Adaptive omnidirectional driving assistance method and device for vehicle based on driving risk field

Abstract

The application relates to an adaptive omnidirectional driving assistance method and device for a vehicle based on a driving risk field, wherein the method comprises the following steps: collecting a driving track of a target vehicle, planning the driving track based on a preset mode, generating a plurality of candidate tracks, and screening the plurality of candidate tracks according to a preset screening strategy to obtain a candidate track cluster; evaluating each candidate track in the candidate track cluster based on a driving risk evaluation model established by a potential energy risk field and a preset loss function to obtain a target track whose evaluation result meets preset requirements; and learning the weight parameters of the loss function corresponding to the target track to perform adaptive learning for omnidirectional assistance driving of the vehicle. Therefore, the problems that functions of existing adaptive driving assistance technologies are isolated from each other, and application scenarios are single are solved.

Description

Technical Field

[0001] This application relates to the field of vehicle driving assistance technology, and in particular to an adaptive omnidirectional driving assistance method and device based on driving risk field. Background Technology

[0002] Advanced driver assistance systems (ADAS) improve driving comfort and safety by reducing the driver's workload or alerting them to potential hazards. To meet the diverse needs of numerous drivers, research is needed into personalized driver assistance methods that adapt to individual driver characteristics. This involves adapting the system to the driver's personal preferences to enhance user experience and improve system performance.

[0003] Generally speaking, adaptive driving assistance systems can be divided into the following two categories:

[0004] 1. Explicit adaptation, which involves surveying drivers' driving preferences through questionnaires or scales and adjusting the mode type and related parameters in the system settings based on the test results;

[0005] 2. Implicit adaptation, which does not involve directly surveying drivers, but rather analyzes their driving behavior to deduce their possible driving preferences and adjusts the model to suit their characteristics. Implicit adaptation methods are more objective and are therefore increasingly becoming the mainstream of adaptive driving assistance methods.

[0006] Depending on the target audience, adaptive methods can be divided into two types: The first is a classification-based method, which first extracts driving features, then classifies drivers based on these features, and finally designs different driving style mode parameters for each type of driver, providing personalized driving assistance control strategies based on driver type; The second is an individual-based method, which, after extracting driving feature parameters, directly provides personalized control parameters for the current individual driver, eliminating the driver classification step. This type of control strategy reproduces the individual's driving style.

[0007] Depending on the controlled object, existing research on adaptive driving assistance systems can be divided into two categories: one is to monitor or control the longitudinal movement of the vehicle, including adaptive cruise control systems and forward collision warning systems; the other is to monitor or control the lateral movement of the vehicle, such as lane departure warning systems and lane keeping assist systems.

[0008] In terms of classification-based longitudinal adaptation, existing methods mainly analyze the characteristic parameters of drivers and classify them using methods such as clustering and fuzzy logic. Then, for each type of driver, different parameters are designed for the reference acceleration curve in ACC (Adaptive Cruise Control), the timing of using and disabling ACC functions, and the parameters of CC (Cruise Control) / ACC switching rules, thereby achieving longitudinal driving assistance adaptation.

[0009] In terms of individual-based longitudinal adaptation, existing methods mainly analyze the characteristics of driver behavior and use different models to adaptively learn these characteristics. For example, some studies establish the relationship between driver actions and TTCi (Inverse of Time-to-Collision) and THW (Time Headway) using linear driver models, and establish a self-learning method based on a recursive least squares algorithm with a forgetting factor. Other studies have proposed an ACC controller architecture based on a linear vehicle tracking model, which also uses a recursive least squares algorithm to solve the problem to reproduce the time headway during manual driving. Still other studies use hidden Markov models, Gaussian mixture regression, random forests, and neural Q-learning to predict future driver actions and achieve adaptation.

[0010] In classification-based lateral adaptive driving, it is typically necessary to categorize drivers into several fixed classes based on different driving characteristic data, and then set different control strategy parameters for different classes of drivers. Several typical methods include:

[0011] ① Based on the offline Gaussian distribution method, the vehicle speed, steering wheel angle, and lateral error of the reference trajectory are used as inputs, and the driver's expected steering ratio is used as the output. The drivers are divided into three categories according to their sensitivity to the error, and different control strategy parameters are set for each category.

[0012] ② Based on the fuzzy rule method, using data on lateral jerk and steering feel factors, drivers are divided into four categories according to their driving aggression, and different assistance torques are provided to drivers of different styles to achieve a personalized LKA (Lane Keeping Assist) system;

[0013] ③ Based on the clustering method, drivers are divided into four categories by the braking deceleration required for the following vehicle to avoid a collision when the vehicle changes lanes. The signal detection theory is used to determine the appropriate threshold for the lane change warning system for each type of driver.

[0014] In terms of individual-based lateral adaptation, existing methods primarily learn to fit the driver's operational characteristics through different models. Several typical methods include:

[0015] ① Use HMM (Hidden Markov Model) and GMR (Gaussian Mixture Regression) to predict the driver's steering actions and provide personalized steering control to prevent the vehicle from deviating from the lane.

[0016] ② Using data such as vehicle speed, relative heading angle, and road curvature, predict the future lateral position deviation of the vehicle based on GMM (Gaussian Mixture Model) and HMM, and then design corresponding lane departure warning strategies.

[0017] ③ Using a steering model composed of the desired path and transfer function, the parameters of the steering model can be fitted based on the driver's driving data in multiple scenarios. The personalized steering model provided for each driver can be used for driving on highways and urban roads.

[0018] The above analysis reveals that existing research can adapt to driver characteristics to a certain extent, achieving single-function lateral or longitudinal driver assistance. However, each driver assistance system operates independently, with varying application scenarios, and decision-making conflicts may arise between systems under complex road conditions. Due to the lack of overall coordinated control among multiple systems, it is difficult to optimize and upgrade the system through simple function aggregation.

[0019] In summary, existing adaptive driving assistance technologies are functionally isolated and have limited application scenarios. There is an urgent need for an omnidirectional driving assistance method that adapts to the characteristics of the driver to overcome these shortcomings. Summary of the Invention

[0020] This application provides an adaptive omnidirectional driving assistance method and device based on driving risk field to solve the problems of isolated functions and limited application scenarios of existing adaptive driving assistance technologies.

[0021] The first aspect of this application provides an adaptive omnidirectional driving assistance method for vehicles based on a driving risk field, comprising the following steps: collecting the driving trajectory of a target vehicle; planning the driving trajectory according to a preset method to generate multiple candidate trajectories; filtering the multiple candidate trajectories according to a preset filtering strategy to obtain a candidate trajectory cluster; evaluating each candidate trajectory in the candidate trajectory cluster based on a driving risk assessment model established by a potential energy risk field and a preset loss function to obtain a target trajectory whose evaluation result meets preset requirements; and learning the weight parameters of the loss function corresponding to the target trajectory to perform adaptive learning for omnidirectional driving assistance.

[0022] Optionally, in one embodiment of this application, the evaluation of each candidate trajectory in the candidate trajectory cluster based on the driving risk assessment model established by the potential energy risk field and the preset loss function includes: obtaining the risk equivalence and front wheel angle probability based on vehicle kinetic energy and front wheel angle data, and calculating the roadside view risk based on the risk equivalence and front wheel angle probability; calculating the driver's view risk through the vehicle's speed, other vehicle speed, and the maximum lateral and longitudinal risk perception distances, and calculating the potential collision risk field using the longitudinal length and lateral width of the target vehicle; and constructing a comprehensive driving risk model based on the roadside view risk, driver's view risk, and potential collision risk.

[0023] Optionally, in one embodiment of this application, the step of planning the driving trajectory based on a preset method to generate multiple candidate trajectories includes: collecting high-speed driving data of the target vehicle, analyzing the lane-changing time and loop speed of different drivers based on the high-speed driving data, obtaining the driving style of the corresponding driver based on the lane-changing time and loop speed of the different drivers; fitting the lateral and longitudinal motion of the vehicle based on a polynomial model to plan the lane-changing trajectory, and generating the multiple candidate trajectories according to the driver's driving style, the planned lane-changing trajectory, and preset trajectory constraints.

