Unmanned system trajectory prediction credibility evaluation method and device based on multi-source data fusion

Abstract

The embodiment of the invention discloses an unmanned system trajectory prediction credibility evaluation method and device based on multi-source data fusion. A specific embodiment of the method comprises the steps of constructing a trajectory prediction data set, and performing model training on a trajectory prediction model to be attacked according to the trajectory prediction data set to obtain a target model; generating a confrontation sample set; and according to the adversarial sample set and the real tracks corresponding to the adversarial samples in the adversarial sample set, carrying out credibility evaluation on the target model to generate a credibility evaluation result. According to the embodiment, the credibility of the trajectory prediction model of the unmanned system in a complex environment can be effectively detected and evaluated, and it is ensured that the unmanned system can still make reasonable decisions under severe conditions.

Description

Technical Field

[0001] The embodiments disclosed herein relate to the field of model credibility assessment, specifically to a method and apparatus for credibility assessment of unmanned system trajectory prediction based on multi-source data fusion. Background Technology

[0002] With the deepening application of artificial intelligence in unmanned systems, trajectory prediction technology has become a core component of unmanned systems such as autonomous driving and intelligent security. Meanwhile, deep learning-based trajectory prediction models (such as LSTM (Long Short-Term Memory) models) have achieved significant breakthroughs in accuracy. However, due to the inherent uninterpretability of deep neural networks, significant safety risks have arisen. For example, researchers at the University of Michigan caused severe deviations in the predicted trajectory by adding perturbations that are difficult for humans to detect to the input trajectory of the trajectory prediction model. Therefore, assessing the reliability of the model is crucial for determining its robustness against perturbations. Summary of the Invention

[0003] The content of this disclosure is used to briefly introduce concepts that will be described in detail in the detailed description section below. The content of this disclosure is not intended to identify key features or essential features of the claimed technical solution, nor is it intended to limit the scope of the claimed technical solution.

[0004] Some embodiments of this disclosure propose a reliable evaluation method and apparatus for trajectory prediction of unmanned systems based on multi-source data fusion, in order to solve the technical problems mentioned in the background section above.

[0005] In a first aspect, some embodiments of this disclosure provide a reliability assessment method for trajectory prediction of unmanned systems based on multi-source data fusion, comprising: constructing a trajectory prediction dataset, and training a target model based on the trajectory prediction dataset to obtain a trajectory prediction model to be attacked, wherein the trajectory prediction data in the trajectory prediction dataset includes the historical trajectory and predicted trajectory corresponding to the agent; generating an adversarial sample set, wherein the adversarial samples in the adversarial sample set are generated through the following steps: perturbing candidate trajectories to obtain perturbed candidate trajectories, wherein the candidate trajectories are the historical trajectories corresponding to the agent; determining a reference adversarial sample set based on the perturbed candidate trajectories, the target model, and a distance-based attention-guided loss function, wherein the reference adversarial samples include: reference adversarial trajectories and predicted reference trajectories; selecting reference adversarial samples that meet the selection criteria from the reference adversarial sample set as adversarial samples, wherein the selection criteria are: the average displacement error corresponding to the reference adversarial sample is the largest; and performing a reliability assessment of the target model based on the adversarial sample set and the real trajectories corresponding to the adversarial samples in the adversarial sample set to generate a reliability assessment result.

[0006] Secondly, some embodiments of this disclosure provide a reliability assessment device for trajectory prediction of unmanned systems based on multi-source data fusion. The device includes: a construction unit configured to construct a trajectory prediction dataset and train a trajectory prediction model to be attacked based on the trajectory prediction dataset to obtain a target model, wherein the trajectory prediction data in the trajectory prediction dataset includes historical trajectories and predicted trajectories corresponding to the agent; a generation unit configured to generate an adversarial sample set, wherein the adversarial samples in the adversarial sample set are generated through the following steps: perturbating candidate trajectories to obtain perturbated candidate trajectories, wherein the candidate trajectories are historical trajectories corresponding to the agent; determining a reference adversarial sample set based on the perturbated candidate trajectories, the target model, and a distance-based attention-guided loss function, wherein the reference adversarial samples include: reference adversarial trajectories and predicted reference trajectories; selecting reference adversarial samples that meet the selection criteria from the reference adversarial sample set as adversarial samples, wherein the selection criteria are: the average displacement error corresponding to the reference adversarial sample is the largest; and a reliability assessment unit configured to perform a reliability assessment on the target model based on the adversarial sample set and the real trajectories corresponding to the adversarial samples in the adversarial sample set to generate a reliability assessment result.

