Vehicle control method, electronic device, and vehicle

Abstract

Embodiments of the present application provide a vehicle control method, an electronic device and a vehicle. The method determines semantic risk feature data and a physical risk score according to environment data and map data. The semantic risk feature data includes static scene labels, dynamic scene chaos degrees, and behavior intentions and intention index parameters of at least one traffic participant. The method determines a semantic risk score according to the semantic risk feature data, and determines a comprehensive risk score according to the physical risk score and the semantic risk score. The method determines a risk level according to the comprehensive risk score and a preset risk threshold, and controls the vehicle to perform a driving operation corresponding to the risk level. The embodiments improve the rationality of vehicle control decisions through multi-dimensional semantic understanding at the static scene level, the dynamic scene level and the intention level. Meanwhile, the embodiments improve the explainability and predictability of vehicle control decisions in complex dynamic environments by fusing semantic risk features and physical state risks, and improve the driving comfort experience.

Description

Technical Field

[0001] This application relates to the field of vehicle control technology, and in particular to a vehicle control method, electronic device, and vehicle. Background Technology

[0002] With the development of artificial intelligence and driver assistance technologies, the demand for intelligent control in autonomous driving is increasing. Current technologies employ rule-based threshold triggering of predetermined actions, relying solely on simple distance and speed sensor data for decision-making. This lack of deep semantic understanding of the environment leads to unstable perception results, causing deviations in subsequent path planning and control strategies, and hindering refined and rational control decisions. For example, different target trajectories need to be planned for pedestrians who are observing from the roadside and those who are jaywalking. Other vehicle control methods use model training to establish a mapping strategy from perception to control commands. However, this approach makes vehicle control adjustments a "black box," resulting in poor interpretability and predictability. In risk scenarios not covered by training data, behavior becomes uncontrollable, reducing the comfort experience for drivers and passengers. Summary of the Invention

[0003] This application provides a vehicle control method, electronic device, and vehicle, aiming to improve the problems of insufficient rationality in driving decisions and poor comfort experience in vehicle control.

[0004] A vehicle control method, comprising: Based on environmental and map data, semantic risk feature data and physical risk scores are determined; the semantic risk feature data includes static scene labels, dynamic scene confusion level, and behavioral intentions and intention index parameters of at least one traffic participant. Based on the semantic risk feature data, a semantic risk score is determined, and based on the physical risk score and the semantic risk score, a comprehensive risk score is determined; Based on the comprehensive risk score and the preset risk threshold, the risk level is determined, and the vehicle is controlled to perform driving operations corresponding to the risk level.

[0005] This embodiment performs multi-dimensional semantic understanding of environmental and map data at the static, dynamic, and intent levels to generate structured semantic risk features. This ensures the comprehensiveness of environmental semantics between vehicles and traffic participants, contributing to improved precision and rationality in subsequent trajectory planning. Furthermore, this embodiment analyzes the physical state risk of vehicles affected by traffic participants based on environmental and map data. By leveraging semantic risk features, it can identify potential risks caused by malicious intent earlier. The comprehensive risk score, which integrates semantic risk features and physical state risk, is used to classify different risk levels and adjust vehicle trajectory planning accordingly. This improves the interpretability and predictability of vehicle control decisions in complex dynamic environments, ultimately enhancing the driving comfort experience.

[0006] Furthermore, the determination of semantic risk feature data based on environmental data and map data includes: Based on environmental and map data, determine static and dynamic scene information; The static scene label is determined based on the road type, traffic signs, and lane topology type in the static scene information; Based on the dynamic scene information, determine the dynamic scene disorder level and at least one of the traffic participants' behavioral intentions and intention index parameters; Semantic risk feature data is determined based on the static scene labels, the dynamic scene disorder, and the behavioral intentions and intention index parameters of all traffic participants.

[0007] This embodiment ensures the environmental semantic comprehensiveness of semantic risk features by performing multi-dimensional semantic understanding of environmental data and map data at the static scene level, dynamic scene level, and intent level, thereby improving the accuracy and interpretability of subsequent risk assessment.

[0008] Further, determining the dynamic scene disorder level and the behavioral intentions and intention index parameters of at least one traffic participant based on the dynamic scene information includes: The degree of chaos in the dynamic scene is determined based on the participant type, movement direction, and speed of all traffic participants in the dynamic scene information. Based on the participant's location, speed, acceleration, and historical trajectory in the dynamic scene information, the behavioral intention and intention index parameters of each participant are determined.

[0009] This embodiment performs targeted analysis of data from different dimensions in dynamic scene information at both the dynamic scene level and the intent level, ensuring the accuracy of dynamic scene disorder and improving the predictability of traffic participants' behavioral intentions and intent indicator parameters.

[0010] Further, determining the dynamic scene disorder level based on the participant type, participant movement direction, and participant speed of all traffic participants in the dynamic scene information includes: Determine the type entropy based on the participant type of all the traffic participants; Determine the directional dispersion based on the direction of movement of all traffic participants; Determine the speed standard deviation based on the participant speeds of all traffic participants; The type entropy, the directional dispersion, and the velocity standard deviation are weighted according to a preset disorder weight to obtain the dynamic scene disorder.

[0011] This embodiment measures the category diversity of each traffic participant through type entropy, the consistency of their movement direction through directional dispersion, and the interaction complexity of their movement through velocity standard deviation, thus ensuring the credibility of the dynamic scene disorder. Furthermore, by weighting and fusing type entropy, directional dispersion, and velocity standard deviation according to different needs, the adaptability of the dynamic scene disorder to complex scenarios is improved.

[0012] Furthermore, the intent indicator parameters include intent determination indicator values ​​and intent threat indicator values; The step of determining the behavioral intent and intent index parameters of each traffic participant based on their position, speed, acceleration, and historical trajectory in the dynamic scene information includes: By using a pre-set behavior prediction model, the location, speed, acceleration, and historical trajectory of each traffic participant are predicted and analyzed to obtain a set of candidate behavioral intentions for the traffic participants. The candidate behavioral intentions are filtered based on the candidate behavioral probabilities in the candidate behavioral intention set to determine the behavioral intentions of the traffic participants and the intention determination index value. The intent threat index value of the traffic participant is determined based on the behavioral intent trajectory corresponding to the traffic participant's behavioral intent and the initial planned trajectory of the vehicle.

[0013] This embodiment improves the accuracy of intent prediction by combining multi-dimensional data such as location, speed, acceleration, and historical trajectory. Probability screening and determination of indicator values ​​quantify the credibility of behavioral intent prediction, while threat indicator values ​​help to adopt different control strategies based on different risk levels, forming a complete closed loop from perception to prediction to assessment.

[0014] Further, the step of determining a semantic risk score based on the semantic risk feature data, and determining a comprehensive risk score based on the physical risk score and the semantic risk score, includes: Based on the static scene labels, feature transformation rules are determined, and based on the feature transformation rules, the dynamic scene disorder degree and the behavioral intentions and intention index parameters of all traffic participants are transformed to obtain a semantic risk score. The physical risk score and the semantic risk score are fused according to a preset fusion coefficient to obtain a comprehensive risk score; wherein, the physical risk score is determined based on all the relative state parameters between the vehicle and each traffic participant obtained from the environmental data and the map data.

