsiveness. By using the environmental feature of attack pattern and the communication feature of credibility score as the adjustment basis, the control strategy can more accurately meet the actual needs, avoiding the problems of control failure or poor driving experience caused by a single strategy.

[0047] Furthermore, in one embodiment, the final control command is obtained based on the adversarial disturbance scenario, the uncertainty set boundary, and a preset rolling optimization strategy, including: constructing a robust prediction model containing the adversarial disturbance scenario; using the uncertainty set boundary as the constraint parameter of the robust prediction model; solving the robust prediction model using a rolling hot-start single-step approximation strategy to generate the optimal control sequence at the current moment; and using the control quantity at the current moment in the optimal control sequence as the final control command.

[0048] In this embodiment, a rolling hot-start single-step approximation strategy is adopted for solving the robust prediction model, and a speed-adaptive prediction time domain is used. For example, when the speed is 16.67 m / s, the number of prediction steps is... Coverage distance m, using velocity-dependent curvature constraints, for example... .

[0049] Furthermore, in one embodiment, reference is made to Figure 3 , Figure 3 This is a schematic diagram of the rolling hot-start single-step approximation process in the special vehicle avoidance method of this application. The rolling hot-start single-step approximation strategy is used to solve the robust prediction model and generate the optimal control sequence at the current moment. The process includes: obtaining the optimal control sequence obtained from the previous control cycle; assigning the values ​​of the optimal control sequence obtained from the previous control cycle to the optimization variables of the current control cycle; performing single-step gradient update on the robust prediction model based on the assigned optimization variables to obtain the solution result of the current control cycle and generate the optimal control sequence at the current moment.

[0050] In this embodiment, the preset speed and the preset boundary are set based on actual needs. This embodiment further limits the conservative control mode to be activated under normal operating conditions when the vehicle speed is too high and / or the uncertainty set boundary is too large. By setting a speed threshold, it is ensured that conservative control only intervenes when the vehicle needs higher safety output. By setting a boundary threshold, it is ensured that the safety margin is not too low, thus avoiding the impact on the reliability of subsequent control drive due to the exhaustion of the margin.

[0051] Furthermore, in one embodiment, a single-step gradient update is performed on the robust prediction model based on the assigned optimized variables to obtain the solution result for the current control cycle. This includes: if the gradient update satisfies the convergence condition, then the initialized variables are accepted as the optimal solution; if the gradient update does not satisfy the convergence condition, then a single-step update is performed, for example, the perturbation from the previous frame. offset of V1 position m, perform a 1-step projective gradient descent (PGD) update: m, solving the control under the worst-case disturbance , The gradient of the cost function with respect to the perturbation variable is used to indicate the direction of perturbation updates.

[0052] In this embodiment, for the combination of attack type and confidence level of medium or low confidence, the matched control strategy is a robust control strategy. Under the robust control strategy, a robust model is adopted, and the control response time needs to be set to a preset response time based on the attack mode.

[0053] For the combination of severe attack types and medium or low confidence levels, a robust control strategy is matched. Under this strategy, a model predictive control (MPC) algorithm is used, driven by exponential collision cost. For example, in a real-world scenario, V1 is still to the left front, and the LiDAR detects... m (lane width) m, where This refers to the lateral deviation distance between the vehicle and the obstacle detected by the lidar. For a safe distance, To buffer the distance, collision cost is activated. , This is the collision cost function value, used to quantify collision risk, due to the extremely low reliability of V2X messages ( MPC identifies V1 as a high-uncertainty obstacle, increases the safety margin, and chooses to slow down at the current position and wait for confirmation of V1's actual position before changing lanes, rather than changing lanes directly to the left based on false V2X messages.

[0054] Secondly, embodiments of this application also provide a special vehicle avoidance system.

[0055] In one embodiment, reference is made to Figure 2 , Figure 2 This is a functional module diagram of an embodiment of the special vehicle avoidance system of this application. Figure 2 As shown, the special vehicle avoidance system includes: a data and evaluation module for real-time acquisition of attack mode vectors and message credibility scores; a scenario generation module for generating adversarial disturbance scenarios based on attack mode vectors using an adversarial generative network; a boundary mapping module for performing two-level fusion credibility calculation based on message credibility scores, mapping the fused credibility to the uncertainty set boundary in the control constraints; and a control solution module for performing robust control solution based on the adversarial disturbance scenario and the uncertainty set boundary using a preset rolling optimization strategy to obtain the final control command and generate active avoidance control commands for controlling the vehicle to avoid the target vehicle.

[0056] Furthermore, in one embodiment, the attack pattern vector is generated based on multi-sensor data of the target vehicle, the message credibility score is generated based on the vehicle-road cooperative communication messages received from the target vehicle, and the data and evaluation module is integrated into the vehicle controller.