[0024] Optionally, in one embodiment of this application, the step of filtering the plurality of candidate trajectories according to a preset filtering strategy to obtain a candidate trajectory cluster includes: transforming the plurality of candidate trajectories from the Frenet coordinate system to the Cartesian coordinate system; and performing collision detection and curvature detection on the plurality of candidate trajectories in the Cartesian coordinate system respectively to obtain the candidate trajectory cluster that meets preset requirements.

[0025] Optionally, in one embodiment of this application, the mathematical expression of the preset loss function is as follows:

[0026] Cosy=αC risk +C speed +C jerk

[0027] Among them, C risk For safety indicators, C speed As an efficiency indicator, C jerk For comfort index, α, β and γ represent the weighting coefficients of the safety index, efficiency index and comfort index, respectively.

[0028] Optionally, in one embodiment of this application, the step of learning the weight parameters of the loss function corresponding to the target trajectory for adaptive learning of omnidirectional assisted driving includes: generating multiple historical trajectories and virtual trajectories based on the target trajectory, wherein the historical trajectories include historical straight-line trajectories and historical lane-changing trajectories, and the virtual trajectories include virtual straight-line trajectories and virtual lane-changing trajectories; calculating the safety index, efficiency index, and comfort index of the historical trajectories and virtual trajectories respectively according to a preset index calculation method to obtain a set of weight parameter inequalities, and performing driving risk weight learning based on the set of weight parameter inequalities and a preset learning method.

[0029] A second aspect of this application provides an adaptive omnidirectional driving assistance device based on a driving risk field, comprising: a screening module for collecting the driving trajectory of a target vehicle, planning the driving trajectory according to a preset method, generating multiple candidate trajectories, and screening the multiple candidate trajectories according to a preset screening strategy to obtain a candidate trajectory cluster; an evaluation module for evaluating each candidate trajectory in the candidate trajectory cluster based on a driving risk assessment model established by a potential energy risk field and a preset loss function to obtain a target trajectory whose evaluation result meets preset requirements; and a learning module for learning the weight parameters of the loss function corresponding to the target trajectory to perform adaptive learning for omnidirectional driving assistance.

[0030] Optionally, in one embodiment of this application, the evaluation module includes: a first calculation unit, used to obtain risk equivalence and front wheel angle probability based on vehicle kinetic energy and front wheel angle data, and calculate roadside view risk based on the risk equivalence and front wheel angle probability; a second calculation unit, used to calculate driver's view risk using the vehicle's speed, other vehicle speed, and maximum lateral and longitudinal risk perception distances, and calculate potential collision risk field using the longitudinal length and lateral width of the target vehicle; and a modeling unit, used to construct a comprehensive driving risk model based on the roadside view risk, driver's view risk, and potential collision risk.

[0031] Optionally, in one embodiment of this application, the screening module includes: an analysis unit, configured to collect high-speed driving data of the target vehicle, and analyze the lane-changing time and loop speed of different drivers based on the high-speed driving data, and obtain the driving style of the corresponding driver based on the lane-changing time and loop speed of the different drivers; and a fitting unit, configured to fit the lateral and longitudinal motion of the vehicle based on a polynomial model to plan the lane-changing trajectory, and generate the multiple candidate trajectories according to the driver's driving style, the planned lane-changing trajectory, and preset trajectory constraints.

[0032] Optionally, in one embodiment of this application, the filtering module further includes: a conversion unit for converting the plurality of candidate trajectories from the Frenet coordinate system to the Cartesian coordinate system; and a detection unit for performing collision detection and curvature detection on the plurality of candidate trajectories in the Cartesian coordinate system respectively, to obtain the candidate trajectory cluster that meets preset requirements.

[0033] Optionally, in one embodiment of this application, the mathematical expression of the preset loss function is as follows:

[0034] Cost = αC risk +C speed +C jerk

[0035] Among them, C risk For safety indicators, c speed As an efficiency indicator, C jerk For comfort index, α, β and γ represent the weighting coefficients of the safety index, efficiency index and comfort index, respectively.

[0036] Optionally, in one embodiment of this application, the learning module includes: a generation unit, configured to generate multiple historical trajectories and virtual trajectories based on the target trajectory, wherein the historical trajectories include historical straight-ahead trajectories and historical lane-changing trajectories, and the virtual trajectories include virtual straight-ahead trajectories and virtual lane-changing trajectories; and a third calculation unit, configured to calculate the safety index, efficiency index, and comfort index of the historical trajectories and virtual trajectories respectively according to a preset index calculation method to obtain a set of weight parameter inequalities, and to perform driving risk weight learning based on the set of weight parameter inequalities and a preset learning method.

[0037] 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, wherein the processor executes the program to implement the adaptive omnidirectional driving assistance method for vehicles based on driving risk fields as described in the above embodiments.

[0038] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described adaptive omnidirectional driving assistance method for vehicles based on driving risk fields.

[0039] Therefore, the embodiments of this application have the following beneficial effects:

[0040] The embodiments of this application can collect the driving trajectory of the target vehicle, plan the driving trajectory based on a preset method to generate multiple candidate trajectories, and filter the multiple candidate trajectories according to a preset screening strategy to obtain a candidate trajectory cluster. Each candidate trajectory in the candidate trajectory cluster is evaluated based on a driving risk assessment model established by a potential energy risk field and a preset loss function to obtain a target trajectory whose evaluation result meets preset requirements. The weight parameters of the loss function corresponding to the target trajectory are learned to perform adaptive learning for omnidirectional assisted driving. Therefore, this application provides a reliable theoretical basis for the design of safety indicators in the loss function, ensures the feasibility of the trajectory through collision detection and curvature detection, and comprehensively considers the safety, efficiency, and comfort of the loss function, setting different parameter weights for different types of drivers, thus realizing adaptive omnidirectional driving assistance based on a driving risk field. This solves the problems of isolated functions and limited application scenarios in existing adaptive driving assistance technologies.

[0041] 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

[0042] 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:

[0043] Figure 1 This is a flowchart of an adaptive omnidirectional driving assistance method for vehicles based on a driving risk field, according to an embodiment of this application.

[0044] Figure 2 An execution logic architecture diagram of an adaptive omnidirectional driving assistance method based on driving risk field is provided as an embodiment of this application;

[0045] Figure 3 (a) A schematic diagram illustrating the change of longitudinal position of different types of driving trajectories over time, provided as an embodiment of this application;

[0046] Figure 3 (b) A schematic diagram illustrating the variation of the lateral position of different types of driving trajectories over time, provided as an embodiment of this application;

[0047] Figure 4 (a) A schematic diagram illustrating the variation of longitudinal velocity over time for different types of driving trajectories, provided as an embodiment of this application;

[0048] Figure 4 (b) A schematic diagram illustrating the variation of lateral velocity over time for different types of driving trajectories, provided as an embodiment of this application;

[0049] Figure 5 A schematic diagram of trajectory fitting results for polynomials of different orders is provided as an embodiment of this application;

[0050] Figure 6 A schematic diagram of trajectory fitting error for polynomials of different orders is provided as an embodiment of this application;

[0051] Figure 7 (a) A schematic diagram illustrating the change of the longitudinal position of a candidate trajectory over time, provided in an embodiment of this application;

[0052] Figure 7 (b) A schematic diagram illustrating the change of the lateral position of a candidate trajectory over time, provided as an embodiment of this application;

[0053] Figure 8 (a) A schematic diagram illustrating the variation of longitudinal velocity of a candidate trajectory over time, provided in an embodiment of this application;

