An autonomous vehicle countermeasure trajectory prediction method, device, equipment and medium

Abstract

The present application provides an adversarial trajectory prediction method, apparatus, device and medium for an autonomous driving vehicle. The method includes: determining a dynamic interaction space between the autonomous driving vehicle and the attacking vehicle based on their respective driving information and their common lane information; converting the multi-dimensional information of the dynamic interaction space into a graph structure; extracting a high-dimensional feature vector from the graph structure; inputting the high-dimensional feature vector into a pre-trained trajectory prediction model, so that the trajectory prediction model maps the high-dimensional feature vector into a low-dimensional latent space feature vector and decodes the low-dimensional latent space feature vector into an adversarial trajectory of the autonomous driving vehicle. The present invention improves the accuracy of trajectory prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of autonomous driving, and in particular, to a method, device, equipment and medium for predicting adversarial trajectories of autonomous vehicles. Background Art

[0002] With the rapid development of autonomous driving technology, ADS (Automotive Driving System) is gradually becoming the core technology in the field of intelligent transportation. However, the safety, reliability and stability of the autonomous driving system are still key issues, especially in complex and changing road environments. Trajectory prediction is an important function in autonomous driving, which involves predicting future trajectories based on information such as the current environment, traffic conditions and vehicle states.

[0003] Most traditional trajectory prediction methods rely on simplified models and assumptions, and cannot effectively cope with complex dynamic environments and potential adversarial attacks, resulting in poor robustness of existing adversarial trajectory prediction models and low accuracy of predicted adversarial trajectories. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a method, device, equipment and medium for predicting adversarial trajectories of autonomous vehicles to improve the accuracy of adversarial trajectory prediction.

[0005] In a first aspect, a method for predicting adversarial trajectories of autonomous vehicles is provided, and the method includes:

[0006] Determine the dynamic interaction space between the autonomous vehicle and the attacking vehicle based on their respective driving information and the common lane information between them; the attacking vehicle refers to a vehicle that poses a driving risk to the autonomous vehicle;

[0007] Convert the multi-dimensional information of the dynamic interaction space into a graph structure;

[0008] Extract high-dimensional feature vectors from the graph structure;

[0009] Input the high-dimensional feature vectors into a pre-trained trajectory prediction model, so that the trajectory prediction model maps the high-dimensional feature vectors into low-dimensional latent space feature vectors and decodes the low-dimensional latent space feature vectors into adversarial trajectories of the autonomous vehicle, so that the autonomous vehicle can adjust its driving strategy according to the adversarial trajectories.

[0010] Optionally, determining the dynamic interaction space between the autonomous vehicle and the attacking vehicle based on their respective driving information and the common lane information between them includes:

[0011] Obtain the driving information of the autonomous vehicle and the attacking vehicle, as well as the common lane information between them;

[0012] Construct road constraint conditions based on the common lane information of both;

[0013] Determine the dynamic constraint conditions of the autonomous driving vehicle based on the driving information of the autonomous driving vehicle; and determine the dynamic drivable area of the autonomous driving vehicle based on the road constraint conditions and the dynamic constraint conditions of the autonomous driving vehicle;

[0014] Determine the intervention strategy constraint conditions of the attacking vehicle based on the driving information of the attacking vehicle, and determine the intervention space of the attacking vehicle based on the road constraint conditions and the intervention strategy constraint conditions;

[0015] Determine the dynamic interaction space between the two based on the dynamic drivable area of the autonomous driving vehicle and the intervention space of the attacking vehicle.

[0016] Optionally, extracting the high-dimensional feature vector in the graph structure includes:

[0017] Extract the high-dimensional feature vector from the graph structure based on the multi-aggregation message passing attention mechanism.

[0018] Optionally, the trajectory prediction model consists of an encoder and a decoder; the encoder includes a first encoder, a second encoder, and a third encoder, and the training processes of the first to third encoders include:

[0019] Input the training samples of the time headway between the autonomous driving vehicle and several attacking vehicles into the first encoder for regression training, and optimize the parameters of the first encoder through the first loss function until the preset number of iterations is reached; the time headway refers to the time difference when the front of the autonomous driving vehicle and the attacking vehicle pass through the same position point; the first loss function is as follows:

[0020]

[0021] Wherein, is the predicted time headway, is the observed true time headway, is the number of training times;

[0022] Input the training samples of the vehicle behavior patterns into the second encoder for classification training, and optimize the parameters of the second encoder through the second loss function until the preset number of iterations is reached; the vehicle behavior patterns at least include going straight, turning left / left lane change, and turning right / right lane change; the second loss function is as follows:

[0023]

[0024] Wherein, is the true behavior pattern label, is the predicted behavior pattern probability;

[0025] Input the training samples of the low-level random features of the attacking vehicle into the third encoder for Gaussian training until the preset number of training iterations is reached. The low-level random features include at least driving details and environmental interference.