[0007] In a third aspect, some embodiments of the present disclosure provide an electronic device comprising: one or more processors; a storage device on which one or more programs are stored, and when the one or more programs are executed by one or more processors, the one or more processors implement the method described in any implementation of the first aspect above.

[0008] In a fourth aspect, some embodiments of the present disclosure provide a computer-readable medium having a computer program stored thereon, wherein when the program is executed by a processor, the method described in any implementation of the first aspect is implemented.

[0009] The various embodiments disclosed herein have the following beneficial effects: The reliability assessment method for unmanned system trajectory prediction based on multi-source data fusion, as described in some embodiments of this disclosure, firstly, combines the physical characteristics of trajectory data with the features of the deep model and employs an improved gradient attack algorithm for evaluation. Secondly, a "physical inertia-preserving" gradient perturbation algorithm is designed, utilizing the momentum mechanism during gradient updates to ensure that the generated adversarial trajectory conforms to the laws of physical motion, generating minimal perturbation. Furthermore, an attention-guided loss function is used to focus the attack on points that the model focuses on when predicting the trajectory, such as the trajectory endpoint, improving the targeting and effectiveness of the assessment. In summary, this invention is applicable to application scenarios such as unmanned vehicles, drone swarms, and unmanned reconnaissance vehicles, effectively detecting and evaluating the reliability of unmanned system trajectory prediction models in complex environments, ensuring that they can still make reasonable decisions under adverse conditions. Attached Figure Description

[0010] The above and other features, advantages, and aspects of the various embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and that components and elements are not necessarily drawn to scale.

[0011] Figure 1 This is a flowchart of some embodiments of the reliability assessment method for trajectory prediction of unmanned systems based on multi-source data fusion according to the present disclosure;

[0012] Figure 2 This is a schematic diagram of the structure of some embodiments of the unmanned system trajectory prediction reliability assessment device based on multi-source data fusion according to the present disclosure;

[0013] Figure 3 It is a structural diagram of an electronic device suitable for implementing some embodiments of the present disclosure. Detailed Implementation

[0014] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments described herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are for illustrative purposes only and are not intended to limit the scope of protection of the present disclosure.

[0015] It should also be noted that, for ease of description, only the parts related to the invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of the present disclosure may be combined with each other.

[0016] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0017] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0018] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0019] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0020] refer to Figure 1 The flowchart 100 illustrates some embodiments of the reliability assessment method for trajectory prediction of unmanned systems based on multi-source data fusion according to this disclosure. This reliability assessment method for trajectory prediction of unmanned systems based on multi-source data fusion includes the following steps:

[0021] Step 101: Construct a trajectory prediction dataset and train the trajectory prediction model to be attacked based on the trajectory prediction dataset to obtain the target model.

[0022] In some embodiments, the implementing entity (e.g., a computing device) of the unmanned system trajectory prediction credibility assessment method based on multi-source data fusion can construct a trajectory prediction dataset and train a trajectory prediction model to be attacked based on the trajectory prediction dataset to obtain the target model.