[0054] Figure 8 (b) A schematic diagram illustrating the variation of the lateral velocity of a candidate trajectory over time, provided as an embodiment of this application;

[0055] Figure 9 A schematic diagram of the split-axis theorem is provided for one embodiment of this application;

[0056] Figure 10 A histogram and probability fitting plot of the frequency distribution of front wheel steering angles are provided as an embodiment of this application;

[0057] Figure 11 A roadside view risk diagram provided for one embodiment of this application;

[0058] Figure 12 A roadside view risk diagram considering vehicle shape is provided as an embodiment of this application;

[0059] Figure 13 (a) A schematic diagram of the horizontal and vertical positional relationship between a historical lane-changing trajectory and a virtual trajectory provided in an embodiment of this application;

[0060] Figure 13 (b) A schematic diagram of the longitudinal velocity of a historical lane-changing trajectory and a virtual trajectory over time, provided as an embodiment of this application;

[0061] Figure 14 (a) A schematic diagram of the horizontal and vertical positional relationship between a historical straight trajectory and a virtual trajectory provided in an embodiment of this application;

[0062] Figure 14(b) A schematic diagram showing the change of longitudinal velocity over time in a historical straight trajectory and a virtual trajectory, provided as an embodiment of this application;

[0063] Figure 15 A risk weight learning effect diagram for an aggressive driver is provided as an embodiment of this application;

[0064] Figure 16 A risk weight learning effect diagram for a typical driver is provided as an embodiment of this application;

[0065] Figure 17 A risk weight learning effect diagram for a conservative driver provided in one embodiment of this application;

[0066] Figure 18 A graph illustrating the learning effect of evaluation index weights for an aggressive driver, provided as an embodiment of this application;

[0067] Figure 19 A learning effect diagram of the evaluation index weights for a typical driver is provided as an embodiment of this application;

[0068] Figure 20 A learning effect diagram of evaluation index weights for a conservative driver is provided as an embodiment of this application;

[0069] Figure 21 This is an example diagram of an adaptive omnidirectional driving assistance device for vehicles based on a driving risk field according to an embodiment of this application;

[0070] Figure 22 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0071] Among them, 10-Adaptive vehicle omnidirectional driving assistance device based on driving risk field, 100-Filtering module, 200-Evaluation module, 300-Learning module, 2201-Memory, 2202-Processor, 2203-Communication interface. Detailed Implementation

[0072] 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.

[0073] The following describes an adaptive omnidirectional driving assistance method and apparatus based on a driving risk field according to embodiments of this application, with reference to the accompanying drawings. Addressing the problems mentioned in the background section, this application provides an adaptive omnidirectional driving assistance method based on a driving risk field. In this method, the driving trajectory of the target vehicle is collected, planned according to a preset method, and multiple candidate trajectories are generated. These candidate trajectories are then filtered according to a preset screening strategy to obtain a candidate trajectory cluster. Each candidate trajectory in the candidate trajectory cluster is evaluated based on a driving risk assessment model established by a potential energy risk field and a preset loss function to obtain a target trajectory whose evaluation result meets preset requirements. The weight parameters of the loss function corresponding to the target trajectory are learned to perform adaptive learning for omnidirectional driving assistance. Therefore, this application provides a reliable theoretical basis for the design of safety indicators in the loss function, ensures the feasibility of the trajectory through collision detection and curvature detection, and comprehensively considers the safety, efficiency, and comfort of the loss function, setting different parameter weights for different types of drivers, thus realizing adaptive omnidirectional driving assistance based on a driving risk field. This solves the problems of isolated functions and limited application scenarios in existing adaptive driving assistance technologies.

[0074] Specifically, Figure 1 This is a flowchart of an adaptive omnidirectional driving assistance method for vehicles based on a driving risk field, provided in an embodiment of this application.

[0075] like Figure 1 As shown, this adaptive omnidirectional driving assistance method based on driving risk field includes the following steps:

[0076] In step S101, the driving trajectory of the target vehicle is collected, the driving trajectory is planned based on a preset method, multiple candidate trajectories are generated, and multiple candidate trajectories are filtered according to a preset filtering strategy to obtain a candidate trajectory cluster.

[0077] In the embodiments of this application, the driving trajectory of the vehicle can first be collected through a real vehicle data acquisition platform and a driving simulator data acquisition platform. Based on the collected driving trajectory, the driver's style is classified and identified, and a lane-changing trajectory is planned to generate candidate trajectories. Then, the candidate trajectories are subjected to operations such as trajectory collision detection and trajectory curvature detection to obtain a candidate trajectory cluster, thereby providing a reliable theoretical basis for the subsequent generation of target trajectories and learning of driving risk weights.

[0078] Optionally, in one embodiment of this application, the driving trajectory is planned based on a preset method to generate multiple candidate trajectories, including: collecting high-speed driving data of the target vehicle, analyzing the lane-changing time and loop speed of different drivers based on the high-speed driving data, obtaining the driving style of the corresponding driver based on the lane-changing time and loop speed of different drivers; fitting the lateral and longitudinal motion of the vehicle based on a multinomial model to plan the lane-changing trajectory, and generating multiple candidate trajectories according to the driver's driving style, the planned lane-changing trajectory and preset trajectory constraints.

[0079] It should be noted that, in the embodiments of this application, when a vehicle is driving on a structured road, the driving behavior is mainly divided into two categories: following and lane changing. Since lane changing behavior is usually more complex than following behavior and can better reflect the driver's driving characteristics, the embodiments of this application can mainly introduce the trajectory planning of the vehicle's lane changing process, and treat the following trajectory as a special case of the lane changing trajectory.

[0080] Specifically, embodiments of this application include steps such as planning driving trajectories and generating multiple candidate trajectories, including trajectory analysis of drivers with different styles, lane-change trajectory planning, and candidate trajectory generation. Figure 2 As shown, the specific process is as follows:

[0081] 1. Trajectory analysis of drivers with different driving styles:

[0082] High-speed driving data is collected, and the driver's lane-changing trajectory is analyzed. Considering that there is no significant difference between left and right lane-changing trajectories, the embodiments of this application can treat the right lane-changing trajectory symmetrically, that is, take the opposite values ​​for lateral position Y, lateral velocity Vy, and lateral acceleration Ay; the average values ​​of aggressive, normal, and conservative driving trajectories are calculated respectively, and the changes of longitudinal position X and lateral position Y with time t are as follows: Figure 3 (a) and Figure 3 As shown in (b), the variation of longitudinal velocity Vx and lateral velocity Vy with time t is as follows: Figure 4 (a) and Figure 4 As shown in (b).

[0083] Depend on Figure 3 (a) Figure 3 (b) and Figure 4 (a) Figure 4 (b) It can be seen that aggressive drivers have shorter lane-changing times and greater lateral speeds during lane-changing; conservative drivers have obvious acceleration behavior during lane-changing and longer lane-changing times; and average drivers have no obvious acceleration behavior during lane-changing and also have longer lane-changing times.

[0084] 2. Lane change trajectory planning:

[0085] Those skilled in the art should understand that existing lane-changing trajectory models mainly include circular arc lane-changing trajectories, trigonometric function lane-changing trajectories, trapezoidal acceleration lane-changing trajectories, spline curve lane-changing trajectories, and polynomial lane-changing trajectories. Among them, the polynomial lane-changing trajectory has a simple expression form, and the fitting accuracy can be improved with the increase of the polynomial order. An appropriate order can be selected according to the requirements. Therefore, the embodiments of this application can use a polynomial model to construct the lane-changing trajectory, and its mathematical form is shown in equation (1):

[0086]

[0087] Taking the lateral motion of a conservative lane-changing trajectory as an example, embodiments of this application construct third, fourth, and fifth-order polynomials to fit it, and the results are as follows: Figure 5 , Figure 6 As shown.