[0026] Optionally, the training process of the first to third encoders further includes:

[0027] Perform discriminative training on the time headway data generated by the first encoder through the first discriminator;

[0028] Perform discriminative training on the behavior pattern data generated by the second encoder through the second discriminator;

[0029] Perform discriminative training on the low-level random features generated by the third encoder through the third discriminator.

[0030] Optionally, the training process of the decoder includes:

[0031] Calculate the centroid in the intersection space of the autonomous driving vehicle and the attacking vehicle based on the multi-dimensional information of the dynamic interaction space;

[0032] Calculate the fourth loss function based on the decoded trajectory points and the centroid until the preset number of iterations is reached. The fourth loss function is as follows:

[0033]

[0034] Wherein, represents the adversarial trajectory of the attacking vehicle, is the distance between the trajectory point of the attacking vehicle and the centroid of the dynamic interaction space.

[0035] On the second aspect, an autonomous driving adversarial trajectory prediction device is provided. The device includes: an intersection modeling module, a graph structure vectorization processing module, a latent space vector mapping module, and a trajectory generation module;

[0036] The intersection modeling module is used to determine the dynamic interaction space of the autonomous driving vehicle and the attacking vehicle based on the respective driving information of the autonomous driving vehicle and the attacking vehicle and the common lane information of the two. The attacking vehicle refers to a vehicle that poses a driving risk to the autonomous driving vehicle;

[0037] The graph structure vectorization processing module is used to convert the multi-dimensional information of the dynamic interaction space into a graph structure and extract the high-dimensional feature vector in the graph structure;

[0038] The latent space vector mapping module is used to map the high-dimensional feature vector into a low-dimensional latent space vector;

[0039] The trajectory generation module is used to decode the low-dimensional latent space vector into the adversarial trajectory of the autonomous driving vehicle so that the autonomous driving vehicle can adjust its driving strategy according to the adversarial trajectory.

[0040] Optionally, the latent space vector mapping module consists of a first encoder, a second encoder, and a third encoder;

[0041] The first encoder is used to extract the distribution characteristics of the time headway between the autonomous vehicle and the attacking vehicle; the time headway refers to the time difference when the front ends of the autonomous vehicle and the attacking vehicle pass through the same position point;

[0042] The second encoder is used to extract the behavior pattern characteristics of the attacking vehicle, and the behavior patterns include going straight, turning left / left lane change, and turning right / right lane change;

[0043] The third encoder is used to extract the low-level random characteristics of the attacking vehicle, and the low-level random characteristics at least include driving details and environmental interference.

[0044] In a third aspect, an electronic device is provided, including a processor, a communication interface, a memory, and a communication bus. Among them, the processor, the communication interface, and the memory complete communication with each other through the communication bus;

[0045] The memory is used to store a computer program;

[0046] The processor is used to implement the method steps described in any one of the first aspects when executing the program stored on the memory.

[0047] In a fourth aspect, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method steps described in any one of the first aspects are implemented.

[0048] A method, device, equipment, and medium for predicting the confrontation trajectory of an autonomous vehicle provided by an embodiment of the present invention. By modeling the intersection space of the autonomous vehicle and the attacking vehicle, the present invention can extract rich semantic information, which can not only reflect the geometric characteristics of the road, but also describe the dynamic interaction relationship between vehicles; provide rich data support for the subsequent trajectory prediction model; and by converting the multi-dimensional information of the dynamic interaction space into graph structure data, the data representation can take into account both local details and global semantics, and at the same time facilitate efficient processing by subsequent algorithms; by using an encoder to map the high-dimensional feature vector into a low-dimensional latent space vector, it can not only capture the high-level semantic features of the vehicle, but also contain low-level random characteristics, thereby improving the understanding and generation ability of the trajectory behavior and enhancing the robustness of the trajectory prediction model.

[0049] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following specifically enumerates preferred embodiments and, in conjunction with the accompanying drawings, makes a detailed description as follows. Description of the Drawings

[0050] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the accompanying drawings required in the embodiments. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0051] Figure 1 Shows a flowchart of a method for predicting adversarial trajectories of an autonomous driving vehicle provided by an embodiment of the present invention;

[0052] Figure 2 Shows a schematic structural diagram of an apparatus for predicting adversarial trajectories of an autonomous driving vehicle provided by an embodiment of the present invention;

[0053] Figure 3 Shows a schematic structural diagram of an electronic device provided by an embodiment of the present invention. Specific Embodiments

[0054] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, rather than all, embodiments of the present invention. Usually, the components of the embodiments of the present invention described and illustrated in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed present invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

[0055] Considering that most traditional trajectory prediction methods rely on simplified models and assumptions and cannot effectively cope with complex dynamic environments and potential adversarial attacks, resulting in low robustness of existing adversarial trajectory prediction models, low accuracy of predicted adversarial trajectories, and poor accuracy. The main problems are as follows:

[0056] 1. Insufficient utilization of semantic information: Although existing adversarial trajectory prediction models perform well in most scenarios, they do not make full use of semantic information in terms of algorithmic strategy guidance. As a result, it does not explicitly affect the actual drivable area of autonomous driving vehicles, weakening the effectiveness of the adversarial trajectory prediction model for autonomous driving.

