Method for generating trusted interaction behavior based on risk-driven hesitation trigger model

Abstract

The application relates to a trusted interaction behavior generation method based on a risk-driven hesitation trigger model, which comprises the following steps: identifying a hesitation section, an execution section and a stable section of a vehicle according to the lateral and longitudinal interaction behavior data of the vehicle; constructing frame-by-frame labels and section-level labels of the hesitation section, the execution section and the stable section, constructing a multi-risk factor matrix, frame-level training corpus, section-level training corpus and a double-branch network model, the frame-level training corpus being obtained by extracting the frame-by-frame labels, the section-level training corpus being obtained by extracting the section-level labels, and the double-branch network model being obtained by training the frame-level training corpus and the section-level training corpus; inputting the multi-risk factor matrix into the double-branch network model to obtain a trigger probability and a duration distribution of the interaction behavior, and forming the trusted interaction behavior of the hesitation section, the execution section and the stable section of the vehicle according to the trigger probability and the duration distribution of the interaction behavior. Thus, a trusted interaction test scene close to human behavior characteristics is generated, and the credibility of intelligent networked vehicle testing is improved.

Description

Technical Field

[0001] This application relates to the fields of intelligent connected vehicle testing and verification and autonomous driving simulation technology, and in particular to a method for generating credible interactive behavior based on a risk-driven hesitation triggering model. Background Technology

[0002] With the continuous expansion of intelligent connected vehicle testing and autonomous driving simulation, the closed-loop method based on "scenario-behavior-risk" is gradually becoming the mainstream path for evaluating system safety and performance. Scenario generation, especially interactive scenarios (such as following a decelerated vehicle, changing lanes / merging, yielding / cutting off), directly determines the test exposure and the credibility of the conclusions. Therefore, there is an urgent need for behavioral models and triggering mechanisms that match the decision-making process of human drivers.

[0003] There are three main types of existing test scenario generation: rule / threshold driven (based on single indicators such as TTC (Time To Collision) and THW (Time Headway) and manual parameters); data playback / stitching (reusing measured trajectory segments in simulation); and learning-based generation (adversarial generation, diffusion, or reinforcement learning). These methods have certain advantages in terms of efficiency and controllability, but they generally treat event triggering as a deterministic threshold crossing or static prior, ignoring the hesitation-execution psychological sequence of humans under uncertain risks, thus leading to systematic deviations in trigger timing, action amplitude, and real driving. In terms of behavioral modeling, traditional car-following models (such as IDM (Intelligent Driver Model)) and lane-changing models (such as heuristic utility or game theory frameworks) mostly describe action choices from a steady-state or utility maximization perspective; although recent end-to-end learning or inverse reinforcement learning can characterize complex strategies, they are still insufficient in terms of interpretability, calibrability, and transferability, making them difficult to directly use for verifiable safety testing. In particular, the absence of hesitation behavior (wait-and-see / hesitation before decision-making) in these models makes it difficult to reproduce the waiting or delay of the simulated driver when the risk approaches a critical state. Regarding risk quantification, the industry widely uses indicators such as TTC, THW, relative speed difference, relative acceleration, and minimum distance, but these are mostly calculated using single-factor or fixed weighting methods, lacking a unified representation of longitudinal / lateral interaction coupling and scene context (speed range, traffic density, road structure), making it difficult to stably map to human triggering tendencies. The relationship between risk and triggering is often simplified to "crossing the boundary triggers," rather than a time-continuous triggering tendency / hazard rate, leading to inconsistencies between the triggering delay distribution and real-world data.

[0004] Therefore, in order to address the above problems, there is an urgent need for a risk-driven hesitation triggering modeling method for interactive scenarios. Summary of the Invention

[0005] This application provides a reliable interactive behavior generation method based on a risk-driven hesitation triggering model to solve the problems of insufficient fit between hesitation triggering behavior and real interactive scenarios and low test reliability caused by the distortion of behavior models under risk association in traditional test scenario generation methods.

[0006] The first aspect of this application provides a method for generating credible interactive behavior based on a risk-driven hesitation triggering model, comprising the following steps: acquiring lateral and longitudinal interactive behavior data of a vehicle, and identifying hesitant segments, execution segments, and stable segments of the vehicle based on the lateral and longitudinal interactive behavior data; constructing frame-by-frame labels and segment-level labels for the hesitant segments, the execution segments, and the stable segments, and constructing a multi-risk factor matrix, frame-level training corpus, segment-level training corpus, and a dual-branch network model, wherein the frame-level training corpus is obtained by extracting the frame-by-frame labels, the segment-level training corpus is obtained by extracting the segment-level labels, and the dual-branch network model is trained by the frame-level training corpus and the segment-level training corpus; inputting the multi-risk factor matrix into the dual-branch network model to obtain the trigger probability and duration distribution of the interactive behavior, and forming credible interactive behaviors for the hesitant segments, execution segments, and stable segments of the vehicle based on the trigger probability and duration distribution of the interactive behavior.

[0007] Optionally, the multi-risk factor matrix is: ; in, For the dimension of risk factors, For the normalized risk factors, Let be the normalized risk vector for frame (t).

[0008] Optionally, the trigger probability and duration distribution of the interactive behavior are as follows: ; in, For the joint probability function of hesitation-interaction triggering, It tends to be triggered instantaneously. Contribution to cumulative time distribution For the hesitation duration distribution parameters, For the accumulated time points, For integration variables, Let be the probability density function of a normal distribution.