[0088] Depend on Figure 5 , Figure 6 It can be seen that the fitting error of the third, fourth, and fifth order polynomial trajectories for the lateral motion of the vehicle decreases sequentially. Therefore, the embodiments of this application calculate the sum of squared residuals and the coefficient of determination R of the third, fourth, and fifth order polynomial fitting results, respectively. 2 Table 1 shows a comparison of the fitting results for third, fourth, and fifth degree polynomials.

[0089] Table 1

[0090]

[0091] Analysis of the vehicle lane-changing process reveals that the vehicle's lateral position at the initial and final moments of the lane change should be at the center of the original lane and the target lane, respectively, with both lateral velocity and lateral acceleration being zero. This ensures the continuity of the velocity and acceleration curves. Therefore, there are six constraints on the initial and final states of the vehicle's lateral motion. Ultimately, this embodiment selects a fifth-order polynomial trajectory for lateral motion planning in the Frenet coordinate system, with the road centerline as the longitudinal (s) direction and the normal direction of the centerline as the lateral (d) direction. Based on the initial and final states of the lane change, the coefficient a is obtained. i The system of equations is shown in equation (2):

[0092] d(t0)=a0+a1t0+a2t0 2 +a3t0 3 +a4t0 4 +a5t0 5

[0093]

[0094]

[0095] d(t1)=a0+a1t1+a2t1 2 +a3t1 3 +a4t1 4 +a5t1 5

[0096]

[0097]

[0098] Where a0, a1, ..., a5 are the coefficients to be solved, and t0 and t1 are the start and end times of the lane change, respectively.

[0099] Given the initial lane-changing state of the vehicle, d(t0) = d0. And the lane change ending state d(t1) = d1, Equation (2) can then be solved to obtain the fifth-order polynomial of the vehicle's lateral motion.

[0100] Similarly, for longitudinal movement of the vehicle, the embodiments of this application require a given initial lane-changing state s(t0). And constrain the final state of the vehicle lane change. The longitudinal position of the vehicle at the end of the lane change is not explicitly required; therefore, there are a total of 5 constraints. The embodiment of this application selects a fourth-order polynomial for longitudinal motion planning; based on the initial and final states, the coefficient b is obtained. i The system of equations is shown in equation (3). Solving it yields the fourth-order polynomial of the vehicle's longitudinal motion.

[0101] s(t0)=b0+b1t0+b2t0 2 +b3t0 3 +b4t0 4

[0102]

[0103]

[0104]

[0105]

[0106] 3. Generate candidate trajectories:

[0107] By providing different initial and final states and lane-changing durations, different lane-changing trajectories can be obtained. Furthermore, by setting the lateral displacement of the vehicle as d(t1) = d(t0), a candidate trajectory cluster for the vehicle following behavior can be obtained.

[0108] When setting the constraints for the trajectory, the embodiments of this application can make targeted and personalized settings according to the lane-changing characteristics of various types of drivers. For example, a shorter lane-changing time can be set for aggressive drivers, i.e., a smaller t1-t0; and a shorter final speed can be set for conservative drivers. Greater than the initial velocity Furthermore, embodiments of this application should also include deceleration trajectories to ensure vehicle safety in emergency situations; all trajectories should conform to the vehicle's dynamic performance, i.e., the maximum acceleration and deceleration should not exceed the vehicle's performance limits. The maximum deceleration of the vehicle is limited by the road surface adhesion coefficient. The peak adhesion coefficient of dry asphalt pavement is approximately 0.8 to 0.9, and the sliding adhesion coefficient is approximately 0.75. Therefore, the maximum deceleration of the vehicle is assumed to be -8 m / s². 2 The maximum acceleration of a vehicle is limited by its driving force, and there are significant differences between different vehicles. Let's assume the maximum acceleration of a vehicle is 3 m / s². 2 .

[0109] Let the initial time of the lane change be t0 = 0s, and the ending times be t1 = 4, 5, 6s respectively. The initial lateral position is d(t0) = 0, and the final lateral positions are d(t1) = -3.5, 0, 3.5m respectively. The initial longitudinal velocity is... The final longitudinal velocities are respectively A set of candidate trajectory samples that meet the vehicle performance limits is obtained, for example Figure 7 (a) Figure 7 (b) and Figure 8 (a) Figure 8 As shown in (b).

[0110] Figure 7 (a) Figure 7 (b) and Figure 8 (a) Figure 8 (b) covers various driving trajectories with different characteristics, which can meet the needs of vehicle driving trajectory planning; in addition, the vehicle speed changes smoothly and has good comfort. In practical applications, the embodiments of this application can adjust the parameters according to the specific characteristics of the driver to generate a more personalized driving trajectory.

[0111] Optionally, in one embodiment of this application, multiple candidate trajectories are filtered according to a preset filtering strategy to obtain a candidate trajectory cluster, including: transforming multiple candidate trajectories from the Frenet coordinate system to the Cartesian coordinate system; performing collision detection and curvature detection on multiple candidate trajectories in the Cartesian coordinate system respectively to obtain a candidate trajectory cluster that meets preset requirements.

[0112] After generating candidate trajectories, embodiments of this application can also transform the trajectories in the Frenet coordinate system to the Cartesian coordinate system, and then perform collision detection and curvature detection on them respectively to filter out a cluster of collision-free candidate trajectories that conform to the vehicle motion characteristics.

[0113] Specifically, the process of screening candidate trajectories in the embodiments of this application is as follows:

[0114] 1. Trajectory Collision Detection:

[0115] After the vehicle's trajectory is planned, there may be potential conflicts between the trajectory of the vehicle and the trajectory of other vehicles in the surrounding environment. Therefore, it is necessary to perform collision detection on the expected positions of the vehicle and other vehicles at each moment in the planned trajectory.

[0116] Common collision detection methods include grid-based methods, circular bounding box-based methods, axis-aligned bounding box-based methods, and directional bounding box-based methods. Grid-based methods divide the map into a grid, where the grid size must be larger than the largest object being detected. For each object, only collisions with its own cell and other objects in the surrounding grid need to be detected. This method can effectively improve detection efficiency when there are many objects to detect. Circular bounding box-based methods use multiple circles to enclose the object being detected, simplifying collision detection between complex shapes to detection between two circles. This method is computationally fast but has lower accuracy. Axis-aligned bounding box-based methods use borders parallel to the coordinate axes to enclose the object and then detect the overlap between the two borders. This method is simple and intuitive, but for tilted or irregularly shaped objects, there is significant redundancy within the borders. Directional bounding box-based methods use rectangles in any direction to enclose the object, providing a tighter enclosure than axis-aligned bounding boxes. This method has higher accuracy but is slower.

[0117] Since the vehicle's planar appearance is approximately rectangular and it veers while driving, this application's embodiments employ a collision detection method based on directional bounding boxes. The separation axis theorem is used to detect whether two directional bounding boxes intersect. The principle is to find a line orthogonal to the surface of the directional bounding box as the separation axis. If the projections of the two directional bounding boxes on each axis overlap, then the two directional bounding boxes intersect. Figure 9 As shown.

[0118] In the embodiments of this application, it is assumed that the center of vehicle 1 is located at point (x). c1 ,y c1 The heading angle is θ1, the vehicle length is L1, and the vehicle width is D1; ​​the center of vehicle 2 is located at point (x). c2 ,y c2Given a heading angle of θ2, vehicle length L2, and vehicle width D2; the condition for applying the separation axle theorem is the projection of vehicle 1. Projection of vehicle 2 Whether they intersect can be determined by whether, in embodiments of this application, half of the sum of the projected lengths of vehicle 1 and vehicle 2 on the separation axis is greater than the projected distance between their centers. The two projections intersect, and the result can be calculated using geometric relationships. The length is shown in equation (4):

[0119]

[0120]

[0121]

[0122]

[0123]

[0124] Therefore, by calculating the four separation axles of the two vehicles in sequence according to formula (4), it can be determined whether the two vehicles have collided.