[0015] This embodiment matches appropriate feature transformation rules based on static scene labels, enabling the quantification of scene-level and intent-level semantics into semantic risk scores using these rules, thus improving the rationality and accuracy of the semantic risk scores. Furthermore, it deeply integrates semantic risk scores with physical collision risks to generate a more comprehensive dynamic risk score, avoiding the biases inherent in risk assessments that rely solely on distance and speed.

[0016] Furthermore, the preset risk threshold includes a first risk threshold and a second risk threshold, wherein the first risk threshold is less than the second risk threshold; the risk level includes low risk level, medium risk level and high risk level; The step of determining the risk level based on the comprehensive risk score and the preset risk threshold includes: When the comprehensive risk score has not yet reached the first risk threshold, the risk level is determined to be low risk level; When the comprehensive risk score reaches the first risk threshold but has not yet reached the second risk threshold, the risk level is determined to be medium risk. When the comprehensive risk score reaches the second risk threshold, the risk level is determined to be high risk.

[0017] This embodiment classifies risk levels by comparing different scoring results, so as to adopt different planning strategies for different risk levels, which helps to achieve precise response of vehicle control.

[0018] Furthermore, the control of the vehicle to perform driving operations corresponding to the risk level includes: Obtain the standard values ​​of all preset control indicators, and correct the standard values ​​of each preset control indicator according to the risk level to obtain the corrected values ​​of each preset control indicator; the preset control indicators include at least one of maximum acceleration, maximum deceleration, minimum following distance, maximum steering angular velocity, lane keeping weight, and lane change decision benefit value. Based on the initial planned trajectory of the vehicle and the index correction values ​​of all the preset control indicators, the target decision function and target constraints are determined. A target planning trajectory is generated based on the target decision function and the target constraints, and the vehicle is controlled to perform driving operations corresponding to the target planning trajectory.

[0019] This embodiment differentiates preset control indicators across different dimensions, including longitudinal, lateral, and decision-making, based on risk levels. The multi-dimensional control capability boundary parameter set, composed of the modified values ​​of these preset control indicators, redefines the "feasible operating domain" of vehicle control for decision optimization, achieving refined vehicle control. Simultaneously, under different risk levels, the vehicle exhibits control capabilities consistent with human driving experience, enhancing the interpretability of vehicle control and improving ride comfort.

[0020] An electronic device includes a processor and a memory, wherein Memory, used to store computer programs; The processor is used to execute the program stored in the memory to implement the above vehicle control method.

[0021] A vehicle comprising the aforementioned electronic equipment. Attached Figure Description

[0022] Figure 1 This is a schematic flowchart of a vehicle control method provided in an embodiment of this application; Figure 2 This is a flowchart illustrating step S101 of a vehicle control method provided in an embodiment of this application; Figure 3 This is a schematic diagram of the data processing flow of a vehicle control method provided in an embodiment of this application; Figure 4 This is a flowchart illustrating step S203 of a vehicle control method provided in an embodiment of this application; Figure 5 This is a flowchart illustrating step S401 of a vehicle control method provided in an embodiment of this application; Figure 6 This is a flowchart illustrating step S402 of a vehicle control method provided in an embodiment of this application; Figure 7 This is a flowchart illustrating step S102 of a vehicle control method provided in an embodiment of this application; Figure 8 This is a schematic diagram of the comprehensive scoring process in a vehicle control method provided in an embodiment of this application; Figure 9 This is a flowchart illustrating step S103 of a vehicle control method provided in an embodiment of this application; Figure 10 This is a schematic diagram of the trajectory optimization process in a vehicle control method provided in an embodiment of this application; Figure 11 This is another schematic flowchart of step S103 in the vehicle control method provided in an embodiment of this application; Figure 12 This is a structural diagram of the electronic device provided in the embodiments of this application. Detailed Implementation

[0023] To make the technical problems, technical solutions, and beneficial effects solved by this application clearer, the following detailed description is provided in conjunction with embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0024] Terminology Explanation In this application, the vehicle control method is applied to the vehicle's Electronic Control Unit (ECU). The ECU communicates with both a data acquisition component and a vehicle control component. The data acquisition component is a terminal used to acquire environmental and map data in real time, including but not limited to radar sensors (such as lidar and millimeter-wave radar), image sensors (such as cameras), navigation map terminals, position sensors, and speed sensors. The vehicle control component refers to components used to change the vehicle's driving state by executing specified actions, including but not limited to the accelerator, brakes, and steering wheel.

[0025] Specifically, in autonomous driving scenarios, the electronic control unit acquires environmental and map data through data acquisition components; based on the environmental and map data, it determines semantic risk feature data and physical risk scores; the semantic risk feature data includes static scene labels, dynamic scene disorder, and behavioral intentions and intention index parameters of at least one traffic participant; based on the semantic risk feature data, it determines the semantic risk score, and based on the physical and semantic risk scores, it determines the comprehensive risk score; based on the comprehensive risk score and a preset risk threshold, it determines the risk level and controls the vehicle to perform driving operations corresponding to the risk level.

[0026] Environmental data comprises detailed information characterizing the surrounding road conditions and traffic participants during the vehicle's current driving process. Map data refers to map records of the surrounding road conditions during the vehicle's current driving process. Traffic participants are objects appearing within a specified area of ​​the vehicle's current driving path, including but not limited to vehicles, pedestrians, non-motorized vehicles, and obstacles. A specific driving environment may have only one traffic participant or multiple traffic participants simultaneously. Semantic risk feature data is a set of data that transforms the driving environment into a semantic-level risk quantification description. Semantic risk feature data can be represented using structured vectors, denoted by F... semantic Specifically, F semantic =[Static Scene Label, Dynamic Scene Chaos Degree, Behavioral Intent and Intent Indicator Parameters of Traffic Participant 1, ..., Behavioral Intent and Intent Indicator Parameters of Traffic Participant i], where i represents the sequence number of the traffic participant, and N represents the total number of traffic participants. The static scene label is a classification identifier for fixed road features in the driving environment. The dynamic scene chaos degree is a quantitative value representing the degree of disorder in the movement of traffic participants in the driving environment. The behavioral intent of a traffic participant refers to the most probable expected action that the participant intends to take in the next moment. The intent indicator parameters are a quantitative description of the probability and consequences of a traffic participant's intention to perform a specified expected action in the next moment. Specifically, the intent indicator parameters include the intent certainty index value and the intent threat index value. The intent certainty index value is a quantitative description of the confidence level of a traffic participant in taking a specified expected action in the next moment, while the intent threat index value is a quantitative description of the risk posed to the current vehicle after a traffic participant takes a specified expected action in the next moment.

[0027] The physical risk score is a quantitative description of the risk of collision based on the relative physical state between the vehicle and other traffic participants. The semantic risk score is a quantitative description of the risk of collision if the vehicle and other traffic participants move along their expected trajectories in the next moment. The comprehensive risk score is a quantitative value representing the combined impact of road factors and traffic participant factors on safe vehicle operation. The preset fusion coefficient is a coefficient used to characterize the importance of the semantic risk score in generating the comprehensive risk score.