[0009] Optionally, before identifying the vehicle's hesitation segment, execution segment, and stable segment based on the lateral and longitudinal interaction behavior data, the process includes: acquiring the vehicle's initial lateral and longitudinal interaction behavior data; performing trajectory smoothing processing and velocity and acceleration reconstruction processing on the initial lateral and longitudinal interaction behavior data to obtain the lateral and longitudinal interaction behavior data.

[0010] Optionally, the smooth trajectory and the processing method for the reconstructed velocity and acceleration are as follows: ; ; ; ; ; ; Where t is time, i is the current frame number, Δt is the time interval between two consecutive frames, x represents the vehicle's horizontal position coordinates, y represents the vehicle's vertical position coordinates, and w is the width of the sliding window. The width of the sliding window is half its width. The horizontal position values ​​are obtained by smoothing the horizontal position sequence (x) using the sliding window method. This refers to the smoothed vertical position values ​​of the vertical position sequence (y) using the sliding window method. For the lateral speed of reconstruction, For the longitudinal velocity of reconstruction, For the lateral acceleration of reconstruction, The longitudinal acceleration is for reconstruction.

[0011] Optionally, identifying the vehicle's hesitation segment, execution segment, and stable segment based on the longitudinal and lateral interaction behavior data includes: constructing an action intensity sequence for longitudinal interaction behavior and an action intensity sequence for lateral interaction behavior, and constructing a risk intensity scalar based on multi-dimensional risk factors; searching for two breakpoints in the entire time domain of the interaction behavior using an enumeration method based on the action intensity sequence of the longitudinal interaction behavior, the action intensity sequence of the lateral interaction behavior, and the risk intensity scalar, dividing the entire time domain into three continuous segments; fitting a univariate model to each of the three continuous segments, calculating the sum of squared residuals for each segment, selecting the segment combination with the smallest total sum of squared residuals as the optimal division, and obtaining three segmented segments after division; identifying the vehicle's hesitation segment, execution segment, and stable segment based on the action intensity and risk intensity of each segment of the three segmented segments after division.

[0012] Optionally, constructing the frame-level training corpus and the segment-level training corpus includes: constructing the frame-level training corpus based on the frame-level labels of the identified hesitation segments of the vehicle, wherein the frame-level training corpus includes a risk vector, a time step count, and an event trigger flag; and constructing the segment-level training corpus based on the segment-level labels of the identified hesitation segments of the vehicle, wherein the segment-level training corpus includes a duration and a segment-level average risk index.

[0013] A second aspect of this application provides a reliable interactive behavior generation system based on a risk-driven hesitation triggering model, comprising: an acquisition module for acquiring lateral and longitudinal interactive behavior data of a vehicle, and identifying hesitant segments, execution segments, and stable segments of the vehicle based on the lateral and longitudinal interactive behavior data; a construction module for constructing frame-by-frame labels and segment-level labels for the hesitant segments, the execution segments, and the stable segments, and constructing a multi-risk factor matrix, frame-level training corpus, segment-level training corpus, and a dual-branch network model, wherein the frame-level training corpus is obtained by extracting the frame-by-frame labels, the segment-level training corpus is obtained by extracting the segment-level labels, and the dual-branch network model is trained from the frame-level training corpus and the segment-level training corpus; and a generation module for inputting the multi-risk factor matrix into the dual-branch network model to obtain the trigger probability and duration distribution of the interactive behavior, and forming reliable interactive behaviors of the hesitant segments, execution segments, and stable segments of the vehicle based on the trigger probability and duration distribution of the interactive behavior.

[0014] Optionally, the multi-risk factor matrix is: ; in, For the dimension of risk factors, For the normalized risk factors, Let be the normalized risk vector for frame (t).

[0015] Optionally, the trigger probability and duration distribution of the interactive behavior are as follows: ; in, For the joint probability function of hesitation-interaction triggering, It tends to be triggered instantaneously. Contribution to cumulative time distribution For the hesitation duration distribution parameters, For the accumulated time points, For integration variables, Let be the probability density function of a normal distribution.

[0016] Optionally, before identifying the vehicle's hesitation segment, execution segment, and stabilization segment based on the lateral and longitudinal interaction behavior data, the acquisition module is further configured to: acquire the vehicle's initial lateral and longitudinal interaction behavior data; and perform trajectory smoothing processing and velocity and acceleration reconstruction processing on the initial lateral and longitudinal interaction behavior data to obtain the lateral and longitudinal interaction behavior data.

[0017] Optionally, the smooth trajectory and the processing method for the reconstructed velocity and acceleration are as follows: ; ; ; ; ; ; Where t is time, i is the current frame number, Δt is the time interval between two consecutive frames, x represents the vehicle's horizontal position coordinates, y represents the vehicle's vertical position coordinates, and w is the width of the sliding window. The width of the sliding window is half its width. The horizontal position values ​​are obtained by smoothing the horizontal position sequence (x) using the sliding window method. This refers to the smoothed vertical position values ​​of the vertical position sequence (y) using the sliding window method. For the lateral speed of reconstruction, For the longitudinal velocity of reconstruction, For the lateral acceleration of reconstruction, The longitudinal acceleration is for reconstruction.