[0057] Furthermore, in one embodiment, the boundary mapping module is used to treat the message credibility score as a local credibility score; to perform fusion calculation on multiple local credibility scores to obtain a global credibility score; and to convert the global credibility score into an uncertain set boundary in the control constraints through a preset mapping function.

[0058] Furthermore, in one embodiment, the data and evaluation module is used to perform consistency verification on the acquired similar sensor data according to a preset factor graph model to obtain a verification result; if there is an error value in the verification result, an attack mode vector is generated based on the error value.

[0059] Furthermore, in one embodiment, the scene generation module is used to input the attack pattern vector as a conditional input vector into a conditional Wasserstein generative adversarial network; identify the attack type and attack intensity in the attack pattern vector through the conditional Wasserstein generative adversarial network; and generate an adversarial perturbation scene against a specific attack based on the attack type and attack intensity, thus obtaining the adversarial perturbation scene. The conditional Wasserstein generative adversarial network is a neural network model stored inside the vehicle controller.

[0060] Furthermore, in one embodiment, the control solution module is used to construct a robust prediction model that includes a scenario against disturbances; use the uncertainty set boundary as the constraint parameter of the robust prediction model; use a rolling hot start single-step approximation strategy to solve the robust prediction model and generate the optimal control sequence at the current time; and use the control quantity at the current time in the optimal control sequence as the final control command.

[0061] Furthermore, in one embodiment, the control solving module is used to obtain the optimal control sequence obtained from the previous control cycle; assign the value of the optimal control sequence obtained from the previous control cycle to the optimization variable of the current control cycle; perform single-step gradient update on the robust prediction model based on the assigned optimization variable to obtain the solution result of the current control cycle and generate the optimal control sequence at the current moment.

[0062] The functions of each module in the above-mentioned special vehicle avoidance system correspond to the steps in the above-mentioned special vehicle avoidance method embodiment, and their functions and implementation processes will not be described in detail here.

[0063] Thirdly, embodiments of this application provide a special vehicle avoidance device, which is a device with data processing functions such as a vehicle controller or a vehicle power domain controller.

[0064] In this embodiment, the special vehicle avoidance device includes a processor, a memory, a communication interface, and a communication bus.

[0065] The communication bus can be of any type and is used to interconnect the processor, memory, and communication interface.

[0066] The communication interface includes input / output interfaces, physical interfaces, and logical interfaces used to interconnect components within the special vehicle obstacle avoidance device, as well as interfaces used to interconnect the special vehicle obstacle avoidance device with other devices. Physical interfaces include Ethernet interfaces, fiber optic interfaces, etc.; user equipment includes displays, keyboards, etc.

[0067] Memory refers to various types of storage media, such as random access memory, read-only memory, non-volatile memory, flash memory, optical storage, hard disk, etc.

[0068] The processor is a general-purpose processor. The general-purpose processor calls a special vehicle avoidance program stored in memory and executes the special vehicle avoidance method provided in the embodiments of this application. For example, the general-purpose processor is a central processing unit. The method executed when the special vehicle avoidance program is called can be referred to in various embodiments of the special vehicle avoidance method of this application.

[0069] Those skilled in the art will understand that the hardware structure shown above does not constitute a limitation of this application, including more or fewer components than shown, or combinations of certain components, or different component arrangements.

[0070] Fourthly, embodiments of this application also provide a computer-readable storage medium.

[0071] The present application provides a special vehicle avoidance program stored on a computer-readable storage medium, wherein when the special vehicle avoidance program is executed by a processor, it implements the steps of the special vehicle avoidance method described above.

[0072] The method implemented when the special vehicle avoidance procedure is executed can be referred to in various embodiments of the special vehicle avoidance method of this application.

[0073] The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application 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 such process, method, product, or apparatus. The terms "first," "second," and "third," etc., are used to distinguish different objects, etc., and do not indicate a sequence, nor do they limit "first," "second," and "third" to different types.

[0074] In the description of the embodiments of this application, terms such as "exemplary," "for example," or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplary," "for example," or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary," "for example," or "for instance" is intended to present the relevant concepts in a concrete manner.

[0075] In the description of the embodiments of this application, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. The "and / or" in the text is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone. In addition, in the description of the embodiments of this application, "multiple" means two or more.

[0076] In some processes described in the embodiments of this application, multiple operations or steps are included in a specific order. However, it should be understood that these operations or steps may not be executed in the order they appear in the embodiments of this application, or they may be executed in parallel. The sequence number of the operation is only used to distinguish different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed sequentially or in parallel, and these operations or steps may be combined.

[0077] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device to execute the methods described in the various embodiments of this application.