[0028] The preset risk threshold is a pre-defined critical value used to determine whether a vehicle's overall risk score reaches different risk levels. It can be set to a default value or adjusted as needed. Risk level is used to distinguish different degrees of risk severity. Risk levels can be divided into two or more levels as needed. Specifically, risk levels are divided into three levels: low risk, medium risk, and high risk. The preset risk threshold includes one or more thresholds used to distinguish different risk levels. When the risk level is divided into three levels, the preset risk threshold includes a first risk threshold and a second risk threshold, with the first risk threshold being lower than the second risk threshold. The first risk threshold is used to determine whether the vehicle's overall risk score reaches the ideal critical value for the medium risk level, and the second risk threshold is used to determine whether the vehicle's overall risk score reaches the ideal critical value for the high risk level. During autonomous driving, different risk levels correspond to different control capability boundaries, different control capability boundaries correspond to different planning trajectory optimization constraints, and thus correspond to different driving operations. Driving operation is a vehicle control method used to perform specific actions through specific components of the vehicle to change the vehicle's driving state. For example, when the vehicle is traveling too fast and is prone to skidding, lane change decisions in the planned trajectory are suppressed, while when the vehicle is traveling at a speed that is not prone to skidding, lane change decisions in the planned trajectory are allowed to be executed.

[0029] The electronic control unit (ECU) analyzes environmental and map data to obtain static and dynamic scene information. Static scene information refers to scene elements that do not change over time, including but not limited to road types (such as highways, ramps, and construction zones), traffic signs (such as speed limits and lane change restrictions), and lane topology types (such as merging and diverging). Dynamic scene information refers to scene elements that change continuously over time, including but not limited to data on the location, speed, acceleration, and direction of movement of traffic participants. Participant type is used to distinguish the behavioral characteristics of different traffic participants, such as vehicles, pedestrians, and non-motorized vehicles. Factors for assessing dynamic scene disorder include type entropy, directional dispersion, and speed standard deviation. Type entropy is a quantitative value used to assess the distribution of traffic participant types; directional dispersion is a quantitative value used to assess the dispersion of traffic participant movement directions; and speed standard deviation is a quantitative value used to assess the differences in speed among traffic participants. The preset disorder weights are fusion weight coefficients used to characterize the degree of influence of each evaluation factor on the disorder of a dynamic scene. They include the type weight (denoted by α) corresponding to type entropy, the direction weight (denoted by β) corresponding to direction dispersion, and the velocity weight (denoted by γ) corresponding to velocity standard deviation. The preset disorder weights can be set to a default value or adjusted as needed, and the sum of the type weight, direction weight, and velocity weight is 1.

[0030] For each traffic participant, a candidate behavioral intention set can be obtained by predicting and analyzing the participant's position, speed, acceleration, and historical trajectory using a pre-set behavior prediction model. The pre-trained behavior prediction model is a neural network model that predicts the probability of different actions occurring in the next moment based on the traffic participant's current motion state (such as position, speed, and acceleration) and historical trajectory. The pre-set behavior prediction model can be based on a Long Short-Term Memory (LSTM) network, a Convolutional Neural Network (CNN), or a transformer-based model. The behavioral intention trajectory is the expected action trajectory formed by the prediction of the traffic participant based on the pre-set behavior prediction model. The candidate behavioral intention set includes at least one candidate behavioral intention and its corresponding candidate behavioral probability. A candidate behavioral intention refers to the expected action that the traffic participant may take in the next moment, as output by the pre-set behavior prediction model, and the candidate behavioral probability refers to the probability of the expected action occurring. The traffic participant's behavioral intention is the expected action with the highest probability of occurring in the next moment, selected from the candidate behavioral intentions.

[0031] Different static scene labels correspond to different feature transformation rules. Feature transformation rules refer to the rules for converting the dynamic scene disorder and the behavioral intentions and intention index parameters of all traffic participants in the semantic risk feature data into semantic risk scores. Relative state parameters are parameters used to characterize the relative motion state between the current vehicle and traffic participants, including but not limited to relative distance, relative speed, and relative acceleration.

[0032] The initial planned trajectory of a vehicle refers to the expected driving trajectory, which is generated by a planning algorithm based on the vehicle's current position, speed, direction, and other state parameters, combined with high-precision maps and environmental perception information, and awaits optimization. Preset control indicators are control constraint parameters set in advance to limit the vehicle's actions in longitudinal, lateral, following, and lane-changing maneuvers. Preset control indicators include at least one of the following: maximum acceleration, maximum deceleration, minimum following distance, maximum steering angular velocity, lane-keeping weight, and lane-changing decision benefit value. Each preset control indicator corresponds to a standard indicator value, which is the indicator value when there is no risk in the vehicle's longitudinal, lateral, following, and lane-changing actions. The indicator correction value refers to the indicator value adjusted after adjusting the standard indicator value under different risk levels.

[0033] A target decision function is constructed based on the expected action type of the initial planned trajectory. This target decision function is a mathematical evaluation function used to assess whether the initial planned trajectory is the optimal trajectory. Based on the indicator correction values ​​of all preset control indicators, target constraints are determined. These target constraints are the restrictions used to evaluate whether the initial planned trajectory is the optimal trajectory. The target planned trajectory refers to the planned trajectory that the vehicle actually needs to execute in the next moment.

[0034] This application provides a vehicle control method, which is illustrated using an example of a vehicle's electronic control unit. Please refer to [link / reference]. Figure 1 This includes the following steps: S101: Based on environmental data and map data, determine semantic risk feature data and physical risk scores; the semantic risk feature data includes static scene labels, dynamic scene confusion degree, and behavioral intentions and intention index parameters of at least one traffic participant; S102: Determine the semantic risk score based on the semantic risk feature data, and determine the comprehensive risk score based on the physical risk score and the semantic risk score; S103: Based on the comprehensive risk score and the preset risk threshold, determine the risk level and control the vehicle to perform driving operations corresponding to the risk level.

[0035] Understandably, the vehicle's electronic control unit (ECU) acquires environmental and map data. On one hand, based on the environmental and map data, the ECU performs multi-dimensional semantic analysis at the static scene level, dynamic scene level, and intent level to obtain semantic risk feature data, which is then converted into a semantic risk score. On the other hand, the ECU analyzes the environmental and map data to obtain a physical risk score. Next, the ECU merges the physical and semantic risk scores to obtain a comprehensive risk score. Finally, the ECU compares the comprehensive risk score with a preset risk threshold to determine the risk level and controls the vehicle to perform driving operations corresponding to that risk level.

[0036] This embodiment performs multi-dimensional semantic understanding of environmental and map data at the static, dynamic, and intent levels to generate structured semantic risk features. This ensures the comprehensiveness of environmental semantics between vehicles and traffic participants, contributing to improved precision and rationality in subsequent trajectory planning. Furthermore, this embodiment analyzes the physical state risk of vehicles affected by traffic participants based on environmental and map data. By leveraging semantic risk features, it can identify potential risks caused by malicious intent earlier. The comprehensive risk score, which integrates semantic risk features and physical state risk, is used to classify different risk levels and adjust vehicle trajectory planning accordingly. This improves the interpretability and predictability of vehicle control decisions in complex dynamic environments, ultimately enhancing the driving comfort experience.