[0023] The trajectory prediction dataset contains trajectory prediction data for intelligent agents. For example, the Apolloscape dataset could be used. An intelligent agent is a proxy capable of environmental perception and executing corresponding actions to achieve a predetermined goal. In the field of autonomous driving, for instance, an intelligent agent can be a proxy simulating vehicle driving. Specifically, the trajectory prediction dataset includes the historical trajectory and predicted trajectory of the intelligent agent. The trajectory data (historical trajectory or predicted trajectory) can be in the form of triplets, such as <agent ID, trajectory node ID of the intelligent agent, coordinates of the trajectory node in two-dimensional space>. For time frames (or "trajectory nodes")... The number of agents observed and received by the trajectory prediction model can be... The trajectory length of the historical trajectories received by the trajectory prediction model can be... (For example, in a trajectory prediction dataset, the value is 6). The trajectory length of the predicted trajectory can be... (For example, in a trajectory prediction dataset, the value is 6). Agent The two-dimensional spatial representation (two-dimensional spatial coordinates) can be Therefore, the historical trajectories of each agent corresponding to the trajectory prediction dataset can be represented as:

[0024] ;

[0025] in, Indicates the first Each agent in a time frame The historical trajectories included in the trajectory prediction data corresponding to the time. Indicates the first The two-dimensional spatial coordinates of an intelligent agent on its historical trajectory.

[0026] The predicted trajectories of each agent corresponding to the trajectory prediction dataset can be represented as:

[0027] ;

[0028] in, Indicates the first Each agent in a time frame The predicted trajectory is included in the trajectory prediction data corresponding to the time. Indicates the first The two-dimensional spatial coordinates of an agent on the predicted trajectory.

[0029] The actual trajectories corresponding to the predicted trajectories in each trajectory prediction dataset can be represented as follows:

[0030] ;

[0031] in, Indicates the first Each agent in a time frame The actual trajectory corresponding to the time. Indicates the first The agent's two-dimensional spatial coordinates on the real trajectory. Specifically, the trajectory collected for the agent is divided into two parts: the first half serves as the corresponding historical trajectory, and the second half as the corresponding real trajectory. The real trajectory is used as sample labels during the training of the trajectory prediction model.

[0032] The trajectory prediction model to be attacked can be any trajectory prediction model used for trajectory prediction and subject to credibility evaluation. For example, a conventional trajectory prediction model can be used, such as:

[0033] FQA model ( https: / / arxiv.org / abs / 2010.15891 );

[0034] GRIP++ model ( https: / / arxiv.org / abs / 1907.07792 );

[0035] Trajectron++ model ( https: / / arxiv.org / abs / 2001.03093 The system uses historical trajectories as the trajectory prediction model to be attacked, and obtains the corresponding model configuration files. It then employs supervised training, using historical trajectories as model input, real trajectories as training samples, and predicted trajectories as model output to train the model, obtaining trained samples which serve as the target model.

[0036] It should be noted that the aforementioned computing devices can be either hardware or software. When the computing device is hardware, it can be implemented as a distributed cluster consisting of multiple servers or terminal devices, or as a single server or a single terminal device. When the computing device is software, it can be installed on the hardware devices listed above. It can be implemented as, for example, multiple software programs or software modules used to provide distributed services, or as a single software program or software module. No specific limitations are made here.

[0037] Step 102: Generate a set of adversarial examples.

[0038] In some embodiments, the aforementioned executing entity can generate an adversarial sample set. The adversarial samples in the adversarial sample set are generated through the following steps:

[0039] Step S1: Add perturbations to the candidate trajectories to obtain perturbation-added candidate trajectories, where the candidate trajectories are the historical trajectories corresponding to the agent.

[0040] Among them, the candidate trajectory is the historical trajectory corresponding to the agent.

[0041] In practice, the historical trajectory included in the trajectory prediction data of any agent can be randomly selected from the trajectory prediction dataset as the trajectory to which the perturbation is added.

[0042] As an example, select the first The historical trajectory corresponding to each agent is used as the candidate trajectory. For candidate trajectories Add perturbation Then the disturbance can be expressed as:

[0043] .

[0044] Candidate trajectories after perturbation It can be represented as:

[0045] .

[0046] Predicted trajectory corresponding to candidate trajectory after perturbation is added It can be represented as:

[0047] .