[0018] Optionally, the acquisition module is further configured to: construct an action intensity sequence of longitudinal interaction behavior and an action intensity sequence of lateral interaction behavior, and construct a risk intensity scalar based on multi-dimensional risk factors; search for two breakpoints in the entire time domain of the interaction behavior using an enumeration method based on the action intensity sequence of the longitudinal interaction behavior, the action intensity sequence of the lateral interaction behavior, and the risk intensity scalar, and divide the entire time domain into three continuous segments; fit a univariate model to each of the three continuous segments, calculate the sum of squared residuals for each segment, select the segment combination with the smallest total sum of squared residuals as the optimal segmentation, and obtain the three segmented segments after division; identify the hesitation segment, execution segment, and stable segment of the vehicle based on the action intensity and risk intensity of each segment of the three segmented segments after division.

[0019] Optionally, the construction module is further configured to: construct the frame-level training corpus based on the frame-level labels of the identified hesitation segments of the vehicle, the frame-level training corpus including a risk vector, a time step count, and an event trigger flag; and construct the segment-level training corpus based on the segment-level labels of the identified hesitation segments of the vehicle, the segment-level training corpus including a duration and a segment-level average risk index.

[0020] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the program to implement the trusted interactive behavior generation method based on the risk-driven hesitation triggering model as described in the above embodiments.

[0021] A fourth aspect of this application provides a computer program product having a computer program stored thereon, which is executed by a processor to implement the trusted interactive behavior generation method based on the risk-driven hesitation triggering model as described in the above embodiments.

[0022] In the above implementation, the vehicle's hesitation, execution, and stability segments are identified based on its lateral and longitudinal interaction behavior data. Frame-by-frame and segment-level labels are constructed for these segments. A multi-risk factor matrix, frame-level training corpus, segment-level training corpus, and a dual-branch network model are also built. The frame-level training corpus is obtained by extracting frame-by-frame labels, the segment-level training corpus by extracting segment-level labels, and the dual-branch network model is trained using both the frame-level and segment-level training corpora. The multi-risk factor matrix is ​​input into the dual-branch network model to obtain the trigger probability and duration distribution of the interaction behavior. Based on these distributions, reliable interaction behaviors for the vehicle's hesitation, execution, and stability segments are formed. This solves the problems of insufficient alignment between hesitant trigger behaviors and real-world interaction scenarios, and low test reliability due to distorted behavioral models under risk association, which are common in traditional test scenario generation methods. It generates reliable interaction test scenarios that are closer to human behavioral characteristics, improving the reliability and effectiveness of the scenarios and effectively enhancing the reliability and effectiveness of intelligent connected vehicle testing.

[0023] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0024] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart of a method for generating trusted interactive behaviors based on a risk-driven hesitation triggering model, according to an embodiment of this application. Figure 2This is an example diagram of the first typical lateral interaction trajectory recognition for highways provided in the embodiments of this application; Figure 3 Example diagram of the construction of the first typical longitudinal interactive trajectory label for a highway provided in this application embodiment; Figure 4 Example diagram of constructing a second typical lateral interaction trajectory label for a highway provided in this application embodiment; Figure 5 Figure showing the training results of the lateral interaction P-hazard and P-duration models provided in the embodiments of this application; Figure 6 The training results of the longitudinal interaction P-hazard and P-duration models provided in the embodiments of this application are shown in the figure. Figure 7 The diagram illustrates the integrated effect of hesitant triggering of interactive behavior in the joint deployment of P-hazardNet and P-durationNet, as provided in the embodiments of this application. Figure 8 This is an example diagram of a trusted interaction behavior generation system based on a risk-driven hesitation triggering model according to an embodiment of this application; Figure 9 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0025] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0026] The following describes a method for generating credible interactive behavior based on a risk-driven hesitation triggering model, according to embodiments of this application, with reference to the accompanying drawings. Addressing the problems mentioned in the background art, such as insufficient fit between hesitation-triggered behavior and real-world interaction scenarios in traditional test scenario generation methods, and the distortion of behavioral models under risk association leading to low test credibility, this application provides a method for generating credible interactive behavior based on a risk-driven hesitation triggering model. In this method, the hesitation segment, execution segment, and stable segment of a vehicle are identified based on the vehicle's horizontal and vertical interactive behavior data; frame-by-frame labels and segment-level labels for the hesitation segment, execution segment, and stable segment are constructed; a multi-risk factor matrix, frame-level training corpus, segment-level training corpus, and a dual-branch network model are constructed. The frame-level training corpus is obtained by extracting frame-by-frame labels, the segment-level training corpus is obtained by extracting segment-level labels, and the dual-branch network model is trained using the frame-level and segment-level training corpora; the multi-risk factor matrix is ​​input into the dual-branch network model to obtain the trigger probability and duration distribution of the interactive behavior; and credible interactive behaviors for the vehicle's hesitation segment, execution segment, and stable segment are formed based on the trigger probability and duration distribution of the interactive behavior. This solves the problems of insufficient alignment between hesitation-triggered behaviors and real-world interaction scenarios, and low test credibility caused by distorted behavioral models under risk association in traditional test scenario generation methods. It generates more credible interactive test scenarios that are closer to human behavioral characteristics, improving the credibility and effectiveness of the scenarios and effectively enhancing the credibility and effectiveness of intelligent connected vehicle testing.

[0027] Specifically, Figure 1 This is a flowchart illustrating a method for generating trusted interactive behaviors based on a risk-driven hesitation triggering model, as provided in an embodiment of this application.