[0057] 2. Redundancy in the high-dimensional feature space: Currently, end-to-end adversarial trajectory prediction models for autonomous vehicles usually optimize the model's parameters in a learning-based manner. Existing models often process available trajectory, map, and other information into feature vectors during training. However, in fact, high-dimensional feature vectors have many redundant dimensions in expressing significant features, that is, dimension information with weak correlation also participates in the parameter iteration process. The expression accuracy of significant features directly affects the performance of the adversarial trajectory prediction model.

[0058] Based on this, embodiments of the present invention provide an adversarial trajectory prediction method and device for autonomous vehicles, which will be described below through embodiments.

[0059] Embodiments of the present invention provide an adversarial trajectory prediction method for autonomous vehicles, as Figure 1 shown, the method includes the following steps:

[0060] Step S101: Determine the dynamic interaction space between the autonomous vehicle and the attacking vehicle based on the driving information of the autonomous vehicle and the attacking vehicle respectively and the common lane information of the two. The attacking vehicle refers to a vehicle that poses a driving risk to the autonomous vehicle.

[0061] In a feasible embodiment, determining the dynamic interaction space between the autonomous vehicle and the attacking vehicle based on the driving information of the autonomous vehicle and the attacking vehicle respectively and the common lane information of the two includes:

[0062] Step S101A: Obtain the driving information of the autonomous vehicle and the attacking vehicle, as well as the common lane information of the two.

[0063] In this step, the driving information of the autonomous vehicle includes the position coordinates, heading angle, and speed of the autonomous vehicle. In one example, the driving information of the autonomous vehicle can be expressed as ; where is the position coordinates of the autonomous vehicle, is the heading angle of the autonomous vehicle, is the speed of the autonomous vehicle. The driving information of the attacking vehicle can be expressed as , where is the position coordinates of the attacking vehicle, is the heading angle of the attacking vehicle; is the speed of the attacking vehicle.

[0064] In one example, the lane information includes road structure, traffic rules, etc.

[0065] Step S101B: Construct road constraint conditions based on the common lane information of the two.

[0066] In this step, road constraint conditions such as lane boundaries and no-go areas, etc., in one example, the road constraint conditions can be expressed as .

[0067] Step S101C: Determine the dynamic constraint conditions of the autonomous driving vehicle based on the driving information of the autonomous driving vehicle; and determine the dynamic drivable area of the autonomous driving vehicle based on the road constraint conditions and the dynamic constraint conditions of the autonomous driving vehicle.

[0068] In this step, the dynamic feasible area refers to the area where the autonomous driving vehicle can legally and safely drive at the current moment. This area is restricted by factors such as road structure, traffic rules, and vehicle dynamic performance. The dynamic constraint conditions of the autonomous driving vehicle are, for example, turning radius, acceleration limit, etc. In one example, the dynamic constraint conditions of the autonomous driving vehicle can be expressed as ;

[0069] The dynamic drivable area (DVR, Dynamic Viable Region) can be expressed as:

[0070] (1);

[0071] Wherein, is the road space, including all lane boundaries.

[0072] Step S101D: Determine the intervention strategy constraint conditions of the attacking vehicle based on the driving information of the attacking vehicle, and determine the intervention space of the attacking vehicle based on the road constraint conditions and the intervention strategy constraint conditions.

[0073] The intervention space (AIS, Adversarial Intervention Space) describes the area range where the attacking vehicle forms potential interference to the autonomous driving vehicle, involving intervention strategies such as lane-changing behavior and speed change.

[0074] Among them, the intervention strategy constraint condition is the lane-changing target area strategy constraint, and this strategy constraint predicts its lane-changing intention based on the current state of the attacking vehicle. In one example, the intervention strategy constraint condition can be expressed as ; Continuing the previous example, the intervention space can be expressed as:

[0075] (2);

[0076] Step S101E: Determine the dynamic interaction space between the two based on the dynamic drivable area of the autonomous driving vehicle and the intervention space of the attacking vehicle.

[0077] Continuing the previous example, the dynamic interaction space can be expressed as:

[0078] (3).