[0048] Optionally, the perturbation added within the candidate trajectory is based on an adaptive step size search of the corresponding perturbation space, where the step size is... The corresponding step size is determined by the following formula:

[0049] ;

[0050] in, Indicates the number of iterations. Indicates the first The step size corresponding to the next iteration. Indicates the first The step size corresponding to the next iteration. Indicates the first The loss value corresponding to the next iteration. Indicates the first The loss value corresponding to the next iteration. Indicates the preset loss threshold

[0051] Step S2: Determine the reference adversarial sample set based on the candidate trajectory added after the perturbation, the target model, and the distance-based attention-guided loss function.

[0052] The reference adversarial samples include: reference adversarial trajectories and predicted reference trajectories.

[0053] Optionally, the distance-based attention-guided loss function is characterized by the following formula:

[0054] ;

[0055] in, Represents the loss function. Indicates a disturbance. Represents the weight matrix. Represents the trace of a matrix. This represents the transpose of the weight matrix. This represents the predicted trajectory corresponding to the candidate trajectory. This represents the perturbation corresponding to the candidate trajectory. This represents the trade-off parameters corresponding to the disturbance. This represents the tradeoff parameters corresponding to the weight matrix, where, and The values ​​are all 0.1. The order of the norm. ,in, Represents intelligent agents Time frame in the weight matrix The corresponding weights.

[0056] In practice, the generated adversarial examples should mislead the trajectory prediction model (target model) as much as possible, causing the output predicted trajectory to deviate from the original true trajectory. Since the trajectory prediction model (target model) typically contains sequence modules, which usually focus on later trajectory points, this disclosure proposes the above loss function to add adaptive attention weights to each trajectory point, ensuring the attack focuses on the trajectory points that the trajectory prediction model (target model) is interested in. Specifically, and The values ​​are all 0.1, which aims to find smaller perturbations and reduce the influence of the weight matrix on the selection process.

[0057] Optionally, the loss value corresponding to the distance-based attention-guided loss function is solved using the stochastic projection gradient descent algorithm, as shown in the following formula:

[0058] ;

[0059] in, Indicates the number of iterations. This represents the Bruch's sphere to be projected. This indicates the step size corresponding to the current iteration number. Indicates the process The perturbation in the next iteration adds candidate trajectories. Indicates the process The perturbation in the next iteration adds candidate trajectories. express The magnitude of the gradient relative to the loss function at the next iteration.

[0060] in, The gradient magnitude with respect to the loss function at the next iteration Characterized by the following formula:

[0061] ;

[0062] in, This indicates the proportion of gradients retained from the previous iteration. express The magnitude of the gradient relative to the loss function at the next iteration. express The corresponding gradient, This represents the target model (trajectory prediction model). Indicates the first The actual trajectory corresponding to each intelligent agent.

[0063] In practice, unlike the traditional projective gradient descent algorithm, this disclosure adopts the momentum approach when solving the gradient, retaining a portion of the gradient values ​​generated in the previous iteration to simulate the "physical inertia" of the trajectory coordinates caused by the perturbation update process. The aim is to generate more covert and realistic adversarial examples.

[0064] In particular, the perturbation corresponding to the distance-based attention-guided loss function Less than or equal to the maximum acceptable coordinate change Specifically, the following constraints must be satisfied:

[0065] .

[0066] Through iterative random projection gradient descent algorithm Next, the reference adversarial trajectories included in the reference adversarial samples are obtained. and predicted reference trajectory .

[0067] In practice, by repeating steps S1 to S2 a preset number of times ( (This is repeated several times) to obtain a set of reference adversarial samples.

[0068] Step S3: Select reference adversarial samples that meet the selection criteria from the set of reference adversarial samples, and use them as adversarial samples.

[0069] The selection criterion is that the average displacement error corresponding to the reference adversarial sample is the largest.

[0070] In practice, the average displacement error is expressed by the following formula:

[0071] .