[0125] 2. Trajectory curvature detection:

[0126] The trajectory curvature detection in the embodiments of this application is divided into two parts. The first part is the minimum turning radius R of the vehicle. min It should be less than or equal to the minimum radius of curvature ρ of the trajectory. min The calculation formulas for the two are shown in equation (5):

[0127]

[0128]

[0129] R min ≤ρ min

[0130] Where, δ max is the maximum turning angle of the vehicle's front wheels, and l is the vehicle's wheelbase.

[0131] The second part aims to prevent the vehicle from skidding during steering. The maximum adhesion provided by the road surface should be greater than the centripetal force required for the vehicle to turn. However, since the tire's lateral deflection characteristics will enter the nonlinear region when the vehicle experiences large lateral acceleration, the embodiments of this application further limit the maximum lateral acceleration to 0.4g. The limiting relationship between the trajectory curvature radius ρ and the vehicle speed v is shown in equation (6).

[0132]

[0133] Therefore, the embodiments of this application sequentially perform curvature detection and screening on the above-obtained vehicle collision-free trajectory using equations (5) and (6) to obtain a cluster of collision-free candidate trajectories that conform to the vehicle motion characteristics, providing reliable data support for the subsequent evaluation of vehicle trajectories.

[0134] In step S102, each candidate trajectory in the candidate trajectory cluster is evaluated based on the driving risk assessment model established by the potential energy risk field and the preset loss function, and the target trajectory whose evaluation result meets the preset requirements is obtained.

[0135] After obtaining the candidate trajectory cluster, the embodiments of this application can further establish a driving risk assessment model and evaluate each candidate trajectory in the obtained candidate trajectory cluster by combining a loss function, thereby obtaining the target trajectory, providing a reliable theoretical basis for the adaptive learning of subsequent omnidirectional assisted driving of vehicles.

[0136] Optionally, in one embodiment of this application, each candidate trajectory in the candidate trajectory cluster is evaluated based on a driving risk assessment model established by the potential energy risk field and a preset loss function. This includes: obtaining risk equivalence and front wheel angle probability based on vehicle kinetic energy and front wheel angle data, and calculating roadside view risk based on risk equivalence and front wheel angle probability; calculating driver's view risk based on the vehicle's speed, other vehicle speed, and maximum lateral and longitudinal risk perception distances, and calculating the potential collision risk field using the longitudinal length and lateral width of the target vehicle; and constructing a comprehensive driving risk model based on roadside view risk, driver's view risk, and potential collision risk.

[0137] Understandably, assessing the risks during vehicle operation is fundamental to researching driver assistance algorithms, and TTC (Time To Collision) and TTCi are among the most widely used metrics by researchers. Another important metric is THW, which drivers typically control within their desired range and keep TTCi near zero during driving. In addition, there are metrics based on distance, acceleration, and potential energy fields.

[0138] Since the potential energy field can measure the all-around collision risk around a vehicle, it is more comprehensive than longitudinal risk assessment indicators such as TTC and THW. Therefore, embodiments of this application can use a potential energy field-based strategy to quantitatively assess the risks encountered by a vehicle during driving.

[0139] Specifically, the main steps for establishing a driving risk assessment model according to the embodiments of this application are as follows:

[0140] 1. Establish a roadside view risk model.

[0141] The roadside view refers to the perspective of a third party, from which all moving traffic participants (vehicles, cyclists, pedestrians, etc.) on the road have a certain amount of kinetic energy. When a traffic accident occurs, the greater the kinetic energy that both parties have before the collision, the greater the damage and loss caused by the collision. On the other hand, considering that with a certain amount of kinetic energy, the farther away from the moving object, the smaller the perceived risk.

[0142] Therefore, embodiments of this application can use kinetic energy divided by distance to characterize the risk at a point P on the road. The dimension of the risk calculated by dividing energy by distance is force, so it can be called equivalent force, as shown in equation (7):

[0143]

[0144] Where E is the kinetic energy of the moving object, m and v are the mass and velocity of the object, respectively, Δx is the distance between the moving object and point P, and F is the magnitude of the equivalent force.

[0145] Since the risk level at a certain point on the road is related not only to the equivalent force F, but also to the relative position of that point and the moving object, and the direction of the object's motion, the embodiments of this application statistically analyze the front wheel steering angle data of all drivers during highway tests, as follows: Figure 10 As shown.

[0146] Depend on Figure 10 It can be seen that the front wheel steering angle data is basically symmetrically distributed, with the probabilities of turning left and right being roughly the same. The shape of this distribution is relatively steep at the center, which does not conform to the normal distribution law. Therefore, the embodiment of this application uses the Laplace distribution to fit the front wheel steering angle probability, and its probability density function is shown in Equation (8):

[0147]

[0148] Where δ is the front wheel steering angle, with parameters λ = 0.27 and μ = 0.

[0149] According to vehicle dynamics, the relationship between the front wheel steering angle and the turning radius when a vehicle is turning is shown in equation (9):

[0150]

[0151] Where R is the turning radius, K is the stability factor, v is the vehicle speed, and l is the vehicle wheelbase.

[0152] Combining equations (7) to (9) above, the roadside visual risk F1 posed by a moving vehicle to a point P in front of the road is obtained as shown in equation (10):

[0153] w = p(δ) / p(0)

[0154]

[0155] Where δ is the front wheel turning angle required for the vehicle to travel to point P, w is the weight of the trajectory chosen by the vehicle to travel to point P, and Δr is the arc length traversed by the vehicle to travel to point P.

[0156] In practical applications, embodiments of this application can generate a cluster of future trajectories for a vehicle under different front wheel steering angles, and calculate the risk change patterns of each point on the trajectory based on the different front wheel steering angles and the different distances from the vehicle.

[0157] Suppose a moving vehicle is located at point X = 10m, Y = 5m, with a speed of v = 20m / s, a mass of m = 1500kg, and a maximum steering angle δ of the front wheels. max =30°, calculate the roadside viewing angle risk generated by this vehicle as follows: Figure 11 As shown, by Figure 11 It can be seen that there is no risk behind and to the left of the vehicle, mainly because the vehicle cannot travel to these positions in its current state due to kinematic constraints. In addition, at the same distance, the risk directly in front of the vehicle is greater than that to the side; and in the same direction, the risk is less at a greater distance from the vehicle than at a closer distance. The above data is consistent with the subjective perception of risk assessment.

[0158] However, this roadside view risk model has shortcomings. When calculating the vehicle trajectory, it treats the vehicle as a point mass and does not consider the shape of the vehicle body. In reality, since the vehicle body has a certain width D and length L, there is a certain risk at all the locations the vehicle body passes through when the vehicle is moving. Therefore, when calculating the risk of a certain point P in space, each point on the vehicle body should be calculated separately, and the maximum value should be taken as the actual risk of point P.

[0159] In the specific implementation process, in order to improve the calculation speed, the embodiments of this application can change the point selection from the vehicle body to the selection from the four sides of the vehicle (front, rear, left, and right), thereby reducing the calculation complexity from O(·L) to O(+L) without affecting the accuracy of the risk calculation results. Figure 12 As shown.

[0160] 2. Establish a driver-perspective risk model:

[0161] It should be noted that the driver's perspective refers to the view from the first point of view of the traffic participant, where other objects on the road have a certain relative speed to the driver. From this perspective, the risk of collision with vehicles traveling in the same direction at similar speeds ahead is relatively low, while the risk of collision with objects with significantly different speeds (such as stationary obstacles or pedestrians crossing the road) is relatively high. Therefore, the embodiments of this application are based on the form of equation (7), and the driver's perspective risk F2 is calculated as shown in equation (11):

[0162]

[0163]

[0164]

[0165] in, and These represent the vector velocities of the driver and other traffic participants, respectively, where Δx is the weighted distance, and D is the vector velocity. r and L r d represents the relative distance between the two vehicles in the horizontal and vertical directions, respectively. max and s max These represent the maximum risk perception distance in the horizontal and vertical directions, respectively.