[0028] like Figure 1 As shown, the method for generating trustworthy interactive behavior based on a risk-driven hesitation triggering model includes the following steps: In step S101, the vehicle's lateral and longitudinal interaction behavior data are acquired, and the vehicle's hesitation segment, execution segment, and stability segment are identified based on the lateral and longitudinal interaction behavior data. Optionally, in some embodiments, before identifying the vehicle's hesitation segment, execution segment, and stable segment based on the lateral and longitudinal interaction behavior data, the method includes: acquiring the vehicle's initial lateral and longitudinal interaction behavior data; performing trajectory smoothing processing and speed and acceleration reconstruction processing on the initial lateral and longitudinal interaction behavior data to obtain the lateral and longitudinal interaction behavior data.

[0029] Optionally, in some embodiments, the smoothing trajectory and the processing of reconstructed velocity and acceleration are as follows:

[0030] ;

[0031] ; (1)

[0032] Where t represents time, usually in seconds (s), i is the current frame number, indicating which frame in the data sequence it is, counting from 1, Δt is the time interval between two consecutive frames, usually in seconds (s), x represents the vehicle's horizontal position coordinates, usually in meters (m), y represents the vehicle's vertical position coordinates, usually in meters (m), and w is the width of the sliding window, with a window size of 2k0+1. This represents the half-width of the sliding window, where k0 is the number of frames extended from the center of the window to both sides, in frames only, and has no unit. The sliding window half-width indicates the number of frames extended forward and backward from the current frame, in frames. The horizontal position values ​​are the smoothed horizontal position sequence (x) using the sliding window method, in meters (m). The vertical position values ​​are the smoothed vertical position sequence (y) using the sliding window method, in meters (m). The reconstructed lateral velocity is calculated by taking the derivative of the smoothed position sequence x using the central difference formula, with units of meters per second (m / s). The longitudinal velocity for reconstruction is calculated by taking the derivative of the smoothed position sequence y using the central difference formula, with units of meters per second (m / s). The reconstructed lateral acceleration is calculated using the central difference formula for the reconstructed lateral velocity. Take the derivative again, with the unit being meters per second squared (m / s²). The longitudinal acceleration for reconstruction is calculated using the central difference formula for the reconstructed longitudinal velocity. Take the derivative again, with the unit being meters per second squared (m / s²).

[0033] Optionally, in some embodiments, identifying the vehicle's hesitation segment, execution segment, and stable segment based on longitudinal and lateral interaction behavior data includes: constructing action intensity sequences for longitudinal and lateral interaction behaviors, and constructing a risk intensity scalar based on multi-dimensional risk factors; searching for two breakpoints in the entire time domain of the interaction behavior using an enumeration method based on the action intensity sequences for longitudinal and lateral interaction behaviors and the risk intensity scalar, dividing the entire time domain into three continuous segments; fitting a univariate model to each of the three continuous segments, calculating the sum of squared residuals for each segment, and selecting the segment combination with the smallest total sum of squared residuals as the optimal segmentation, resulting in three segmented segments; and identifying the vehicle's hesitation segment, execution segment, and stable segment based on the action intensity and risk intensity of each segment of the three segmented segments.

[0034] Specifically, interaction behaviors are extracted in advance according to longitudinal and lateral interaction behaviors. Based on the segment feature analysis of the vehicle's longitudinal and lateral interaction behaviors, interaction behavior extraction rules are designed.

[0035] The extraction rule for typical vertical interaction behavior is as follows: For vertical interaction, first calculate the total weight (THW) of the two interacting vehicles, requiring... The vehicle in front should slow down first. Furthermore, during the interaction period, TTC dropped below the threshold, among which... For the moment when the vehicle in front slows down, The above indicates that the deceleration time of the following vehicle lags behind the deceleration time of the preceding vehicle. The extraction rule for typical lateral interaction behaviors is as follows: For lateral lane change interactions, first calculate the lateral displacement of the interacting main vehicle, requiring the lateral displacement within a certain time t. Lane width Furthermore, the vehicle was in a stable driving phase before changing lanes.

[0036] Based on the collected real vehicle trajectory dataset, in order to suppress vehicle sensing noise, a window moving average is first applied to the position sequence, and then the velocity and acceleration are reconstructed using the central difference, which makes the hesitation segmentation and risk calculation more robust.

[0037] In step S102, frame-by-frame labels and segment-level labels are constructed for the hesitant segment, the execution segment, and the stable segment. A multi-risk factor matrix, frame-level training corpus, segment-level training corpus, and a dual-branch network model are also constructed. The frame-level training corpus is obtained by extracting frame-by-frame labels, the segment-level training corpus is obtained by extracting segment-level labels, and the dual-branch network model is trained using the frame-level training corpus and the segment-level training corpus.

[0038] In step S103, the multi-risk factor matrix is ​​input into the dual-branch network model to obtain the trigger probability and duration distribution of the interaction behavior, and the reliable interaction behavior of the vehicle in the hesitation segment, execution segment and stable segment is formed according to the trigger probability and duration distribution of the interaction behavior.

[0039] Optionally, in some embodiments, constructing frame-level training corpora and segment-level training corpora includes: constructing frame-level training corpora based on frame-level labels of the identified vehicle's hesitation segments, wherein the frame-level training corpora include risk vectors, time step counts, and event trigger flags; and constructing segment-level training corpora based on segment-level labels of the identified vehicle's hesitation segments, wherein the segment-level training corpora include duration and segment-level average risk index.

[0040] Optionally, in some embodiments, the multi-risk factor matrix is: (2) in, For the dimension of risk factors, For the normalized risk factors, Let be the normalized risk vector for frame (t).

[0041] Optionally, in some embodiments, the probability of triggering the interaction behavior and its duration are distributed as follows: (3) in, For the joint probability function of hesitation-interaction triggering, It tends to be triggered instantaneously. Contribution to cumulative time distribution For the hesitation duration distribution parameters, For the accumulated time points, For integration variables, Let be the probability density function of a normal distribution.