[0072] Step 103: Based on the adversarial sample set and the real trajectories corresponding to the adversarial samples in the adversarial sample set, perform a credibility assessment on the target model to generate a credibility assessment result.

[0073] In some embodiments, the aforementioned execution entity can perform a credibility assessment of the target model based on the adversarial sample set and the real trajectories corresponding to the adversarial samples in the adversarial sample set, so as to generate a credibility assessment result.

[0074] The aforementioned reliability assessment results include: average displacement error and robustness evaluation, with the robustness evaluation controlled by deviation probability. The average displacement error is calculated using the formula corresponding to the average displacement error in step S3 of step 102. For the average displacement error, the larger the value, the greater the difference between the trajectory coordinates predicted by the trajectory prediction model and the original predicted trajectory, and the greater the impact on the adversarial sample misleading model. The higher the deviation probability, the lower the robustness evaluation (specifically, the average robustness = ...). ,in, The deviation probability (denoted as ) represents the probability that the difference between the predicted reference trajectory and the corresponding real trajectory, which includes reference adversarial examples, is greater than a preset difference. Specifically, the smaller the average displacement error, the higher the robustness score, and the more reliable the model.

[0075] As an example, in the field of autonomous vehicle trajectory prediction, a preset difference can represent the width of half a lane (e.g., 1.8 meters), meaning there is a time frame (or "trajectory node"). ,have In (Two-dimensional spatial coordinates) and the actual trajectory In The difference between (two-dimensional spatial coordinates) is less than a preset difference, specifically represented by the following formula:

[0076] .

[0077] The various embodiments disclosed herein have the following beneficial effects: The reliability assessment method for unmanned system trajectory prediction based on multi-source data fusion, as described in some embodiments of this disclosure, firstly, combines the physical characteristics of trajectory data with the features of the deep model and employs an improved gradient attack algorithm for evaluation. Secondly, a "physical inertia-preserving" gradient perturbation algorithm is designed, utilizing the momentum mechanism during gradient updates to ensure that the generated adversarial trajectory conforms to the laws of physical motion, generating minimal perturbation. Furthermore, an attention-guided loss function is used to focus the attack on points that the model focuses on when predicting the trajectory, such as the trajectory endpoint, improving the targeting and effectiveness of the assessment. In summary, this invention is applicable to application scenarios such as unmanned vehicles, drone swarms, and unmanned reconnaissance vehicles, effectively detecting and evaluating the reliability of unmanned system trajectory prediction models in complex environments, ensuring that they can still make reasonable decisions under adverse conditions.

[0078] Further reference Figure 2 As an implementation of the methods shown in the above figures, this disclosure provides some embodiments of a reliable assessment device for trajectory prediction of unmanned systems based on multi-source data fusion. These device embodiments are similar to... Figure 1Corresponding to the method embodiments shown, this unmanned system trajectory prediction reliability assessment device based on multi-source data fusion can be specifically applied to various electronic devices.

[0079] like Figure 2 As shown, a reliability assessment device 200 for unmanned system trajectory prediction based on multi-source data fusion in some embodiments includes: a construction unit 201, a generation unit 202, and a reliability assessment unit 203. The construction unit 201 is configured to construct a trajectory prediction dataset and train a target model based on the trajectory prediction dataset to obtain a target model. The trajectory prediction dataset contains historical trajectories and predicted trajectories corresponding to the agent. The generation unit 202 is configured to generate an adversarial sample set. The adversarial samples in the adversarial sample set are generated through the following steps: adding perturbations to candidate trajectories to obtain perturbation samples. Add candidate trajectories, where the candidate trajectories are the historical trajectories corresponding to the agent; determine a set of reference adversarial samples based on the added candidate trajectories, the target model, and a distance-based attention-guided loss function, where the reference adversarial samples include: reference adversarial trajectories and predicted reference trajectories; select reference adversarial samples that meet the selection criteria from the set of reference adversarial samples as adversarial samples, where the selection criteria are: the average displacement error corresponding to the reference adversarial sample is the largest; the credibility evaluation unit 203 is configured to perform credibility evaluation on the target model based on the set of adversarial samples and the real trajectories corresponding to the adversarial samples in the set of adversarial samples, so as to generate credibility evaluation results.