[0037] In one embodiment, please refer to Figure 2 In step S101, namely determining semantic risk feature data based on environmental data and map data, the following steps are included: S201: Determine static scene information and dynamic scene information based on environmental data and map data; S202: Determine the static scene label based on the road type, traffic signs, and lane topology type in the static scene information; S203: Based on the dynamic scene information, determine the dynamic scene disorder level and at least one of the traffic participants' behavioral intentions and intention index parameters; S204: Determine semantic risk feature data based on the static scene labels, the dynamic scene disorder, and the behavioral intentions and intention index parameters of all traffic participants.

[0038] Understandably, please refer to Figure 3 The electronic control unit (ECU) outputs structured semantic risk feature data through multi-dimensional semantic parsing at the static scene level, dynamic scene level, and intent level. On one hand, based on Simultaneous Localization and Mapping (SLAM) algorithms, image semantic segmentation models, and map matching algorithms, the ECU performs static scene-level semantic parsing on static scene information to obtain static scene labels. Specifically, road types include highways, urban roads, ramps, tunnels, and construction zones, i.e., "road_type: {highway, urbanroad, ramp, tunnel, construction zone}". Traffic rule indicators include speed limit 60 km / h, no lane change, yield, and priority road, i.e., "traffic rule: {speed limit: 60, no lane change, yield, priorityroad}". Lane topology types include merging, diverging, intersections, and roundabouts, i.e., "lane topology: {merging,diverging, intersection, roundabout}". When the static scene label is scene_label: "urbanroad-unsignalized-intersection", it indicates that the static scene is an unsignalized intersection on an urban road. On the other hand, the electronic control unit performs dynamic scene-level semantic parsing of dynamic scene information to obtain the dynamic scene disorder level. For example, it counts the number of traffic participants, their type diversity, and the dispersion of movement directions to quantify the complexity and unpredictability of the scene, obtaining a dynamic scene disorder level index, denoted by C.s Furthermore, the electronic control unit performs intent-level semantic parsing of dynamic scene information to obtain the behavioral intentions and intent index parameters of all traffic participants. For example, it predicts the behavior of traffic participants (vehicles, pedestrians, and non-motorized vehicles) and outputs behavioral intentions (such as "go straight", "cut left", "cut right", "slow down and yield", "speed up and overtake"), as well as the corresponding intent index parameters (such as confidence level).

[0039] This embodiment ensures the environmental semantic comprehensiveness of semantic risk features by performing multi-dimensional semantic understanding of environmental data and map data at the static scene level, dynamic scene level, and intent level, thereby improving the accuracy and interpretability of subsequent risk assessment.

[0040] In one embodiment, please refer to Figure 4 In step S203, namely, determining the dynamic scene chaos level and the behavioral intentions and intention index parameters of at least one traffic participant based on the dynamic scene information, the steps include: S401: Determine the degree of chaos in the dynamic scene based on the participant type, participant movement direction, and participant speed of all traffic participants in the dynamic scene information; S402: Based on the participant's position, speed, acceleration, and historical trajectory of each traffic participant in the dynamic scene information, determine the behavioral intention and intention index parameters of each traffic participant.

[0041] Understandably, the electronic control unit (ECU) quantifies the complexity and unpredictability of the scenario for all traffic participants from three dimensions: participant type, participant direction of movement, and participant speed, obtaining dynamic scenario disorder, such as statistical type diversity and movement direction dispersion. Based on four dimensions—participant position, participant speed, participant acceleration, and participant historical trajectory—the ECU predicts the behavior of each traffic participant, determining their behavioral intentions and intention index parameters.

[0042] This embodiment performs targeted analysis of data from different dimensions in dynamic scene information at both the dynamic scene level and the intent level, ensuring the accuracy of dynamic scene disorder and improving the predictability of traffic participants' behavioral intentions and intent indicator parameters.

[0043] In one embodiment, please refer to Figure 5 In step S401, which involves determining the dynamic scene disorder level based on the participant type, movement direction, and speed of all traffic participants in the dynamic scene information, the following steps are included: S501: Determine the type entropy based on the participant type of all the traffic participants; S502: Determine the directional dispersion based on the movement directions of all traffic participants; S503: Determine the speed standard deviation based on the participant speeds of all the traffic participants; S504: The type entropy, the directional dispersion, and the velocity standard deviation are weighted according to a preset disorder weight to obtain the dynamic scene disorder.

[0044] Understandably, in one specific embodiment, the total number of traffic participants within the current vehicle's perception range is N, and the traffic participants are numbered i. They are categorized into three types: vehicles, pedestrians, and non-motorized vehicles. Each participant type corresponds to a certain number of traffic participants. The type entropy is represented by H. type The specific calculation formula is H type =-Σ[P(type j ) ×log2(P(type j ))], where P(type j ) = Number of traffic participants of type j / N, j = 1, 2, 3. The directional dispersion is represented by σ. θ Specifically, it refers to the standard deviation of the movement directions of all traffic participants (the larger the standard deviation, the higher the dispersion). Calculating the standard deviation requires first calculating the average direction angle θ of the movement directions of all traffic participants. mean The standard deviation of velocity is expressed as σ. v The larger the speed standard deviation, the more complex the interaction. The disorder level of dynamic scenes is represented by C. s The specific calculation formula is C. s =α ×normalize(H type ) + β × normalize(σ θ ) + γ × normalize(σ v In this equation, the type weight corresponding to the type entropy is represented by α, the direction weight corresponding to the direction dispersion is represented by β, and the velocity weight corresponding to the velocity standard deviation is represented by γ (e.g., α=0.3, β=0.4, γ=0.3). The normalize() function normalizes each component to the [0,1] interval, and finally C... s ∈[0,1].

[0045] This embodiment measures the category diversity of each traffic participant through type entropy, the consistency of their movement direction through directional dispersion, and the interaction complexity of their movement through velocity standard deviation, thus ensuring the credibility of the dynamic scene disorder. Furthermore, by weighting and fusing type entropy, directional dispersion, and velocity standard deviation according to different needs, the adaptability of the dynamic scene disorder to complex scenarios is improved.

[0046] In one embodiment, the intent indicator parameters include an intent determination indicator value and an intent threat indicator value. Please refer to [reference needed]. Figure 6 In step S402, namely, determining the behavioral intention and intention index parameters of each traffic participant based on the participant's position, speed, acceleration, and historical trajectory in the dynamic scene information, includes the following steps: S601: By using a preset behavior prediction model, predict and analyze the participant's position, speed, acceleration, and historical trajectory for each traffic participant to obtain a set of candidate behavioral intentions for the traffic participants; S602: Filter candidate behavioral intentions based on the candidate behavioral probabilities in the candidate behavioral intention set to determine the behavioral intention of the traffic participant and the intention determination index value; S603: Determine the intent threat index value of the traffic participant based on the behavioral intent trajectory corresponding to the traffic participant's behavioral intent and the initial planned trajectory of the vehicle.