[0079] In the embodiment of the present invention, by cross - modeling the dynamic viable region (DVR, Dynamic Viable Region) of the autonomous vehicle and the adversarial intervention space (AIS, Adversarial Intervention Space) of the attacking vehicle, a semantically rich feature representation is generated. This representation can not only reflect the geometric characteristics of the road but also describe the dynamic interaction relationships between vehicles, realizing the full utilization of semantic information.

[0080] Step S102: Convert the multi - dimensional information in the dynamic interaction space into a graph structure.

[0081] By converting the complex multi - dimensional information of the dynamic interaction space J into graph - structured data, this graph structure enables the data representation to take into account both local details and global semantics, while facilitating efficient processing by subsequent algorithms.

[0082] In a feasible implementation, the conversion process of this graph structure is as follows:

[0083] The first step is to construct the vertex set in the graph structure , and the vertex set is the discretized position points or dynamic state points in the dynamic interaction space. Let the dynamic interaction space be , and the vertex set is generated by uniform sampling or semantic importance sampling: ;

[0084] where represents a semantic point generated by the combination of the state of the autonomous vehicle and the state of the attacking vehicle.

[0085] The second step is to construct the edge set between the vertices. If two vertices share the same semantic features (such as being in the same lane), construct an edge connection relationship . In an example, the semantic dependency determination function can be used to construct the edge connection relationship . When is true, takes the value of 1; otherwise takes the value of 0. Taking the value of 1 means that there are the same semantic features between two nodes, and an edge connection relationship can be constructed between the two nodes; Taking the value of 0 means not constructing an edge relationship between the two nodes.

[0086] If an edge connection relationship is established between two nodes, corresponding weights can also be set for each edge. Specifically, the following three weights can be attached: a distance weight is attached to each edge , and this distance weight measures the direct distance between two vehicles; a dynamically changing weight , and this dynamically changing weight measures the change in the position of the vehicle in different states; a semantic similarity weight , when is true, then takes a value of 1; otherwise takes a value of 0.

[0087] Step 3: Obtain the attributes of the vertices, including four types: spatial attributes, dynamic attributes, interference attributes, and semantic attributes. Among them, the position of the vertex in the physical space serves as the spatial attribute; the headings and speeds of the two vehicles serve as the dynamic attributes; the interference intensity is an index of the anti-interference function in the semantic space , and the interference intensity serves as the interference attribute; the lane number, priority identifier, traffic rule status, etc. serve as the semantic attributes.

[0088] Finally, a graph structure is formed , where is the vertex set, is the edge set, with vertex attributes attached and edge attributes .

[0089] Step S103: Extract the high-dimensional feature vectors from the graph structure.

[0090] In a feasible implementation, based on the Mamba (Multi-Aggregation Message-Based Attention) model, high-dimensional feature vectors are extracted from the graph structure. Semantic features are extracted from the graph structure to provide a powerful input representation for latent space optimization.

[0091] The Mamba model extracts salient features from the graph structure. Its core idea is to combine multiple aggregation strategies and attention mechanisms to dynamically learn the relationship between vertices and their neighbors.

[0092] Specifically, the extraction process is that in each round of propagation, each vertex receives information from its neighbors , and updates its own state:

[0093] (4);

[0094] Among them, is a vertex at the layer feature is a trainable weight matrix for the vertex neighbor set , is a message generation function is the attention weight, expressed as:

[0095] (5);

[0096] wherein is the edge of the graph structure is the number of propagation layers

[0097] After the propagation ends, through a global pooling operation, a high-dimensional feature vector of the entire graph is extracted :

[0098] (6).

[0099] Step S104: Input the high-dimensional feature vector into a pre-trained trajectory prediction model, so that the trajectory prediction model maps the high-dimensional feature vector into a low-dimensional latent space feature vector, and decodes the low-dimensional latent space feature vector into an adversarial trajectory of the autonomous vehicle, so that the autonomous vehicle adjusts its driving strategy according to the adversarial trajectory

[0100] The low-dimensional latent space feature vector is the core part of the entire autonomous driving trajectory prediction, aiming to represent the high-dimensional information in the vehicle trajectory in a compact and interpretable manner. These latent space feature vectors, by combining domain knowledge and statistical distribution modeling, can not only capture the high-level semantic features of the vehicle, but also contain low-level stochastic characteristics, thereby enhancing the understanding and generation ability of the trajectory behavior

[0101] Mapping the high-dimensional feature vector into a low-dimensional latent space feature vector is mainly extracted from three dimensions, one is the longitudinal dimension, one is the lateral dimension, and one is the detail dimension. The longitudinal dimension refers to the position of the vehicle in the driving direction and the change information in time. The lateral dimension refers to the change information when the vehicle changes lanes left and right laterally. The detail dimension refers to the driving details of the vehicle, surrounding environment and other information

[0102] In the embodiment of the present invention, the trajectory prediction model consists of an encoder and a decoder; the encoder includes a first encoder, a second encoder and a third encoder, and the first to third encoders are trained respectively, and the training process includes:

[0103] Step S104A: Input the training samples of the time headway between the autonomous vehicle and a number of attacking vehicles into the first encoder for regression training, and optimize the parameters of the first encoder through the first loss function until the preset number of iterations is reached; the time headway refers to the time difference when the fronts of the autonomous vehicle and the attacking vehicle pass through the same position point; the first loss function is as follows:

[0104] (7);

[0105] where is the predicted time headway, is the observed true time headway, is the number of training times.