[0166] 3. Establish a potential collision risk model:

[0167] In order to avoid other road users, drivers should leave a certain safe distance when passing each other. The greater the relative speed between the two parties, the greater the safe distance should be. Therefore, the embodiments of this application can define the potential collision risk field F3 as shown in equation (12):

[0168]

[0169] Where L and D are the longitudinal length and lateral width of the vehicle, respectively, and w L and w D These are the corresponding proportionality coefficients. L r and D r These represent the relative distances between the two vehicles, both longitudinally and laterally.

[0170] 4. Establish a comprehensive driving risk model:

[0171] Based on the above analysis, the embodiments of this application can obtain a comprehensive driving risk F. total As shown in equation (13):

[0172]

[0173] F1, F2, and F3 represent roadside view risk, driver view risk, and potential collision risk, respectively; α1, α2, and α3 are their corresponding weighting coefficients.

[0174] Therefore, the embodiments of this application establish a comprehensive driving risk model by creating a roadside view risk model, a driver's view risk model, and a potential collision risk model, thereby effectively and reliably quantifying the risks encountered by the vehicle during driving.

[0175] Optionally, in one embodiment of this application, the mathematical expression of the preset loss function is as follows:

[0176] Cost = αC risk +C speed +C jerk

[0177] Among them, C risk For safety indicators, C speed As an efficiency indicator, C jerk For comfort index, α, β, and γ represent the weighting coefficients of safety index, efficiency index, and comfort index, respectively.

[0178] It should be noted that after obtaining the cluster of collision-free executable trajectories and establishing a comprehensive driving risk model, the embodiments of this application also need to use a loss function to evaluate each trajectory, and finally output the trajectory with the minimum loss function as the target trajectory. The ideal target trajectory should meet the following characteristics:

[0179] (1) High feasibility: The acceleration and deceleration of the planned trajectory should meet the dynamic performance of the vehicle, and there are also corresponding requirements for the trajectory curvature and the lateral acceleration of the vehicle to ensure the feasibility of subsequent vehicle control.

[0180] (2) High safety: Avoid collisions with other road users, maintain a certain distance from other vehicles while driving, and leave a suitable safety space;

[0181] (3) High efficiency: Under the premise of meeting safety requirements, the vehicle travels at a relatively high speed to reach the destination as soon as possible and shorten the travel time;

[0182] (4) Good comfort: Try to avoid sudden acceleration and deceleration when driving to improve the driving experience.

[0183] The feasibility of the trajectory, as mentioned above, has already been ensured in the trajectory selection stage. Therefore, embodiments of this application can design a loss function with safety, efficiency, and comfort as objectives, discretizing the total trajectory duration into t1,2,… n If there are n points in total, the loss function is as shown in equation (14):

[0184]

[0185]

[0186]

[0187] Cost = αC risk +C speed +C jerk

[0188] Where α1, α2 and α3 are the weighting coefficients of the three driving risks, and α, β and γ represent the weighting coefficients of the three evaluation indicators.

[0189] Safety Index C risk A driving risk assessment model is used to measure the safety of the trajectory, where F1(t) is used in the formula. i ), F2(t i ) and F3(t i ) represent the trajectory at time t. i The risk from the roadside view, the risk from the driver's view, and the potential collision risk at any given moment are represented by α1, α2, and α3, which are their respective weighting coefficients; n in the denominator represents C. risk The average risk represents the entire trajectory and is independent of the degree of discretization; the 10 in the denominator 3 Indicates that C risk The order of magnitude is reduced to be similar to that of other indicators to avoid situations where the weighting coefficients differ too much;

[0190] High efficiency index C speed Using the target vehicle speed v target The vehicle speed v(t) at each moment in the trajectory i The efficiency of a trajectory is measured by the square of the difference between the two terms. Since drivers' actual need is not to achieve the highest efficiency within a few seconds on a single trajectory, but rather to find routes with high long-term efficiency, t is used in the formula. i 2 Indicates the t-th i Weighted average of driving efficiency at each moment, using t i 2 The mathematical properties of the equation allow for a greater weighting of vehicle speeds closer to the end of the trajectory, thus identifying "more promising" target trajectories; the denominator contains... This indicates the normalization process for the weights, 10. 2 Used to transfer C speed The order of magnitude decreased to be similar to other indicators;

[0191] Comfort Index C jerk Use jerk (accelerometer) i The degree of perceived comfort is measured by the square of the term (c), with a smaller rate of change in acceleration indicating a better driving experience; the denominator n represents C. jerk The average comfort level of the entire trajectory is independent of the degree of discretization.

[0192] Therefore, the embodiments of this application, by designing the above-mentioned loss function, fully consider safety, efficiency and comfort, and ensure the safety and reliability of the adaptive learning of subsequent omnidirectional assisted driving of the vehicle.

[0193] In step S103, the weight parameters of the loss function corresponding to the target trajectory are learned to perform adaptive learning for omnidirectional assisted driving of the vehicle.

[0194] After obtaining the target trajectory, the embodiments of this application can further obtain the relevant weight parameters of the target trajectory based on a series of calculations, thereby realizing adaptive learning of omnidirectional assisted driving of the vehicle.

[0195] Optionally, in one embodiment of this application, learning the weight parameters of the loss function corresponding to the target trajectory to perform adaptive learning for omnidirectional assisted driving of the vehicle includes: generating multiple historical trajectories and virtual trajectories based on the target trajectory, wherein the historical trajectories include historical straight-ahead trajectories and historical lane-changing trajectories, and the virtual trajectories include virtual straight-ahead trajectories and virtual lane-changing trajectories; calculating the safety index, efficiency index, and comfort index of the historical trajectories and virtual trajectories respectively according to a preset index calculation method to obtain a set of weight parameter inequalities, and performing driving risk weight learning based on the set of weight parameter inequalities and a preset learning method.

[0196] It should be noted that when each driver remains rational while driving, their acceleration, deceleration, and steering operations represent the optimal choices they make under the current circumstances, and the trajectory of the vehicle driven by the driver is the trajectory with the best virtual comprehensive evaluation index in their mind, then the embodiments of this application can learn the weight parameters of the loss function from the driver's historical driving trajectory.

[0197] Specifically, firstly, the embodiments of this application can extract the driver's lane-changing trajectory, and then, based on information such as initial and final speeds and whether a lane change occurred, generate multiple virtual trajectories around the actual lane-changing trajectory. Table 2 shows the parameter settings for historical lane-changing trajectories and virtual trajectories.

[0198] Table 2

[0199]

[0200] Randomly select a real lane-change trajectory, and generate historical and virtual trajectories based on the parameters in Table 2, as shown below. Figure 13 (a) Figure 13 As shown in (b).

[0201] Furthermore, embodiments of this application calculate the safety, efficiency, and comfort indices of historical and virtual trajectories, respectively, and the loss function Cost of the historical lane-changing trajectory. ta0 It should be less than or equal to the loss function Cost of each virtual trajectory. tai As shown in equation (15):

[0202] Cost ta0 =αC risk_ta0 +βC speed_ta0 +γC jerk_ta0

[0203] Cost tai =αC risk_tai+βC speed_tai +γC jerk_tai (15)

[0204] Cost ta0 ≤Cost tai (i = 1, 2, ..., 7)

[0205] Furthermore, the timing of the driver's lane change is also crucial. If the driver fails to change lanes within a few seconds before the lane change begins, it indicates that the loss function of choosing the straight-ahead trajectory at that time is less than that of an immediate lane change. Therefore, embodiments of this application can also generate multiple virtual trajectories around the historical straight-ahead trajectory prior to the lane change. Table 3 shows the parameter settings for the historical straight-ahead trajectory and the virtual trajectories.