[0047] Understandably, the intent indicator parameters include an intent determination indicator value and an intent threat indicator value, with the intent determination indicator value denoted as I. c The intention threat index is measured by the probability distribution entropy of predicted actions in the behavioral intention trajectory, and the intention threat index value is expressed as I. t The system predicts the potential conflict severity of the action to the current vehicle by analyzing the behavioral intent trajectory, combined with the behavioral intent trajectory of the traffic participant and the initial planned trajectory of the current vehicle. The electronic control unit (ECU) compares the behavioral intent trajectory corresponding to the traffic participant's behavioral intent with the initial planned trajectory of the current vehicle to determine the conflict type, remaining conflict duration, and spatial overlap. Based on these factors, it determines the traffic participant's intent threat index value. Conflict type refers to the classification used to distinguish different directions of collision between the traffic participant and the current vehicle under a specified behavioral intent, including head-on collision, side collision, and rear-end collision. Remaining conflict duration refers to the time from the current time to the expected collision time, also known as time-to-collision (TTC). Spatial overlap refers to the degree of spatial overlap between the behavioral intent trajectory and the initial planned trajectory. Each traffic participant corresponds to one behavioral intent, one intent determination index value, and one intent threat index value.

[0048] In one specific embodiment, firstly, the electronic control unit (ECU) uses a preset behavior prediction model to predict and analyze the position, speed, acceleration, and historical trajectory of each traffic participant, generating a set of possible future behavior hypotheses and their probabilities for each participant, i.e., a candidate behavior intention set. For example, if the traffic participant is a vehicle in the adjacent lane, the candidate behavior intention set is Hypothesis = {keep_lane: 70%, cut_in_left: 25%, cut_out_right: 5%}. In this case, the probability of the vehicle in the adjacent lane maintaining a straight path is 70%, the probability of changing lanes to the left is 25%, and the probability of changing lanes to the right is 5%. Next, the ECU calculates the probability distribution entropy of the probabilities in the candidate behavior intention set of the traffic participants. The formula for calculating the probability distribution entropy is Entropy = -Σ(P(m) × log2(P(m))), where m represents the type of candidate behavior intention in the candidate behavior intention set, and P(m) represents the probability of the candidate behavior. The formula for calculating the intention determination index value is I. c =1-normalize(Entropy), where normalize represents normalization to [0,1]. For a candidate behavioral intent, if I c A value closer to 1 indicates a stronger likelihood of the candidate's intended action (e.g., a 99% probability of maintaining lane direction). If I... c If the value is close to 0, the intent to commit the candidate behavior is uncertain (e.g., all candidate behavior intentions have an equal probability of occurrence). The electronic control unit (ECU) filters out the behavior intention with the highest probability from the candidate behavior intentions and obtains the corresponding intent determination index value. The formula for calculating the intent threat index value is I. t =f(conflict type, remaining conflict duration, spatial overlap), where the conflict type is weighted by W. critical (e.g., opposing collision = 1.0) > W major (e.g., lateral scrape = 0.7) > W minor (e.g., rear-end collision = 0.5). The smaller the remaining time to conflict (TTC), the greater the threat level. The simplified formula for calculating the intention threat index value is I. t =W critical × exp(-k × TTC), where k represents the attenuation coefficient. Structured semantic risk feature data is represented by F. semantic For example, F semantic = [scene_label: "unsignalized_intersection", C s : 0.84, target_1: {id: 101, type: vehicle, intent: "cut_in_left", I c0.40, I t : 0.65},target_2: {id: 205, type: pedestrian, intent: "crossing", I c 0.90, I t : 0.80},target_3: {id: 308, type: vehicle, intent: "go_straight", I c 0.70, I t :0.29},...]. Where, target represents the traffic participant.

[0049] This embodiment improves the accuracy of intent prediction by combining multi-dimensional data such as location, speed, acceleration, and historical trajectory. Probability screening and determination of indicator values ​​quantify the credibility of behavioral intent prediction, while threat indicator values ​​help to adopt different control strategies based on different risk levels, forming a complete closed loop from perception to prediction to assessment.

[0050] In one embodiment, the handling anomaly index value includes a steering anomaly index value and a braking anomaly index value. Please refer to... Figure 7 In step S102, namely, determining the semantic risk score based on the semantic risk feature data, and determining the comprehensive risk score based on the physical risk score and the semantic risk score, the following steps are included: S701: Determine feature transformation rules based on the static scene labels, and transform the dynamic scene disorder degree and the behavioral intentions and intention index parameters of all traffic participants according to the feature transformation rules to obtain a semantic risk score; S702: The physical risk score and the semantic risk score are fused according to a preset fusion coefficient to obtain a comprehensive risk score; wherein, the physical risk score is determined based on all the relative state parameters between the vehicle and each of the traffic participants after obtaining the relative state parameters between the vehicle and each of the traffic participants based on the environmental data and the map data.

[0051] Understandably, the electronic control unit (ECU) looks up the feature transformation rule corresponding to the actual static scene label from the label rule mapping table. The label rule mapping table is a pre-generated data table used to record the mapping relationships between different static scene labels and functions of different feature transformation rules. The ECU substitutes the dynamic scene chaos level and the behavioral intentions and intention index parameters of all traffic participants into the function of the feature transformation rule for calculation, obtaining a semantic risk score. Then, the ECU fuses the physical risk score and the semantic risk score according to a preset fusion coefficient to obtain a comprehensive risk score.

[0052] In one specific embodiment, please refer to Figure 8 The semantic risk feature data is represented as F semantic The physical risk score is represented as R. physics The semantic risk score is represented as R. semantic The comprehensive risk score is represented as R(t). The electronic control unit (ECU) obtains the relative state parameters (such as relative distance D, relative speed V, and relative acceleration A) between the current vehicle and each traffic participant based on environmental and map data, and calculates the physical risk score based on all relative state parameters. The ECU can use an improved collision time algorithm (considering relative acceleration) to calculate the physical risk score, or it can use a P-Metric model to calculate the physical risk score. The rule function g(·) represents the feature transformation rule determined based on the static scene label, and the general formula for calculating the semantic risk score is R. semantic = g(C s , I t , 1-I c The specific calculation formula is R. semantic = x × C s + y×Σ(I t × (1 - I c ), where x and y are weighting coefficients, Σ(I t × (1 - I c This represents the sum of the converted intent index parameters of all traffic participants. Higher intent threat levels, lower intent certainty, and higher dynamic scene chaos all lead to higher semantic risk scores. A comprehensive risk score is generated by fusing physical and semantic risks; the specific fusion calculation formula is R(t) = R physics × (1 +z × R semantic ), where z is a preset fusion coefficient used to adjust the weight of semantic risk scoring. The comprehensive risk score R(t) not only reflects the immediate possibility of physical collision, but also anticipates and amplifies the potential risks brought about by malicious intent and chaotic scenarios.