[0106] The first encoder is used to extract the longitudinal features of the vehicle, and the longitudinal features are represented by the time headway. In one example, the time headway in an urban driving scenario follows a log-normal distribution. Its probability density function is defined as:

[0107] (8);

[0108] where is the vehicle position parameter, representing the mean of the normal distribution; is the scale parameter, representing the standard deviation of the distribution. Through this function, the time headway of the vehicle can be predicted.

[0109] Step S104B: Input the training samples of the vehicle behavior patterns into the second encoder for classification training, and optimize the parameters of the second encoder through the second loss function until the preset number of iterations is reached; the vehicle behavior patterns at least include going straight, turning left / merging left, and turning right / merging right; the second loss function is as follows:

[0110] (9);

[0111] where is the true behavior pattern label, is the predicted behavior pattern probability.

[0112] The second encoder is used to extract the lateral features of the vehicle, and the lateral features are represented by the vehicle behavior patterns. In one example, the lateral feature latent space vector can be extracted through the second encoder, where 、 、 respectively represent the probabilities of the vehicle being in three behavior patterns.

[0113] Step S104C: Input the training samples of the low-level random features of the attacking vehicle into the third encoder for Gaussian training until the preset number of training iterations is reached. The low-level random features include at least driving details and environmental interference.

[0114] In addition to semantic features, there are many low-level random characteristics in vehicle trajectories, such as driving details, environmental interference, etc. These characteristics often cannot be captured by explicit semantic classification or regression modeling. Therefore, in the embodiments of the present invention, a Gaussian encoder G is used to extract them.

[0115] High-dimensional feature vector Through the Gaussian encoder Mapped to a random latent space vector , and this latent space vector satisfies a Gaussian distribution: ; where and are the mean and covariance matrix of the Gaussian distribution, predicted by the Gaussian encoder G.

[0116] To ensure the authenticity of the latent space feature vector, in the embodiments of the present invention, an adversarial generation loss is introduced for regularization, that is, the fake samples and real samples generated by each encoder are jointly input into the discriminator for discrimination, and adversarial training is performed to force the latent space feature vector generated by the encoder to be close to the true value.

[0117] Therefore, on the basis of the above embodiments, the training process of the first to third encoders further includes:

[0118] Step S104D: Perform discriminant training on the time headway data generated by the first encoder through the first discriminator.

[0119] The first discriminator Mainly targets the time headway To verify whether it conforms to the lognormal distribution:

[0120] (10);

[0121] Where Represents the probability density function of the lognormal distribution. By optimizing the following adversarial loss function, the longitudinal latent space distribution generated by the first encoder is constrained:

[0122] (11);

[0123] Where Represents the true distribution of the time headway; Represents the distribution of the time headway predicted by the first encoder.

[0124] Step S104E: Use the second discriminator to perform discriminative training on the behavior pattern data generated by the second encoder.

[0125] The second discriminator mainly targets the horizontal characteristics to verify whether it conforms to the classification distribution :

[0126] (12);

[0127] Constrain the horizontal characteristics through a similar adversarial loss function:

[0128] (13);

[0129] where represents the classification distribution of the true behavior pattern; represents the classification distribution of the behavior pattern predicted by the second encoder.

[0130] Step S104F: Use the third discriminator to perform discriminative training on the low-level random features generated by the third encoder.

[0131] Random latent space is a feature with strong randomness extracted by the third encoder from the high-dimensional feature vector, and its target distribution is the standard Gaussian distribution . The discriminator is used to verify whether it satisfies the standard Gaussian distribution:

[0132] (14);

[0133] Optimize the distribution alignment of the random latent space vector through the following adversarial loss function:

[0134] (15);

[0135] where is the distribution of the true random latent space vector; is the distribution of the random latent space vector predicted by the third encoder.

[0136] The encoder and discriminator in the embodiment of the present invention form an adversarial autoencoder, which is trained through a generative adversarial training mechanism. The low-dimensional latent space variables are optimized through the trained adversarial autoencoder, which can automatically extract low-dimensional salient features and force the generated latent space vectors to approach the target distribution. This improves the robustness of adversarial trajectory learning. When the trajectory prediction model faces potential adversarial attacks, it can better adapt and defend, thereby enhancing the safety and stability of the autonomous driving system.