[0078] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for avoiding obstacles on special vehicles, characterized in that, The methods for avoiding special vehicles include: The attack pattern vector and message credibility score are acquired in real time. The attack pattern vector is generated based on the multi-sensor data of the target vehicle, and the message credibility score is generated based on the V2X communication messages sent by the target vehicle. Based on the obtained attack pattern vectors and the pre-set adversarial generation network, adversarial perturbation scenarios are generated, and the credibility scores of the obtained messages are fused in two levels to calculate the credibility and generate the uncertainty set boundary. Based on the aforementioned adversarial disturbance scenario, the uncertainty set boundary, and the preset rolling optimization strategy, the final control command is obtained, generating an active avoidance control command for controlling the vehicle to avoid the target vehicle.

2. The special vehicle avoidance method as described in claim 1, characterized in that, The step of generating adversarial perturbation scenarios based on the attack pattern vector through an adversarial generative network includes: The attack pattern vector is input as a conditional input vector into the conditional Wasserstein generative adversarial network. The conditional Wasserstein generative adversarial network identifies the attack type and attack intensity in the attack pattern vector. Based on the attack type and attack intensity, an adversarial perturbation scenario is generated for the specific attack, thus obtaining the adversarial perturbation scenario.

3. The special vehicle avoidance method as described in claim 1, characterized in that, The step of performing a two-level fusion credibility calculation on the obtained message credibility score to generate an uncertainty set boundary includes: The message credibility score is used as a local credibility score; The global credibility score is obtained by fusing and calculating multiple local credibility scores. The global credibility score is converted into an uncertain set boundary in the control constraints using a preset mapping function.

4. The special vehicle avoidance method as described in claim 3, characterized in that, The process of fusing multiple local credibility scores to obtain a global credibility score includes: Obtain the relative distance between the target vehicle and the current vehicle that provides each of the aforementioned local confidence scores; The consensus weight is calculated using a distance-weighted Bayesian consensus algorithm based on the relative distance. The global credibility score is obtained by weighting and summing the multiple local credibility scores using the consensus weights.

5. The special vehicle avoidance method as described in claim 1, characterized in that, The robust control solution, based on the adversarial disturbance scenario and the uncertainty set boundary, is obtained through a preset rolling optimization strategy to obtain the final control command, including: Construct a robust prediction model that includes the aforementioned adversarial perturbation scenario; The boundary of the uncertainty set is used as the constraint parameter of the robust prediction model; The robust prediction model is solved using a rolling hot-start single-step approximation strategy to generate the optimal control sequence at the current moment. The current control quantity in the optimal control sequence is used as the final control command.

6. The special vehicle avoidance method as described in claim 5, characterized in that, The step of solving the robust prediction model using a rolling hot-start single-step approximation strategy to generate the optimal control sequence at the current moment includes: Obtain the optimal control sequence obtained from the previous control cycle; The value of the optimal control sequence obtained from the previous control cycle is assigned to the optimization variable of the current control cycle; Based on the assigned optimization variables, a single-step gradient update is performed on the robust prediction model to obtain the solution result of the current control cycle and generate the optimal control sequence at the current moment.

7. The special vehicle avoidance method as described in claim 1, characterized in that, The attack pattern vector is generated based on multi-sensor data collected from the target vehicle, including: Consistency verification was performed on the acquired sensor data of the same type based on the preset factor graph model, and the verification results were obtained. If the verification result contains an error value, then an attack pattern vector is generated based on the error value.

8. A special vehicle avoidance system, characterized in that, The special vehicle avoidance system includes: The data and evaluation module is used to acquire attack pattern vectors and message credibility scores in real time. The attack pattern vectors are generated based on the multi-sensor data of the target vehicle, and the message credibility scores are generated based on the V2X communication messages sent by the target vehicle. The scene generation module is used to generate adversarial perturbation scenes based on the attack mode vector through an adversarial generation network. The boundary mapping module is used to perform two-level fusion credibility calculation based on the message credibility score, and map the fused credibility to the uncertainty set boundary in the control constraints. The control solution module is used to perform robust control solution based on the adversarial disturbance scenario and the uncertainty set boundary through a preset rolling optimization strategy, obtain the final control command, and generate an active avoidance control command for controlling the vehicle to avoid the target vehicle.

9. A special vehicle avoidance device, characterized in that, The special vehicle avoidance device includes a processor, a memory, and a special vehicle avoidance program stored in the memory and executable by the processor, wherein when the special vehicle avoidance program is executed by the processor, it implements the steps of the special vehicle avoidance method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a special vehicle avoidance program, wherein when the special vehicle avoidance program is executed by a processor, it implements the steps of the special vehicle avoidance method as described in any one of claims 1 to 7.

Images