[0053] This embodiment matches appropriate feature transformation rules based on static scene labels, enabling the quantification of scene-level and intent-level semantics into semantic risk scores using these rules, thus improving the rationality and accuracy of the semantic risk scores. Furthermore, it deeply integrates semantic risk scores with physical collision risks to generate a more comprehensive dynamic risk score, avoiding the biases inherent in risk assessments that rely solely on distance and speed.

[0054] In one embodiment, the preset risk threshold includes a first risk threshold and a second risk threshold, wherein the first risk threshold is less than the second risk threshold; the risk level includes low risk level, medium risk level, and high risk level. Please refer to... Figure 9 In step S103, namely, determining the risk level based on the comprehensive risk score and the preset risk threshold, the following steps are included: S901: When the comprehensive risk score has not yet reached the first risk threshold, the risk level is determined to be a low risk level; S902: When the comprehensive risk score reaches the first risk threshold but has not yet reached the second risk threshold, the risk level is determined to be a medium risk level. S903: When the comprehensive risk score reaches the second risk threshold, the risk level is determined to be a high-risk level.

[0055] Understandably, please refer to Figure 10 In one specific embodiment, the preset risk threshold includes a first risk threshold and a second risk threshold, wherein the first risk threshold is denoted as R. low And R low =0.2, the second risk threshold is expressed as R high And R high =0.6. The electronic control unit compares the comprehensive risk score with the preset risk threshold and divides it into three intervals, corresponding to three risk levels. Wherein, when R(t)∈[0, R... low When R(t) ∈ (R) it is in the low-risk range, and the risk level is low-risk. low ,R high When R(t)∈(R_high, 1], it is in the medium-risk range and the risk level is medium-risk. When R(t)∈(R_high, 1], it is in the high-risk range and the risk level is high-risk.

[0056] This embodiment classifies risk levels by comparing different scoring results, so as to adopt different planning strategies for different risk levels, which helps to achieve precise response of vehicle control.

[0057] In one embodiment, please refer to Figure 11 In step S103, the controlled vehicle performs a driving operation corresponding to the risk level, including the following steps: S1101: Obtain the standard values ​​of all preset control indicators, and correct the standard values ​​of each preset control indicator according to the risk level to obtain the corrected values ​​of each preset control indicator; the preset control indicators include at least one of maximum acceleration, maximum deceleration, minimum following distance, maximum steering angular velocity, lane keeping weight, and lane change decision benefit value. S1102: Determine the target decision function and target constraints based on the initial planned trajectory of the vehicle and the index correction values ​​of all the preset control indicators; S1103: Generate a target planning trajectory based on the target decision function and the target constraints, and control the vehicle to perform driving operations corresponding to the target planning trajectory.

[0058] Understandably, please refer to Figure 10 The electronic control unit selects a nonlinear mapping function corresponding to each preset control indicator from a pre-defined cluster of nonlinear mapping functions based on the risk level. Then, it corrects the standard value of each preset control indicator according to the selected nonlinear mapping function, obtaining the corrected value for each preset control indicator. The set of corrected values ​​for all preset control indicators generates a multi-dimensional control capability boundary parameter set, denoted by P. bound(t) The preset control indicators include at least one of the following: maximum acceleration, maximum deceleration, minimum following distance, maximum steering angle, lane keeping weight, and lane change decision benefit value. Maximum acceleration, maximum deceleration, and minimum following distance are control boundary parameters in the longitudinal dimension; maximum steering angle and lane keeping weight are control boundary parameters in the lateral dimension; and the lane change decision benefit value is the control boundary parameter in the decision dimension. The electronic control unit (ECU) does not directly control the vehicle to execute the initial planned trajectory. Instead, it determines the target decision function based on the expected action at the next moment in the current vehicle's initial planned trajectory (e.g., if the expected action at the next moment in the initial planned trajectory is a lane change, then a target decision function with a safe and smooth lane change as the objective is constructed). It then determines the target constraints based on the indicator correction values ​​of all preset control indicators (e.g., target constraints with maximum acceleration, maximum deceleration, and maximum steering angle as constraints). Finally, the ECU generates the target planned trajectory based on the target decision function and target constraints, that is, it performs optimization decision-making and execution within the corrected control capability boundaries.

[0059] The pre-defined nonlinear mapping function set is a dataset used to record the nonlinear mapping functions corresponding to different pre-defined control indicators at different risk levels. Each combination of a pre-defined control indicator and a risk level corresponds to a nonlinear mapping function. Pre-defined control indicators may correspond to different nonlinear mapping functions at different risk levels, allowing for varying degrees of correction to the standard values ​​of the indicators. For example, for the same pre-defined control indicator, when the risk level is low, the vehicle can utilize its full performance; to pursue efficiency and comfort, the nonlinear mapping function changes smoothly, or even does not shrink its capabilities. When the risk level is medium, environmental complexity, uncertainty, or threat increases, and the nonlinear mapping function needs to shrink smoothly and gradually to achieve a balance between safety and efficiency. When the risk level is high, the physical collision risk is extremely high or the intent is extremely dangerous and uncertain; the vehicle enters a minimum risk strategy, and the nonlinear mapping function needs to shrink strictly, with collision avoidance as the highest priority.

[0060] A pre-defined family of nonlinear mapping functions designs independent but coordinated nonlinear mapping functions for each pre-defined control index. The general form of the control capability boundary parameters is P. bound(R) = f(P nominal , R, strategy), where P bound(R) To define the boundary values ​​of the preset control index R, P nominal R is the standard value of the preset control indicator when there is zero risk, and strategy is the control strategy corresponding to the preset control indicator R.

[0061] In one specific embodiment, the maximum acceleration is represented by a_max_acc(R), with a corresponding standard value of a_nom_acc = 3.0 m / s². The maximum deceleration is represented by a_max_dec(R), with a corresponding standard value of a_nom_dec = -4.0 m / s². The minimum following distance is represented by TTC_min(R), with a corresponding standard value of TTC_nom = 1.5 s. The maximum steering angular velocity is represented by ω_max(R), with a corresponding standard value of ω_nom = 30 deg / s. The lane keeping weight is represented by W_lc(R), with a corresponding standard value of W_nom = 1.0. The lane change decision benefit is represented by Threshold_lc(R), with a corresponding standard value of Threshold_nom_lc = 0.6. Furthermore, preset risk thresholds include a first risk threshold (R_low = 0.2) and a second risk threshold (R_high = 0.6).

[0062] The corrected value of the maximum acceleration indicator remains consistent with the standard indicator value in the low-risk interval, and a quadratic attenuation function is adopted as the nonlinear mapping function in the medium and high-risk intervals to suppress aggressive behavior at an early stage and limit rapid acceleration. The nonlinear mapping functions corresponding to the maximum acceleration at different risk levels are as follows: At the low risk level, if R<= R_low: a_max_acc(R) = a_nom_acc. At the medium risk level, elif R_low<R<= R_high: a_max_acc(R) = a_nom_acc × [1 -((R - R_low) / (R_high - R_low))^2]. At the high risk level, else: / / R>R_high: a_max_acc(R)= a_nom_acc × 0.1. For example, when the comprehensive risk score R=0.4, the risk level is the medium risk level, and the corrected indicator value of the maximum acceleration a_max_acc = 3.0 * [1 - ((0.4-0.2) / 0.4)^2] = 3.0 * [1 - 0.25]=2.25 m / s². The acceleration capability is limited to 75% of the standard indicator value.