[0137] By optimizing the low-dimensional latent space vector, not only the convergence speed of the adversarial learning of the trajectory prediction model is improved, but also the prediction effect of the model is optimized, the computational cost during training is reduced, and at the same time the final trajectory prediction accuracy and reliability are enhanced.

[0138] In the embodiment of the present invention, by mapping the high-dimensional feature vector into a low-dimensional latent space vector, it is realized that more representative feature information is expressed using less dimensional information, and the redundancy of the high-dimensional feature vector is reduced.

[0139] Based on the above embodiment, the training process of the decoder includes:

[0140] Step S104G: Calculate the centroid in the intersection space of the autonomous driving vehicle and the attacking vehicle based on the multi-dimensional information of the dynamic interaction space.

[0141] The centroid position reflects the most representative area of the interaction between the dynamic drivable region DVR of the autonomous driving vehicle and the intervention space AIS of the attacking vehicle, and is the key point most likely to interfere with the autonomous driving vehicle. Let be the area of the dynamic interaction space, then the calculation formula of the centroid is:

[0142] (16);

[0143] Among them, is the centroid; is the region of the dynamic interaction space.

[0144] Step S104H: Calculate the fourth loss function based on the decoded trajectory points and the centroid until the preset number of iterations is reached. The fourth loss function is as follows:

[0145] (17);

[0146] Among them, represents the adversarial trajectory of the attacking vehicle, is the distance between the trajectory point of the attacking vehicle and the centroid of the dynamic interaction space. The smaller this distance is, the smaller the loss is, and the optimized trajectory is closer to the centroid.

[0147] The decoded trajectory point is the trajectory point of the attacking vehicle. The trajectory optimization goal of this attacking vehicle is to make its trajectory points gradually approach the centroid position, so as to enhance the interference effect on the autonomous driving vehicle.

[0148] In another embodiment, after the training is completed, the effectiveness and naturalness of the trajectory predicted by the trajectory prediction model are optimized. The generated trajectory can be iteratively optimized by a multi-objective optimization method. The optimization objectives include the interference effect of the adversarial trajectory on the autonomous vehicle and the physical feasibility, smoothness, and human-like driving characteristics (naturalness) of the adversarial trajectory, ensuring that the adversarial trajectory meets the actual application requirements. After the optimization is completed, the effectiveness and naturalness of the trajectory can be evaluated respectively. Specifically, the evaluation of effectiveness includes quantitative indicators such as path deviation, speed change, and safety impact. The evaluation of naturalness includes evaluating whether the generated adversarial trajectory conforms to physical constraints and traffic rules, ensuring that the trajectory has smoothness, continuity, and practical executability.

[0149] By optimizing and evaluating the effectiveness of the trajectory, it is ensured that the generated adversarial trajectory has an attack effect in actual applications, thereby enhancing the ability to evaluate the safety and robustness of the autonomous driving system. By optimizing and evaluating the naturalness of the trajectory, it is ensured that the trajectory has natural driving characteristics in terms of rationality and reality. Ensure that the adversarial trajectory has both interference effects and is close to the real scenario.

[0150] In addition, a virtual-real combined evaluation system can be used to conduct simulation tests and real-vehicle verification on the adversarial trajectories after evaluation and screening, comprehensively evaluate the effectiveness and transferability of the adversarial testing method in safety-critical scenarios, and provide a reliable basis for the safety testing of the autonomous driving system.

[0151] It can be understood from the above embodiments that the present invention can extract rich semantic information by modeling the intersection space of the autonomous vehicle and the attacking vehicle, which can not only reflect the geometric characteristics of the road but also describe the dynamic interaction relationship between vehicles; provide rich data support for the subsequent trajectory prediction model; and by converting the multi-dimensional information of the dynamic interaction space into graph-structured data, the data representation can take into account both local details and global semantics, while facilitating efficient processing by subsequent algorithms; by using an encoder to map the high-dimensional feature vector into a low-dimensional latent space vector, it can not only capture the high-level semantic features of the vehicle but also include low-level random characteristics, thereby enhancing the understanding and generation ability of the trajectory behavior and improving the robustness of the trajectory prediction model.

[0152] Based on the same inventive concept, an autonomous driving adversarial trajectory prediction device is provided, as Figure 2 shown. The device includes: an intersection modeling module 201, a graph structure vectorization processing module 202, a latent space vector mapping module 203, and a trajectory generation module 204;

[0153] The intersection modeling module 201 is used to determine the dynamic interaction space between the autonomous vehicle and the attacking vehicle based on the driving information of each of the autonomous vehicle and the attacking vehicle and the common lane information of the two; the attacking vehicle refers to a vehicle that poses a driving risk to the autonomous vehicle;

[0154] The graph structure vectorization processing module 202 is used to convert the multi-dimensional information in the dynamic interaction space into a graph structure and extract the high-dimensional feature vectors in the graph structure;

[0155] The latent space vector mapping module 203 is used to map the high-dimensional feature vectors into low-dimensional latent space vectors;

[0156] The trajectory generation module 204 is used to decode the low-dimensional latent space vectors into adversarial trajectories of the autonomous vehicle, so that the autonomous vehicle can adjust its driving strategy according to the adversarial trajectories.