[0080] It is understandable that the various units and references recorded in the multi-source data fusion-based unmanned system trajectory prediction reliability assessment device 200 are related to... Figure 1 The steps described in the method correspond to each other. Therefore, the operations, features, and beneficial effects described above for the method are also applicable to the reliability evaluation device 200 for trajectory prediction of unmanned systems based on multi-source data fusion and the units contained therein, and will not be repeated here.

[0081] The following is for reference. Figure 3 It illustrates a schematic diagram of the structure of an electronic device (e.g., a computing device) suitable for implementing some embodiments of the present disclosure. Figure 3 The electronic device shown is merely an example and should not be construed as limiting the functionality or scope of the embodiments of this disclosure. Figure 3As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The memory may include a non-volatile storage medium and internal memory. The non-volatile storage medium may store an operating system and a computer program. The computer program includes program instructions that, when executed, cause the processor to perform any of the methods described above. The processor provides computational and control capabilities to support the operation of the entire computer device. The internal memory provides an environment for the execution of the computer program in the non-volatile storage medium; when executed by the processor, the computer program causes the processor to perform any of the methods described above. The network interface is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that... Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present disclosure and does not constitute a limitation on the computer device to which the present disclosure is applied. A specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0082] It should be understood that the processor can be a Central Processing Unit (CPU), but it can also be 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. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0083] In one embodiment, the processor is configured to run a computer program stored in a memory to perform the following steps: constructing a trajectory prediction dataset and training a target model based on the trajectory prediction dataset to obtain a target model, wherein the trajectory prediction data in the trajectory prediction dataset includes the historical trajectory and predicted trajectory corresponding to the agent; generating an adversarial sample set, wherein the adversarial samples in the adversarial sample set are generated through the following steps: perturbing candidate trajectories to obtain perturbed candidate trajectories, wherein the candidate trajectories are the historical trajectories corresponding to the agent; determining a reference adversarial sample set based on the perturbed candidate trajectories, the target model, and a distance-based attention-guided loss function, wherein the reference adversarial samples include: reference adversarial trajectories and predicted reference trajectories; selecting reference adversarial samples that meet the selection criteria from the reference adversarial sample set as adversarial samples, wherein the selection criteria are: the average displacement error corresponding to the reference adversarial sample is the largest; and performing a credibility assessment on the target model based on the adversarial sample set and the real trajectories corresponding to the adversarial samples in the adversarial sample set to generate a credibility assessment result.

[0084] This disclosure also provides a computer-readable storage medium storing a computer program, the computer program including program instructions, and the method implemented when the program instructions are executed can be referred to the various embodiments of the methods described above.

[0085] The aforementioned computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiments, such as the hard disk or memory of the computer device. Alternatively, the aforementioned computer-readable storage medium may be an external storage device of the computer device, such as a plug-in hard disk, SmartMedia Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the computer device.

[0086] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0087] The above descriptions are merely some preferred embodiments of the present disclosure and illustrate the underlying technical principles. Those skilled in the art should understand that the scope of the invention encompassed by the embodiments of the present disclosure is not limited to technical solutions formed by specific combinations of the aforementioned technical features. It also encompasses other technical solutions formed by any combination of the aforementioned technical features or their equivalents, without departing from the aforementioned inventive concept. For example, a technical solution formed by replacing the aforementioned features with (but not limited to) technical features with similar functions disclosed in the embodiments of the present disclosure.