[0042] Based on the obtained smooth trajectory and kinematic quantities, the hesitant segment, the executive segment, and the stable segment are identified, and hesitant labels (frame-by-frame labels and segment-level labels) that can be used for learning are constructed.

[0043] Constructing action intensity sequence This is used to indicate whether explicit execution has commenced. Depending on the interaction type, the following forms can be selected: Vertical interaction: ; Lateral lane change interaction: (4) Construct risk intensity scalar The result is obtained by weighting the factors in each dimension of the multi-risk factor matrix: (5) in Let (j) be the normalized value of the (j)th risk factor (e.g., TTC, THW). The weights are (either given equally or learned implicitly during training). The discrimination mode of the hesitant segment is: (6) in The duration of the hesitation phase. As the lowest risk threshold, This is the upper limit of "non-executive action intensity". This is the minimum hesitation time threshold.

[0044] The hesitation criterion is "risk background exists + low action intensity + persistent delay". That is, based on the action intensity sequence, the interaction process is divided into three segments: P (hesitation), E (execution), and S (stable). Using an enumeration method, two breakpoints are searched throughout the entire time domain to divide the process into three segments. A simple univariate model is fitted to each segment, and the segment combination with the smallest sum of squared residuals is selected as the optimal segmentation. The statistics of each segment are then mapped to... .

[0045] (7) ; in, The sequence duration of each segment is represented by t, where t represents the current time and T represents the end time of this trajectory sequence.

[0046] (8) in, The model is fitted to segment (s). and This represents the positions of two breakpoints, i.e., an interactive trajectory sequence. If it is to be divided into three segments, then two points are required because... This is represented as the dividing point between the hesitation phase and the execution phase. This indicates the dividing point between the execution segment and the stable segment. This represents the sum of squared residuals during the hesitation segment. This represents the sum of squared residuals of the execution segment. This represents the sum of squared residuals during the stable phase. , SSEs represents the position of the points after segmentation, Ω1, Ω2, and Ω3 represent the time combination of each segment after identification.

[0047] Furthermore, a multi-risk factor matrix and training corpus are constructed based on the scene interaction characteristics. The risk matrix gathers interaction risk factors from multiple dimensions (such as vertical speed difference, horizontal spacing, relative acceleration, TTC, etc.), and after normalization, forms a multi-risk factor matrix, as shown in Equation (8): ,in The number of risk indicators selected determines the outcome. This represents the normalized risk factor, calculated by normalizing the statistics in the training set.

[0048] And record the corresponding state sequence: (9) in, This represents the state at time t.

[0049] For each P segment, construct a P-Hazard-Net training corpus within a time window near the end, including risk vectors, time step counts, and event trigger flags: (10) in, The training corpus for P-Hazard-Net, Indicates relative clock counting within the segment. As an indicator variable for risk events, Used to align sample lengths.

[0050] Finally, P-Duration-Net training corpus is constructed for each P segment by segment-level summary, and the duration is extracted. Segment-level average risk indicators : ; (11) Where Tend represents the end time of the hesitant segment and tstart represents the start time of the hesitant segment.

[0051] Two types of training samples were generated: frame-level triggering P-Hazard-Net and segment-level duration P-Duration-Net, providing a data foundation for subsequent risk and trigger latency modeling.

[0052] A dual-branch joint triggering network is constructed, comprising an interaction frame-level triggering network P-Hazard-Net and a hesitation duration segment-level distribution network P-Duration-Net. The former performs frame-by-frame inference on the risk time series, while the latter models the overall risk statistics. The two jointly output the trigger probability and duration distribution of the interaction behavior.

[0053] Trusted interactive behaviors are generated based on the hesitation-execution pattern. The "hesitation-execution pattern" refers to the state sequence and its parameterized description, consisting of the hesitation phase (P), the execution phase (E), and the stable phase (S), under given risk levels and interaction conditions. During the generation phase, trusted interactive behaviors are generated for each type of interaction scenario according to the following process: 1. Online calculation of risk matrix: During simulation time steps The scenario is pushed forward, and the risk vector is calculated frame by frame. .

[0054] 2. Frame-by-frame inference of trigger probability and departure motivation: Input the current state of the interactive vehicle and risk variables into the dual-branch joint triggering network. ,calculate .

[0055] 3. Determine the trigger time Threshold decision: when At that time, determine , The moment when the interaction is triggered ends due to hesitation. This is the trigger threshold; 4. Generation of P-segment (hesitation segment) behavior: In During this period, the constraint intensity remains low and allows for tentative small-scale operations to satisfy the hesitation characteristic (e.g., restrictions). (etc.), but the risks can gradually change as the interaction evolves.

[0056] 5. Generate E-segment (execution segment) behavior: In Then, based on E, a specific action is generated, and the lateral lane change is generated in the form of a fifth-order polynomial.

[0057] 6. Generation of S-segment (stable segment) behavior: After execution, it enters stabilization control, adjusting the execution intensity. The indicators fall back and remain within a stable range, such as maintaining target headway, maintaining lane centering, and maintaining a stable speed.