[0157] Based on the above embodiments, the latent space vector mapping module 203 is composed of a first encoder, a second encoder and a third encoder;

[0158] The first encoder is used to extract the distribution characteristics of the time headway between the autonomous vehicle and the attacking vehicle; the time headway refers to the time difference when the front ends of the autonomous vehicle and the attacking vehicle pass through the same position point;

[0159] The second encoder is used to extract the behavior pattern characteristics of the attacking vehicle, and the behavior patterns include going straight, turning left / left lane change and turning right / right lane change;

[0160] The third encoder is used to extract the low-level random characteristics of the attacking vehicle, and the low-level random characteristics at least include driving details and environmental interference.

[0161] Based on the same technical concept, an embodiment of the present invention also provides an electronic device, as Figure 3 shown, including a processor 301, a communication interface 302, a memory 303 and a communication bus 304, wherein the processor 301, the communication interface 302, and the memory 303 communicate with each other through the communication bus 304.

[0162] The memory 303 is used to store a computer program;

[0163] The processor 301 is used to implement the steps of the autonomous vehicle adversarial trajectory prediction method when executing the program stored in the memory 303.

[0164] The communication bus mentioned in the above electronic device may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of simplicity, only a thick line is used to represent it in the figure, but it does not mean that there is only one bus or one type of bus.

[0165] The communication interface is used for communication between the above-mentioned electronic device and other devices.

[0166] The memory may include a Random Access Memory (RAM), or may also include a Non-Volatile Memory (NVM), such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0167] The aforementioned processor may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0168] The computer program product for the method of predicting adversarial trajectories of an autonomous vehicle provided by an embodiment of the present invention includes a computer-readable storage medium storing program code, and the instructions included in the program code can be used to execute the method described in the foregoing method embodiment. For specific implementation, reference can be made to the method embodiment, which will not be elaborated here.

[0169] The device for predicting adversarial trajectories of an autonomous vehicle provided by an embodiment of the present invention may be specific hardware on the device, or software or firmware installed on the device, etc. For the device provided by an embodiment of the present invention, its implementation principle and the technical effects produced are the same as those of the foregoing method embodiment. For the sake of brief description, for the parts not mentioned in the device embodiment, reference can be made to the corresponding content in the foregoing method embodiment. Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the aforementioned system, device, and unit can all refer to the corresponding processes in the above method embodiment, which will not be elaborated here.

[0170] Finally, it should be noted that the above-described embodiments are only specific embodiments of the present invention, used to illustrate the technical solutions of the present invention, rather than limiting it. The protection scope of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that any technician familiar with the technical field of the present invention can still modify the technical solutions recorded in the foregoing embodiments or can easily think of changes, or perform equivalent replacements on some of the technical features; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. All should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. An anti-trajectory prediction method for autonomous vehicles, characterized in that The method includes: Obtaining the driving information of the autonomous driving vehicle and the attacking vehicle, as well as the lane information common to both; Constructing road constraint conditions based on the lane information common to both; Determining the dynamic constraint conditions of the autonomous driving vehicle based on the driving information of the autonomous driving vehicle; and determining the dynamic drivable area of the autonomous driving vehicle based on the road constraint conditions and the dynamic constraint conditions of the autonomous driving vehicle; Determining the intervention strategy constraint conditions of the attacking vehicle based on the driving information of the attacking vehicle, and determining the intervention space of the attacking vehicle based on the road constraint conditions and the intervention strategy constraint conditions; Determining the dynamic interaction space between the two based on the dynamic drivable area of the autonomous driving vehicle and the intervention space of the attacking vehicle; the attacking vehicle refers to a vehicle that poses a driving risk to the autonomous driving vehicle; Converting the multi-dimensional information of the dynamic interaction space into a graph structure; Extracting high-dimensional feature vectors from the graph structure; Inputting the high-dimensional feature vectors into a pre-trained trajectory prediction model, so that the trajectory prediction model maps the high-dimensional feature vectors into low-dimensional latent space feature vectors, and decodes the low-dimensional latent space feature vectors into adversarial trajectories of the autonomous driving vehicle, so that the autonomous driving vehicle adjusts its driving strategy according to the adversarial trajectories.