Claims

1. A reliable assessment method for trajectory prediction of unmanned systems based on multi-source data fusion, characterized in that, include: Construct a trajectory prediction dataset and train a trajectory prediction model against the target model based on the trajectory prediction dataset to obtain the target model. The trajectory prediction data in the trajectory prediction dataset includes the historical trajectory and predicted trajectory of the agent. Generate an adversarial sample set, wherein the adversarial samples in the adversarial sample set are generated through the following steps: The candidate trajectory is perturbed to obtain the perturbated candidate trajectory, where the candidate trajectory is the historical trajectory corresponding to the agent; Based on the candidate trajectory after perturbation addition, the target model, and the distance-based attention-guided loss function, a set of reference adversarial samples is determined, wherein the reference adversarial samples include: reference adversarial trajectories and predicted reference trajectories; Select reference adversarial samples that meet the selection criteria from the set of reference adversarial samples, and use them as adversarial samples. The selection criteria is: the reference adversarial sample has the largest average displacement error. Based on the set of adversarial examples and the real trajectories corresponding to the adversarial examples in the set of adversarial examples, the credibility of the target model is evaluated to generate credibility evaluation results.

2. The method according to claim 1, characterized in that, The perturbations added within the candidate trajectory are searched for in the corresponding perturbation space based on an adaptive step size. The step size is determined by the following formula: ； in, Indicates the number of iterations. Indicates the first The step size corresponding to the next iteration. Indicates the first The step size corresponding to the next iteration. Indicates the first The loss value corresponding to the next iteration. Indicates the first The loss value corresponding to the next iteration. This indicates the preset loss threshold.

3. The method according to claim 2, characterized in that, The distance-based attention-guided loss function is characterized by the following formula: ； in, Represents the loss function. Indicates a disturbance. Represents the weight matrix. Represents the trace of a matrix. This represents the transpose of the weight matrix. This represents the predicted trajectory corresponding to the candidate trajectory. This represents the perturbation corresponding to the candidate trajectory. This represents the trade-off parameters corresponding to the disturbance. This represents the tradeoff parameters corresponding to the weight matrix, where, and The values ​​are all 0.

1. The order of the norm. ,in, Represents intelligent agents Time frame in the weight matrix The corresponding weights.

4. The method according to claim 3, characterized in that The loss value corresponding to the distance-based attention-guided loss function is solved by the stochastic projection gradient descent algorithm.

5. The method according to claim 4, characterized in that, The credibility assessment results include: average displacement error and robustness evaluation. The robustness evaluation is controlled by deviation probability. The higher the deviation probability, the lower the robustness evaluation. The deviation probability is used to characterize the probability that the difference between the predicted reference trajectory and the corresponding real trajectory in the reference adversarial sample is greater than a preset difference.

6. A reliable assessment device for trajectory prediction of unmanned systems based on multi-source data fusion, characterized in that, include: The building unit is configured to build a trajectory prediction dataset and train a trajectory prediction model to be attacked based on the trajectory prediction dataset to obtain the target model. The trajectory prediction data in the trajectory prediction dataset includes the historical trajectory and predicted trajectory of the agent. The generation unit is configured to generate a set of adversarial examples, wherein the adversarial examples in the set are generated through the following steps: The candidate trajectory is perturbed to obtain the perturbated candidate trajectory, where the candidate trajectory is the historical trajectory corresponding to the agent; Based on the candidate trajectory after perturbation addition, the target model, and the distance-based attention-guided loss function, a set of reference adversarial samples is determined, wherein the reference adversarial samples include: reference adversarial trajectories and predicted reference trajectories; Select reference adversarial samples that meet the selection criteria from the set of reference adversarial samples, and use them as adversarial samples. The selection criteria is: the reference adversarial sample has the largest average displacement error. The credibility assessment unit is configured to perform credibility assessment on the target model based on the adversarial sample set and the real trajectories corresponding to the adversarial samples in the adversarial sample set, so as to generate credibility assessment results.

7. An electronic device, characterized in that, include: One or more processors; A storage device on which one or more programs are stored; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1 to 5.

8. A computer-readable medium, characterized in that, It stores a computer program thereon, wherein the computer program, when executed by a processor, implements the method as described in any one of claims 1 to 5.

Images