[0058] Taking the generation of realistic and reliable interactive behaviors in a highway scene as an example, we first extract typical longitudinal and lateral interactive behaviors based on the highway scene dataset. According to the extraction rules for typical longitudinal interactive behaviors: longitudinal interaction first calculates the total power (THW) of the two interacting vehicles, requiring... The vehicle in front should slow down first. Furthermore, during the interaction period, the TTC drops below the threshold; the extraction rules for typical lateral interaction behaviors are as follows: for lateral lane change interactions, first calculate the lateral displacement of the interacting main vehicle, requiring the lateral displacement within a certain time t. The vehicle reaches the lane width and has a stable driving phase before changing lanes.

[0059] Secondly, a center-based segmentation method is used to smooth the trajectory data of interactive vehicle behavior in the highway dataset, eliminating unreasonable spikes in position, speed, and acceleration, as well as decoupled regions. ; ; ; ; ; .

[0060] After obtaining the smooth interactive behavior trajectory, the interaction process is divided into three segments: P (preparation segment), E (execution segment), and S (stable segment). Using an enumeration method, two breakpoints are searched throughout the entire time domain to divide the process into three segments. A simple univariate model is fitted to each segment, and the segment combination with the smallest sum of squared residuals is selected as the optimal segmentation. The segment is then mapped to its statistical value based on the statistics of each segment. Typical lateral interaction trajectories on highways are as follows: Figure 2 As shown, the label construction results of the lateral interaction behavior trajectory are as follows: Figure 3 As shown, the label construction results of the vertical interaction behavior trajectory are as follows: Figure 4 As shown.

[0061] Based on the aforementioned rules for constructing hesitation labels, a dataset of interactive behavior trajectory data is built. Simultaneously, scene interaction characteristics are constructed to build a multi-risk factor matrix and training corpus. A dual-branch network model is constructed, including an interaction frame-level triggering P-Hazard-Net and a hesitation duration segment-level distribution P-Duration-Net. The former performs frame-by-frame reasoning on the risk time series, while the latter models the overall risk statistics. The two jointly output the trigger probability and duration distribution of the interactive behavior.

[0062] The training metrics for the lateral interaction models P-Hazard-Net and P-Duration-Net are as follows: Figure 5 As shown in Table 1, the training results metrics for the horizontal interaction P-hazard and P-duration models are presented.

[0063] Table 1

[0064] The training metrics for the longitudinal interaction models P-Hazard-Net and P-Duration-Net are as follows: Figure 6 As shown in Table 2, the training results metrics for the longitudinal interaction P-hazard and P-duration models are presented.

[0065] Table 2

[0066] Based on the trigger probability and duration distribution of the joint output interaction behavior, the environmental vehicles in the test scenario generation process are controlled, and the interaction duration distribution of the three vehicles can be obtained as follows: Figure 7 As shown in the figure, the blue shaded area represents the vehicle's hesitant behavior trajectory before lateral interaction, and the pink area represents the vehicle's hesitant behavior trajectory before longitudinal interaction.

[0067] The reliable interactive behavior generation method based on a risk-driven hesitation triggering model proposed in this application identifies the hesitation segment, execution segment, and stable segment of a vehicle based on its horizontal and vertical interactive behavior data. It constructs frame-by-frame and segment-level labels for these segments, and builds a multi-risk factor matrix, frame-level training corpus, segment-level training corpus, and a dual-branch network model. The frame-level training corpus is obtained by extracting frame-by-frame labels, the segment-level training corpus is obtained by extracting segment-level labels, and the dual-branch network model is trained using both the frame-level and segment-level training corpora. The multi-risk factor matrix is ​​input into the dual-branch network model to obtain the trigger probability and duration distribution of interactive behavior. Based on these distributions, reliable interactive behaviors for the vehicle's hesitation, execution, and stable segments are formed. This solves the problems of insufficient fit between hesitation triggering behavior and real-world interactive scenarios, and low test reliability due to distorted behavioral models under risk association in traditional test scenario generation methods. It generates reliable interactive test scenarios that are closer to human behavioral characteristics, improving the reliability and effectiveness of the scenarios and effectively enhancing the reliability and effectiveness of intelligent connected vehicle testing.

[0068] Next, referring to the accompanying drawings, a trustworthy interactive behavior generation system based on a risk-driven hesitation triggering model, according to an embodiment of this application, is described.

[0069] Figure 8 This is a block diagram of a trusted interactive behavior generation system based on a risk-driven hesitation triggering model, according to an embodiment of this application.

[0070] like Figure 8 As shown, the trustworthy interaction behavior generation system 10 based on the risk-driven hesitation triggering model includes: an acquisition module 100, a construction module 200, and a generation module 300.

[0071] The system comprises the following modules: an acquisition module 100, which acquires lateral and longitudinal interaction behavior data of the vehicle and identifies the vehicle's hesitation, execution, and stability segments based on this data; a construction module 200, which constructs frame-by-frame and segment-level labels for the hesitation, execution, and stability segments, and builds a multi-risk factor matrix, frame-level training corpus, segment-level training corpus, and a dual-branch network model. The frame-level training corpus is obtained by extracting frame-by-frame labels, the segment-level training corpus is obtained by extracting segment-level labels, and the dual-branch network model is trained using both the frame-level and segment-level training corpora; and a generation module 300, which inputs the multi-risk factor matrix into the dual-branch network model to obtain the trigger probability and duration distribution of the interaction behavior, and forms reliable interaction behaviors for the vehicle's hesitation, execution, and stability segments based on these distributions.

[0072] Optionally, in some embodiments, the multi-risk factor matrix is: ; in, For the dimension of risk factors, For the normalized risk factors, Let be the normalized risk vector for frame (t).