2. The method according to claim 1, wherein The extracting the high-dimensional feature vectors from the graph structure includes: Extracting high-dimensional feature vectors from the graph structure based on a multi-aggregation message passing attention mechanism.

3. The method according to claim 1, wherein The trajectory prediction model consists of an encoder and a decoder; the encoder includes a first encoder, a second encoder, and a third encoder, and the training process of the first to third encoders includes: Inputting the training samples of the time headway between the autonomous driving vehicle and several attacking vehicles into the first encoder for regression training, and optimizing the parameters of the first encoder through a first loss function until a preset number of iterations is reached; the time headway refers to the time difference when the fronts of the autonomous driving vehicle and the attacking vehicle pass through the same position point; the first loss function is as follows: wherein, is the predicted time headway, is the observed true time headway, is the number of training times; Inputting the training samples of the vehicle behavior patterns into the second encoder for classification training, and optimizing the parameters of the second encoder through a second loss function until a preset number of iterations is reached; the vehicle behavior patterns at least include going straight, turning left / left lane change, and turning right / right lane change; the second loss function is as follows: Among them, is the real behavior pattern label, is the predicted probability of the behavior pattern; Inputting the training samples of the low-level random features of the attacking vehicle into the third encoder for Gaussian training until a preset number of training iterations is reached, and the low-level random features at least include driving details and environmental interference.

4. The method according to claim 3, wherein The training process of the first to third encoders further includes: Performing discriminant training on the time headway data generated by the first encoder through a first discriminator; Performing discriminant training on the behavior pattern data generated by the second encoder through a second discriminator; Performing discriminant training on the low-level random features generated by the third encoder through a third discriminator.

5. The method according to claim 4, characterized in that The training process of the decoder includes: Calculating the centroid in the intersection space between the autonomous driving vehicle and the attacking vehicle based on the multi-dimensional information of the dynamic interaction space; Calculate a fourth loss function based on the decoded trajectory points and the centroid until a preset number of iterations is reached. The fourth loss function is as follows: Among them, represents the confrontation trajectory of the attacking vehicle, which is the distance between the trajectory point of the attacking vehicle and the centroid of the dynamic interaction space.

6. An automatic driving countermeasure trajectory prediction device, characterized in that The device includes: a cross-modeling module, a graph structure vectorization processing module, a latent space vector mapping module, and a trajectory generation module; The cross-modeling module is used to determine the dynamic interaction space between the autonomous driving vehicle and the attacking vehicle based on the respective driving information of the autonomous driving vehicle and the attacking vehicle and the common lane information of the two. The attacking vehicle refers to a vehicle that poses a driving risk to the autonomous driving vehicle; Specifically, the cross-modeling module is used to obtain the driving information of the autonomous driving vehicle and the attacking vehicle, as well as the common lane information of the two; Construct road constraint conditions based on the common lane information of the two; Determine the dynamic constraint conditions of the autonomous driving vehicle based on the driving information of the autonomous driving vehicle; and determine the dynamic drivable area of the autonomous driving vehicle based on the road constraint conditions and the dynamic constraint conditions of the autonomous driving vehicle; Determine the intervention strategy constraint conditions of the attacking vehicle based on the driving information of the attacking vehicle, and determine the intervention space of the attacking vehicle based on the road constraint conditions and the intervention strategy constraint conditions; Determine the dynamic interaction space between the two based on the dynamic drivable area of the autonomous driving vehicle and the intervention space of the attacking vehicle; The graph structure vectorization processing module is used to convert the multi-dimensional information of the dynamic interaction space into a graph structure and extract the high-dimensional feature vectors in the graph structure; The latent space vector mapping module is used to map the high-dimensional feature vectors into low-dimensional latent space vectors; The trajectory generation module is used to decode the low-dimensional latent space vectors into adversarial trajectories of the autonomous driving vehicle so that the autonomous driving vehicle can adjust its driving strategy according to the adversarial trajectories.

7. The automatic driving countermeasure trajectory prediction device according to claim 6, wherein The latent space vector mapping module consists of a first encoder, a second encoder, and a third encoder; The first encoder is used to extract the distribution characteristics of the time headway between the autonomous driving vehicle and the attacking vehicle. The time headway refers to the time difference when the front ends of the autonomous driving vehicle and the attacking vehicle pass through the same position point; The second encoder is used to extract the behavior pattern characteristics of the attacking vehicle. The behavior patterns include going straight, turning left / left lane change, and turning right / right lane change; The third encoder is used to extract the low-level random characteristics of the attacking vehicle. The low-level random characteristics at least include driving details and environmental interference.

8. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus. Among them, the processor, the communication interface, and the memory complete communication with each other through the communication bus; The memory is used to store computer programs; The processor is used to implement the method steps described in any one of claims 1-7 when executing the programs stored on the memory.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, it implements the method steps described in any one of claims 1-7.

Images