[0206] Table 3

[0207]

[0208] Based on the parameters in Table 3, the historical trajectory and virtual trajectory are generated as follows: Figure 14 (a) Figure 14 As shown in (b).

[0209] Similarly, embodiments of this application also require calculating the safety, efficiency, and comfort indices of the historical trajectory and the virtual trajectory separately. The loss function Cost for the historical straight-line trajectory... tbs It should be less than or equal to the loss function Cost of each virtual trajectory. tbi As shown in equation (16):

[0210] Cost tb0 =C risk_ +C speed_ +C jerk_

[0211] Cost tbi =αC risk_tbi +βC speed_tbi +γC jerk_tbi (16)

[0212] Cost tb0 ≤Cost tbi (=1,2,…,7)

[0213] For each historical trajectory, a corresponding virtual trajectory is set. The embodiments of this application can be calculated according to equations (15) and (16) to obtain thousands of inequalities about weights α1, α2, α3, α, β and γ.

[0214] In an ideal situation, solving this system of inequalities can learn the range of values of the weight coefficients of the virtual evaluation index in the driver's mind. However, in actual modeling of the loss function, it is impossible to be exactly the same as the driver's perception. Therefore, there are certain errors in some data, resulting in no solution to the system of inequalities. In this regard, the embodiments of this application can change the goal to finding a non-zero range of weight values, that is, the solution set, to make the number of inequalities that hold in the system of inequalities the largest.

[0215] Specifically, the embodiments of this application can set the number of weights as r, the total number of linear inequalities as N, and the maximum number of simultaneously established inequalities as M, where r ≤ M < N; when r = 2, the solution set of the system of linear inequalities in the two-dimensional plane is surrounded by multiple straight lines; when r = 3, the solution set of the system of linear inequalities in the three-dimensional space is surrounded by multiple two-dimensional planes; when r > 3, the solution set of the system of linear inequalities in the r-dimensional space is surrounded by multiple (r - 1)-dimensional hyperplanes; in addition, the vertices of the solution set are the cases where r inequalities take equal signs and the other M - r inequalities all hold.

[0216] For all N inequalities, take any r inequalities to take equal signs. According to permutations and combinations, there are a total of intersection points. These intersection points contain all the vertices of the solution set. Substitute each intersection point into the original N inequalities in turn, and calculate the number of inequalities that hold respectively. The intersection point that makes the number of inequalities that hold the largest is the vertex of the solution set; select all the established inequalities corresponding to the vertex of the solution set. The range of weight values composed of these inequalities is the solution set; the time complexity of the above calculation process is

[0217] In the embodiments of this application, there are a total of r = 6 weights, namely α1, α2, α3, α, β, and γ. Since only the proportional relationship between weights is concerned when calculating the loss function, rather than the absolute magnitude of the weight values, the sum of weights can be set as a fixed constant c. One of the weights is expressed by subtracting other weights from the constant c, so as to reduce the number of weights to be solved. For example, it can be set that α3 = c - α1 - α2, and γ = c - α - β, then two unknowns can be reduced, and the remaining four weights to be solved. Substitute r = 4 into the calculation of the time complexity, and get O(N 4+1 ) = O(N 5 ).

[0218] Furthermore, considering that the longitudinal planning of trajectories ta0 and ta5, and tb0 and tb5 are exactly the same, and their efficiency and comfort index calculation results are the same, the α, β, and γ terms in the inequality can be eliminated, resulting in inequalities only concerning α1, α2, and α3. To further improve computational efficiency, the embodiments of this application can first calculate the driving risk weight coefficients α1, α2, and α3 based on trajectories ta0 and ta5, and tb0 and tb5, and then calculate the weight coefficients α, β, and γ of the evaluation index based on the remaining trajectories. Therefore, the embodiments of this application can ultimately split the entire calculation process into two steps, further reducing the time complexity of each step to O(N). 2+1 )=O(N 3 ).

[0219] As described above, embodiments of this application can use trajectories ta0 and ta5, tb0 and tb5 to calculate a set of inequalities regarding risk weights according to equations (15) and (16). Let the result of rearranging and simplifying any two inequalities be as shown in equation (17):

[0220] a1α1+b1α2+c1α3≥0

[0221] a1α1+b2α2+c2α3≥0 (17)

[0222] α1 + α2 + α3 = 100

[0223] Where α1, α2, and α3 are the weighting coefficients for roadside view risk, driver view risk, and potential collision risk, respectively. i b i and c i , where are known constants, and represent the results of subtracting the three driving risks from the historical trajectory and the virtual trajectory, respectively.

[0224] Furthermore, in the embodiments of this application, the inequality in equation (17) is equalized, and the weight values ​​of the intersection points are obtained as shown in equation (18):

[0225]

[0226]

[0227]

[0228] Substitute the intersection points (α1, α2, α3) into the original system of inequalities, calculate and record the number of true inequalities, and define the proportion of true inequalities to the total number of inequalities as the learning effect of the risk weight. For all N inequalities, select two inequalities in turn and repeat the above steps to obtain the total number of true inequalities. The solution set is the set of inequalities that hold true at the intersection point with the best learning effect.

[0229] This embodiment of the application, based on the driving style classification results, uses data from three drivers with an aggressive driving style to calculate the α1, α2, and α3 weight coefficients for aggressive drivers. This data generates 188 inequalities. After taking the equality sign, 17578 intersection points are obtained. Among all intersection points, a maximum of 135 inequalities are true, corresponding to a learning effect of 0.718. The risk weight learning effect for aggressive drivers is as follows: Figure 15 As shown, Figure 15 The redder the color, the more inequalities the corresponding weight coefficient can make true, meaning the better the learning effect on driver characteristics.

[0230] The embodiments of this application use 24-bit driver data to calculate the weight coefficients for ordinary drivers. The data generates 1370 inequalities with 937765 intersection points. Among all intersection points, a maximum of 989 inequalities are true, corresponding to a learning effect of 0.722. The risk weight learning effect for ordinary drivers is as follows: Figure 16 As shown.

[0231] The embodiments of this application use data from three drivers with a conservative driving style to calculate the weight coefficients for conservative drivers. The data generates 166 inequalities and 13,695 intersection points. Among all intersection points, a maximum of 122 inequalities are true, corresponding to a learning effect of 0.735. The risk weight learning effect for conservative drivers is as follows: Figure 17 As shown.

[0232] Based on the range of values ​​for the system of inequalities that yielded the best learning results, and under the conditions that α1>0, α2>0, and α3>0, the final weight coefficients for the three types of driving risks for aggressive, average, and conservative drivers were determined. Table 4 shows the weight values ​​for the driving risks of the three types of drivers.

[0233] Table 4

[0234]

[0235] Furthermore, embodiments of this application can learn the weights of the aforementioned evaluation indicators, as described in the following process:

[0236] Based on the driving risk weight values ​​in Table 4 and the loss function calculation formula in Equation (14), the safety index C for each trajectory is calculated. risk High efficiency index C speed and comfort index C jerk The specific values ​​are determined using a similar method as described above. Using the other trajectories in Tables 2 and 3 (excluding ta5 and tb5), a set of inequalities concerning the weights of safety, efficiency, and comfort indicators is calculated according to equations (15) and (16). Let any two inequalities, after rearranging and simplifying, be as shown in equation (19):

[0237]

[0238] In the formula, α, β, and γ are the weighting coefficients for safety, efficiency, and comfort indicators, respectively. i b i and c i This is a known constant, which is the result of subtracting the three evaluation indicators from the historical trajectory and the virtual trajectory.

[0239] Similarly, by taking the equality sign of the inequalities in equation (19), we can solve for the weight values ​​of the intersection points. For all N inequalities, we can select two inequalities in turn and repeat the above steps to obtain the total weight values. The solution set is the set of inequalities that hold true at the intersection point with the best learning effect.