[0073] Optionally, in some embodiments, the probability of triggering the interaction behavior and its duration are distributed as follows: ; in, For the joint probability function of hesitation-interaction triggering, It tends to be triggered instantaneously. Contribution to cumulative time distribution For the hesitation duration distribution parameters, For the accumulated time points, For integration variables, Let be the probability density function of a normal distribution.

[0074] Optionally, in some embodiments, before identifying the vehicle's hesitation segment, execution segment, and stabilization segment based on the lateral and longitudinal interaction behavior data, the acquisition module 100 is further configured to: acquire the vehicle's initial lateral and longitudinal interaction behavior data; perform smooth trajectory processing and reconstruct speed and acceleration processing on the initial lateral and longitudinal interaction behavior data to obtain the lateral and longitudinal interaction behavior data.

[0075] Optionally, in some embodiments, the smoothing trajectory and the processing of reconstructed velocity and acceleration are as follows: ; ; ; ; ; ; Where t is time, i is the current frame number, Δt is the time interval between two consecutive frames, x represents the vehicle's horizontal position coordinates, y represents the vehicle's vertical position coordinates, and w is the width of the sliding window. The width of the sliding window is half its width. The horizontal position values ​​are obtained by smoothing the horizontal position sequence (x) using the sliding window method. This refers to the smoothed vertical position values ​​of the vertical position sequence (y) using the sliding window method. For the lateral speed of reconstruction, For the longitudinal velocity of reconstruction, For the lateral acceleration of reconstruction, The longitudinal acceleration is for reconstruction.

[0076] Optionally, in some embodiments, the acquisition module 100 is further configured to: construct the action intensity sequence of longitudinal interaction behavior and the action intensity sequence of lateral interaction behavior, and construct a risk intensity scalar based on multi-dimensional risk factors; search for two breakpoints in the entire time domain of interaction behavior using an enumeration method based on the action intensity sequence of longitudinal interaction behavior, the action intensity sequence of lateral interaction behavior and the risk intensity scalar, and divide the entire time domain into three continuous segments; fit a univariate model to each of the three continuous segments, calculate the residual sum of squares for each segment, select the segment combination with the smallest total residual sum of squares as the optimal segmentation, and obtain the three segmented segments after division; identify the vehicle's hesitation segment, execution segment and stable segment based on the action intensity and risk intensity of each segment of the three segmented segments after division.

[0077] Optionally, in some embodiments, the construction module 200 is further configured to: construct frame-level training corpus based on frame-level labels of the identified vehicle's hesitation segment, the frame-level training corpus including risk vector, time step count and event trigger flag; and construct segment-level training corpus based on segment-level labels of the identified vehicle's hesitation segment, the segment-level training corpus including duration and segment-level average risk index.

[0078] It should be noted that the foregoing explanation of the embodiment of the trusted interactive behavior generation method based on the risk-driven hesitation triggering model also applies to the trusted interactive behavior generation system based on the risk-driven hesitation triggering model of this embodiment, and will not be repeated here.

[0079] The trusted interactive behavior generation system based on a risk-driven hesitation triggering model proposed in this application identifies the hesitation segment, execution segment, and stable segment of a vehicle based on its lateral and longitudinal interactive behavior data. It constructs frame-by-frame and segment-level labels for these segments, and builds a multi-risk factor matrix, frame-level training corpus, segment-level training corpus, and a dual-branch network model. The frame-level training corpus is obtained by extracting frame-by-frame labels, the segment-level training corpus is obtained by extracting segment-level labels, and the dual-branch network model is trained using both the frame-level and segment-level training corpora. The multi-risk factor matrix is ​​input into the dual-branch network model to obtain the trigger probability and duration distribution of interactive behaviors. Based on these distributions, trusted interactive behaviors for the vehicle's hesitation, execution, and stable segments are formed. This solves the problems of insufficient relevance between hesitant triggering behaviors and real-world interactive scenarios, and low test credibility due to distorted behavioral models under risk association in traditional test scenario generation methods. It generates trusted interactive test scenarios that are closer to human behavioral characteristics, improving the credibility and effectiveness of the scenarios and effectively enhancing the credibility and effectiveness of intelligent connected vehicle testing.

[0080] Figure 9 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: The memory 901, the processor 902, and the computer program stored on the memory 901 and capable of running on the processor 902.

[0081] When the processor 902 executes the program, it implements the trusted interaction behavior generation method based on the risk-driven hesitation triggering model provided in the above embodiments.

[0082] Furthermore, electronic devices also include: Communication interface 903 is used for communication between memory 901 and processor 902.

[0083] The memory 901 is used to store computer programs that can run on the processor 902.

[0084] The memory 901 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0085] If the memory 901, processor 902, and communication interface 903 are implemented independently, then the communication interface 903, memory 901, and processor 902 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 9 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0086] Optionally, in a specific implementation, if the memory 901, processor 902, and communication interface 903 are integrated on a single chip, then the memory 901, processor 902, and communication interface 903 can communicate with each other through an internal interface.

[0087] The processor 902 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0088] This application also provides a computer program product on which a computer program is stored, which, when executed by a processor, implements the above-described method for generating trusted interactive behaviors based on a risk-driven hesitation triggering model.

[0089] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0090] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0091] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0092] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequential list of executable instructions for implementing logical functions, and can be specifically implemented in any computer program product for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer program product" can be any means that can contain, store, communicate, propagate, or transmit a program for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples of computer program products (a non-exhaustive list) include the following: an electrical connection having one or N wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic device, and portable optical disc read-only memory (CDROM). Furthermore, the computer program product can even be paper or other suitable medium on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0093] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0094] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer program product, and when executed, it includes one or a combination of the steps of the method embodiments.