[0063] The maximum deceleration can be slightly limited at low and medium risk levels to ensure comfort, and sufficient emergency braking capability must be guaranteed at the high risk level. The nonlinear mapping functions corresponding to the maximum deceleration at different risk levels are as follows: if R<= R_low: |a_max_dec(R)| = |a_nom_dec| × 0.8 / / At the low risk level, the maximum deceleration is limited to improve comfort. elif R_low<R<= R_high: |a_max_dec(R)| = |a_nom_dec| × (0.8 + 0.2×(R-R_low) / (R_high-R_low)) / / At the medium risk level, the value linearly returns to the standard indicator value. else: |a_max_dec(R)|= |a_nom_dec| × 1.2 / / At the high risk level, a higher upper limit of deceleration is allowed (e.g., triggering AEB). For example, when the comprehensive risk score R=0.4, the risk level is the medium risk level, and the corrected indicator value of the maximum deceleration |a_max_dec| =4.0 × (0.8 + 0.2 × 0.5) = 4.0 × 0.9 = 3.6 m / s².

[0064] As the risk increases, a larger safety distance needs to be maintained. Therefore, an exponential growth function is used as the nonlinear mapping function to ensure sufficient distance buffer at high risk levels. The nonlinear mapping function for the minimum following distance is the same for all risk levels: TTC_min(R) = TTC_nom × exp(μ × R), where μ is the growth coefficient, such as μ=0.8. For example, when the comprehensive risk score R=0.4, the risk level is medium risk, and the index correction value for the maximum deceleration TTC_min = 1.5 × exp(0.8 × 0.4) ≈ 1.5 × 1.377 ≈ 2.07 s.

[0065] The lateral control aggression of the maximum steering angular velocity should be strictly suppressed; therefore, an exponential decay function is used as the nonlinear mapping function. The nonlinear mapping function for the maximum steering angular velocity is as follows for all risk levels: ω_max(R) = ω_nom × exp(-k_ω × R), where k_ω is the decay coefficient, such as k_ω=1.5. For example, when the comprehensive risk score R=0.4, the risk level is medium risk, and the index correction value of the maximum steering angular velocity ω_max =30 × exp(-1.5 × 0.4) = 30 × exp(-0.6) ≈ 30 × 0.549 ≈ 16.5 deg / s, significantly limiting the steering rate.

[0066] In the trajectory optimization cost function, lane keeping weight increases the tracking weight of the lane centerline, causing the vehicle to lean more towards the lane center. The nonlinear mapping function of lane keeping weight corresponding to different risk levels is as follows: When R <= R_high, corresponding to low and medium risk levels, W_lc(R) = W_nom + (W_max - W_nom) × (R / R_high). When R > R_high, corresponding to high risk level, W_lc(R) = W_max, / / assuming W_nom = 1.0, W_max = 5.0. For example, when the comprehensive risk score R = 0.4, the risk level is medium risk, and the index correction value of lane keeping weight W_lc = 1.0 + (5.0-1.0)×(0.4 / 0.6) ≈ 1.0 + 4.0 × 0.667 ≈ 3.67.

[0067] The threshold for lane-changing decision-making returns varies with risk; higher risk necessitates a higher net return for a lane-changing decision to be permitted. The nonlinear mapping function for the lane-changing decision-making return threshold across different risk levels is as follows: Threshold_lc(R) = Threshold_nom_lc + Δ_th × R, where Δ_th is the increment, such as 0.3. For example, when the overall risk score R = 0.4, the risk level is medium risk, and the adjusted threshold for lane-changing decision-making returns is Threshold_lc = 0.6 + 0.3 × 0.4 = 0.72. This makes the lane-changing decision more conservative.

[0068] The electronic control unit integrates a trajectory tracking controller or behavior decision module. After generating the initial planned trajectory, it needs to generate the target planned trajectory based on the objective decision function and objective constraints, and execute the control commands (accelerator / brake / steering commands) corresponding to the target planned trajectory. For example, in the Model Predictive Control (MPC) optimization problem, the initial planned trajectory is the expected lane change action. The maximum acceleration a_max_acc(R), the maximum deceleration a_max_dec(R), and the maximum steering angular velocity ω_max(R) can be used as inequality constraints, or the lane change action can be executed only when Threshold_lc(R) is satisfied. The objective decision function for lane change decision evaluation is: Net_Benefit = Benefit_lane_change - Cost_lane_change, and the objective constraint is to execute the lane change when Net_Benefit > Threshold_lc(R). Here, Benefit_lane_change represents the lane change benefit (such as saving time, avoiding congestion, and improving safety), Cost_lane_change represents the lane change cost (such as collision risk and passenger discomfort), and Net_Benefit represents the net benefit. The standard value for the lane change decision benefit is Threshold_nom_lc = 0.6. When the overall risk score R = 0.4, the risk level is medium risk, and the adjusted value for the lane change decision benefit threshold is Threshold_lc = 0.6 + 0.3 × 0.4 = 0.72. The lane change decision becomes more conservative than in the nominal state (Threshold = 0.6). That is, a lane change opportunity with a net benefit of 0.62 in the initial planned trajectory would be executed in the nominal state, but it will be suppressed under the current risk level. The current vehicle will choose to stay in its lane on the target planned trajectory instead of performing a lane change action according to the initial planned trajectory.

[0069] This embodiment, through refined constraints on different dimensions of control indicators, can retain necessary maneuverability as much as possible while ensuring safety. The multi-dimensional control capability boundary parameter set, composed of the comprehensive risk score and the indicator correction values ​​of the preset control indicators, are all monitorable intermediate variables. Engineers can easily define and calibrate the vehicle's "driving style" (such as more sporty or more comfortable) in different scenarios by adjusting the mapping function, realizing personalized customization of vehicle control strategies. For example, at a medium-risk level, lane changes and aggressive steering may be strictly limited, but appropriate acceleration while following other vehicles may still be allowed, thereby reducing the disturbance to the overall traffic flow caused by excessive conservatism (such as unwarranted sudden braking).

[0070] After controlling the vehicle to execute driving operations corresponding to the target planned trajectory, the electronic control unit acquires new environmental and map data at fixed time intervals to begin the next vehicle control process. For example, in the next control cycle (e.g., after 100ms), it recalculates the multi-dimensional control capability boundary parameter set, consisting of the comprehensive risk score and the corrected values ​​of all preset control indicators. If the intention of the adjacent vehicle to cut in disappears (intent threat index value I...), the control unit will continue to control the vehicle. t If the risk decreases, the overall risk score will drop, and the constraints in the multidimensional control capability boundary parameter set will automatically and smoothly relax, allowing the vehicle to regain more aggressive control capabilities. Conversely, if the risk increases, the constraints will tighten further. The vehicle's control capabilities are no longer fixed or abruptly changing, but rather continuously and subtly adjust their "agility" and "caution" in accordance with the dynamic changes in environmental risks.

[0071] This embodiment differentiates preset control indicators across different dimensions, including longitudinal, lateral, and decision-making, based on risk levels. The multi-dimensional control capability boundary parameter set, composed of the modified values ​​of these preset control indicators, redefines the "feasible operating domain" of vehicle control for decision optimization, achieving refined vehicle control. Simultaneously, under different risk levels, the vehicle exhibits control capabilities consistent with human driving experience, enhancing the interpretability of vehicle control and improving ride comfort.

[0072] This application embodiment also provides a vehicle control device, including: a semantic feature analysis module, used to determine semantic risk feature data and physical risk score based on environmental data and map data; the semantic risk feature data includes static scene labels, dynamic scene confusion degree, and behavioral intentions and intention index parameters of at least one traffic participant; a comprehensive scoring module, used to determine a semantic risk score based on the semantic risk feature data, and to determine a comprehensive risk score based on the physical risk score and the semantic risk score; and a control execution module, used to determine a risk level based on the comprehensive risk score and a preset risk threshold, and to control the vehicle to perform driving operations corresponding to the risk level.

[0073] This application also provides an electronic device 120, please refer to... Figure 12 It includes a memory 121 and a processor 122, wherein the memory 121 is used to store computer programs; the processor 122 is used to execute the programs stored in the memory 121 to implement the vehicle control method described in any embodiment of this application.

[0074] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the vehicle control method described in any embodiment of this application.

[0075] In this application, "multiple" refers to two or more.

[0076] The terms “first,” “second,” “third,” “fourth,” etc., used in this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0077] In this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, in this application, the character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0078] Unless otherwise specified, all steps in this application may be performed sequentially or randomly. For example, if the method includes steps A and B, it means that the method may include steps A and B performed sequentially, or it may include steps B and A performed sequentially. For example, if the method may also include step C, it means that step C may be added to the method in any order. For example, the method may include steps A, B, and C, or it may include steps A, C, and B, or it may include steps C, A, and B, etc.

[0079] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A vehicle control method, characterized in that, include: Based on environmental and map data, semantic risk features and physical risk scores are determined. The semantic risk feature data includes static scene labels, dynamic scene confusion level, and at least one traffic participant's behavioral intention and intention index parameters. Based on the semantic risk feature data, a semantic risk score is determined, and based on the physical risk score and the semantic risk score, a comprehensive risk score is determined; Based on the comprehensive risk score and the preset risk threshold, the risk level is determined, and the vehicle is controlled to perform driving operations corresponding to the risk level.

2. The vehicle control method according to claim 1, characterized in that, The determination of semantic risk feature data based on environmental data and map data includes: Based on environmental and map data, determine static and dynamic scene information; The static scene label is determined based on the road type, traffic signs, and lane topology type in the static scene information; Based on the dynamic scene information, determine the dynamic scene disorder level and at least one of the traffic participants' behavioral intentions and intention index parameters; Semantic risk feature data is determined based on the static scene labels, the dynamic scene disorder, and the behavioral intentions and intention index parameters of all traffic participants.

3. The vehicle control method according to claim 2, characterized in that, The step of determining the dynamic scene disorder level and the behavioral intentions and intention index parameters of at least one traffic participant based on the dynamic scene information includes: The degree of chaos in the dynamic scene is determined based on the participant type, movement direction, and speed of all traffic participants in the dynamic scene information. Based on the participant's location, speed, acceleration, and historical trajectory in the dynamic scene information, the behavioral intention and intention index parameters of each participant are determined.

4. The vehicle control method according to claim 3, characterized in that, The determination of dynamic scene disorder based on the participant type, movement direction, and speed of all traffic participants in the dynamic scene information includes: Determine the type entropy based on the participant type of all the traffic participants; Determine the directional dispersion based on the direction of movement of all traffic participants; Determine the speed standard deviation based on the participant speeds of all traffic participants; The type entropy, the directional dispersion, and the velocity standard deviation are weighted according to a preset disorder weight to obtain the dynamic scene disorder.

5. The vehicle control method according to claim 3, characterized in that, The intent indicator parameters include intent determination indicator values ​​and intent threat indicator values; The step of determining the behavioral intent and intent index parameters of each traffic participant based on their position, speed, acceleration, and historical trajectory in the dynamic scene information includes: By using a pre-set behavior prediction model, the location, speed, acceleration, and historical trajectory of each traffic participant are predicted and analyzed to obtain a set of candidate behavioral intentions for the traffic participants. The candidate behavioral intentions are filtered based on the candidate behavioral probabilities in the candidate behavioral intention set to determine the behavioral intentions of the traffic participants and the intention determination index value. The intent threat index value of the traffic participant is determined based on the behavioral intent trajectory corresponding to the traffic participant's behavioral intent and the initial planned trajectory of the vehicle.

6. The vehicle control method according to claim 1, characterized in that, The step of determining a semantic risk score based on the semantic risk feature data, and determining a comprehensive risk score based on the physical risk score and the semantic risk score, includes: Based on the static scene labels, feature transformation rules are determined, and based on the feature transformation rules, the dynamic scene disorder degree and the behavioral intentions and intention index parameters of all traffic participants are transformed to obtain a semantic risk score. The physical risk score and the semantic risk score are fused according to a preset fusion coefficient to obtain a comprehensive risk score; wherein, the physical risk score is determined based on all the relative state parameters between the vehicle and each traffic participant obtained from the environmental data and the map data.

7. The vehicle control method according to claim 1, characterized in that, The preset risk thresholds include a first risk threshold and a second risk threshold, wherein the first risk threshold is less than the second risk threshold; the risk levels include low risk level, medium risk level and high risk level. The step of determining the risk level based on the comprehensive risk score and the preset risk threshold includes: When the comprehensive risk score has not yet reached the first risk threshold, the risk level is determined to be low risk level; When the comprehensive risk score reaches the first risk threshold but has not yet reached the second risk threshold, the risk level is determined to be medium risk. When the comprehensive risk score reaches the second risk threshold, the risk level is determined to be high risk.

8. The vehicle control method according to claim 1, characterized in that, The control of the vehicle to perform driving operations corresponding to the risk level includes: Obtain the standard values ​​of all preset control indicators, and correct the standard values ​​of each preset control indicator according to the risk level to obtain the corrected values ​​of each preset control indicator; the preset control indicators include at least one of maximum acceleration, maximum deceleration, minimum following distance, maximum steering angular velocity, lane keeping weight, and lane change decision benefit value; Based on the initial planned trajectory of the vehicle and the index correction values ​​of all the preset control indicators, the target decision function and target constraints are determined. A target planning trajectory is generated based on the target decision function and the target constraints, and the vehicle is controlled to perform driving operations corresponding to the target planning trajectory.

9. An electronic device, characterized in that, Including processor and memory, among which Memory, used to store computer programs; A processor for executing a program stored in memory to implement the vehicle control method according to any one of claims 1-8.

10. A vehicle, characterized in that, It includes the electronic device as described in claim 9.

Images