[0095] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer program product.

[0096] The computer program product mentioned above may be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A method for generating trustworthy interactive behaviors based on a risk-driven hesitation triggering model, characterized in that, Includes the following steps: Acquire lateral and longitudinal interaction behavior data of the vehicle, and identify the vehicle's hesitation segment, execution segment, and stable segment based on the lateral and longitudinal interaction behavior data; Frame-by-frame and segment-level labels are constructed for the hesitant segment, the execution segment, and the stable segment. A multi-risk factor matrix, frame-level training corpus, segment-level training corpus, and a dual-branch network model are also constructed. The frame-level training corpus is obtained by extracting the frame-by-frame labels, the segment-level training corpus is obtained by extracting the segment-level labels, and the dual-branch network model is trained using the frame-level and segment-level training corpuses. The multi-risk factor matrix is ​​input into the dual-branch network model to obtain the trigger probability and duration distribution of the interaction behavior, and the reliable interaction behavior of the vehicle in the hesitation segment, execution segment and stable segment is formed according to the trigger probability and duration distribution of the interaction behavior.

2. The method according to claim 1, characterized in that, The multi-risk factor matrix is ​​as follows: ； in, For the dimension of risk factors, For the normalized risk factors, Let be the normalized risk vector for frame (t).

3. The method according to claim 1, characterized in that, The probability distribution of the interaction behavior and its duration are as follows: ； in, For the joint probability function of hesitation-interaction triggering, It tends to be triggered instantaneously. Contribution to cumulative time distribution For the hesitation duration distribution parameters, For the accumulated time points, For integration variables, Let be the probability density function of a normal distribution.

4. The method according to claim 1, characterized in that, Before identifying the vehicle's hesitation phase, execution phase, and stability phase based on the aforementioned lateral and longitudinal interaction behavior data, the process includes: Acquire the initial lateral and longitudinal interaction behavior data of the vehicle; The initial horizontal and vertical interaction behavior data is processed by smoothing the trajectory and reconstructing the velocity and acceleration to obtain the horizontal and vertical interaction behavior data.

5. The method according to claim 4, characterized in that, The processing methods for the smooth trajectory and the reconstructed velocity and acceleration are as follows: ； ； ； ； ； ； Where t is time, i is the current frame number, Δt is the time interval between two consecutive frames, x represents the vehicle's horizontal position coordinates, y represents the vehicle's vertical position coordinates, and w is the width of the sliding window. The width of the sliding window is half its width. The horizontal position values ​​are obtained by smoothing the horizontal position sequence (x) using the sliding window method. This refers to the smoothed vertical position values ​​of the vertical position sequence (y) using the sliding window method. For the lateral speed of reconstruction, For the longitudinal velocity of reconstruction, For the lateral acceleration of reconstruction, The longitudinal acceleration is for reconstruction.

6. The method according to claim 1, characterized in that, The step of identifying the vehicle's hesitation phase, execution phase, and stability phase based on the horizontal and vertical interaction behavior data includes: Construct action intensity sequences for vertical and horizontal interactive behaviors, and construct risk intensity scalars based on multi-dimensional risk factors; Based on the action intensity sequence of the vertical interaction behavior, the action intensity sequence of the horizontal interaction behavior, and the risk intensity scalar, an enumeration method is used to search for two breakpoints in the entire time domain of the interaction behavior, and the entire time domain is divided into three continuous segments. A univariate model is fitted to each of the three continuous segments, the sum of squared residuals for each segment is calculated, and the segment combination with the smallest total sum of squared residuals is selected as the optimal partition, resulting in three segmented segments. The vehicle's hesitation segment, execution segment, and stability segment are identified based on the action intensity and risk intensity of each of the three segmented segments.

7. The method according to claim 1, characterized in that, Constructing the frame-level training corpus and the segment-level training corpus includes: The frame-level training corpus is constructed based on the frame-level labels of the identified hesitation segments of the vehicle. The frame-level training corpus includes risk vectors, time step counts, and event trigger flags. A segment-level training corpus is constructed based on the segment-level labels of the identified hesitation segments of the vehicle, and the segment-level training corpus includes the duration and the segment-level average risk index.

8. A trustworthy interactive behavior generation system based on a risk-driven hesitation triggering model, characterized in that, include: The acquisition module is used to acquire the vehicle's lateral and longitudinal interaction behavior data, and to identify the vehicle's hesitation segment, execution segment, and stable segment based on the lateral and longitudinal interaction behavior data; A construction module is used to construct frame-by-frame labels and segment-level labels for the hesitant segment, the execution segment, and the stable segment, and to construct a multi-risk factor matrix, frame-level training corpus, segment-level training corpus, and a dual-branch network model. The frame-level training corpus is obtained by extracting the frame-by-frame labels, the segment-level training corpus is obtained by extracting the segment-level labels, and the dual-branch network model is trained using the frame-level training corpus and the segment-level training corpus. The generation module is used to input the multi-risk factor matrix into the dual-branch network model to obtain the trigger probability and duration distribution of the interaction behavior, and to form the reliable interaction behavior of the vehicle's hesitation segment, execution segment and stable segment based on the trigger probability and duration distribution of the interaction behavior.

9. An electronic device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the trusted interactive behavior generation method based on the risk-driven hesitation triggering model as described in any one of claims 1-7.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the trusted interactive behavior generation method based on the risk-driven hesitation triggering model as described in any one of claims 1-7.

Images