[0240] Then, the weight coefficients of the three evaluation indicators for aggressive, average, and conservative drivers were calculated separately, and their learning effects were as follows: Figures 18 to 20 As shown, based on the range of values ​​for the system of inequalities that yields the best learning effect, and under the conditions that α>0, β>0, and γ>0, the final weight coefficients for the three evaluation indicators for aggressive, average, and conservative drivers are determined. Table 5 shows the weight values ​​for the evaluation indicators for the three types of drivers.

[0241] Table 5

[0242]

[0243] Therefore, the embodiments of this application can comprehensively consider driver safety, efficiency and comfort, and learn the characteristic parameters of drivers through their historical driving trajectories, and set different parameter weights for different types of drivers, effectively realizing the adaptive adjustment of driving risk weights and evaluation index weights.

[0244] The adaptive omnidirectional driving assistance method based on a driving risk field proposed in this application collects the driving trajectory of the target vehicle, plans the trajectory according to a preset method to generate multiple candidate trajectories, and filters these candidate trajectories according to a preset screening strategy to obtain a candidate trajectory cluster. Each candidate trajectory in the candidate trajectory cluster is evaluated based on a driving risk assessment model established by the potential energy risk field and a preset loss function to obtain a target trajectory whose evaluation result meets preset requirements. The weight parameters of the loss function corresponding to the target trajectory are learned to perform adaptive learning for omnidirectional driving assistance. Therefore, this application provides a reliable theoretical basis for the design of safety indicators in the loss function, ensures the feasibility of the trajectory through collision detection and curvature detection, and comprehensively considers the safety, efficiency, and comfort of the loss function, setting different parameter weights for different types of drivers, thus realizing adaptive omnidirectional driving assistance based on a driving risk field.

[0245] Next, referring to the accompanying drawings, an adaptive omnidirectional driving assistance device for vehicles based on driving risk fields is described according to an embodiment of this application.

[0246] Figure 21 This is a block diagram of an adaptive omnidirectional driving assistance device for vehicles based on a driving risk field according to an embodiment of this application.

[0247] like Figure 21 As shown, the adaptive omnidirectional driving assistance device 10 based on driving risk field includes: a screening module 100, an evaluation module 200, and a learning module 300.

[0248] The filtering module 100 is used to collect the driving trajectory of the target vehicle, plan the driving trajectory based on a preset method, generate multiple candidate trajectories, and filter the multiple candidate trajectories according to a preset filtering strategy to obtain a candidate trajectory cluster.

[0249] The evaluation module 200 is used to evaluate each candidate trajectory in the candidate trajectory cluster based on the driving risk assessment model established by the potential energy risk field and the preset loss function, so as to obtain the target trajectory whose evaluation results meet the preset requirements.

[0250] The learning module 300 is used to learn the weight parameters of the loss function corresponding to the target trajectory in order to perform adaptive learning for omnidirectional assisted driving of the vehicle.

[0251] Optionally, in one embodiment of this application, the evaluation module 200 includes: a first calculation unit, a second calculation unit, and a modeling unit.

[0252] The first calculation unit is used to obtain the risk equivalence and front wheel angle probability based on the vehicle's kinetic energy and front wheel angle data, and to calculate the roadside view risk based on the risk equivalence and front wheel angle probability.

[0253] The second calculation unit is used to calculate the driver's perspective risk by using the speed of the vehicle itself, the speed of other vehicles, and the maximum risk perception distance in the lateral and longitudinal directions, and to calculate the potential collision risk field using the longitudinal length and lateral width of the target vehicle.

[0254] The modeling unit is used to construct a comprehensive driving risk model based on roadside view risk, driver view risk, and potential collision risk.

[0255] Optionally, in one embodiment of this application, the screening module 100 includes an analysis unit and a fitting unit.

[0256] The analysis unit is used to collect high-speed driving data of the target vehicle, and analyze the lane-changing time and loop speed of different drivers based on the high-speed driving data, and obtain the driving style of the corresponding driver based on the lane-changing time and loop speed of different drivers.

[0257] The fitting unit is used to fit the lateral and longitudinal motion of the vehicle based on a polynomial model to plan the lane-changing trajectory, and to generate multiple candidate trajectories according to the driver's driving style, the planned lane-changing trajectory and preset trajectory constraints.

[0258] Optionally, in one embodiment of this application, the screening module 100 further includes a conversion unit and a detection unit.

[0259] The transformation unit is used to transform multiple candidate trajectories from the Frenet coordinate system to the Cartesian coordinate system.

[0260] The detection unit is used to perform collision detection and curvature detection on multiple candidate trajectories in the Cartesian coordinate system to obtain a cluster of candidate trajectories that meet preset requirements.

[0261] Optionally, in one embodiment of this application, the mathematical expression of the preset loss function is as follows:

[0262] Cosy=αC risk +C speed +C jerk

[0263] Among them, C risk For safety indicators, C speed As an efficiency indicator, C jerk For comfort index, α, β, and γ represent the weighting coefficients of safety index, efficiency index, and comfort index, respectively.

[0264] Optionally, in one embodiment of this application, the learning module 300 includes a generation unit and a third computing unit.

[0265] The generation unit is used to generate multiple historical trajectories and virtual trajectories based on the target trajectory. The historical trajectories include historical straight-ahead trajectories and historical lane-changing trajectories, and the virtual trajectories include virtual straight-ahead trajectories and virtual lane-changing trajectories.

[0266] The third calculation unit is used to calculate the safety, efficiency and comfort indicators of historical and virtual trajectories according to the preset indicator calculation method to obtain a set of weight parameter inequalities, and to learn the driving risk weights based on the set of weight parameter inequalities and the preset learning method.

[0267] It should be noted that the foregoing explanation of the embodiment of the adaptive omnidirectional driving assistance method based on driving risk field also applies to the adaptive omnidirectional driving assistance device based on driving risk field in this embodiment, and will not be repeated here.

[0268] The adaptive omnidirectional driving assistance device based on the driving risk field proposed in this application constructs an omnidirectional risk assessment model based on the driving safety field by considering various risks during vehicle driving, providing a theoretical basis for the design of safety indicators in the loss function; it also proposes a candidate trajectory generation strategy based on a fifth-order polynomial, and ensures the feasibility of the trajectory through collision detection and curvature detection; in addition, it designs a loss function that considers safety, efficiency and comfort, realizing the adaptation of driving risk weights and evaluation indicator weights.

[0269] Figure 22 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include:

[0270] The memory 2201, the processor 2202, and the computer program stored on the memory 2201 and executable on the processor 2202.

[0271] When the processor 2202 executes the program, it implements the adaptive omnidirectional driving assistance method for vehicles based on driving risk field provided in the above embodiments.

[0272] Furthermore, electronic devices also include:

[0273] Communication interface 2203 is used for communication between memory 2201 and processor 2202.

[0274] The memory 2201 is used to store computer programs that can run on the processor 2202.

[0275] The memory 2201 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0276] If the memory 2201, processor 2202, and communication interface 2203 are implemented independently, then the communication interface 2203, memory 2201, and processor 2202 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 divided into address buses, data buses, control buses, etc. For ease of representation, Figure 22 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.

[0277] Optionally, in a specific implementation, if the memory 2201, processor 2202, and communication interface 2203 are integrated on a single chip, then the memory 2201, processor 2202, and communication interface 2203 can communicate with each other through an internal interface.

[0278] The processor 2202 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.

[0279] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described adaptive omnidirectional driving assistance method for vehicles based on driving risk fields.

[0280] 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.

[0281] 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.

[0282] 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 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.

[0283] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium 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-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more 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 devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0284] 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. 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.

[0285] 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-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0286] 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-readable storage medium.

[0287] The storage medium mentioned above can 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. An adaptive omnidirectional driving assistance method for vehicles based on driving risk fields, characterized in that, Includes the following steps: The driving trajectory of the target vehicle is collected, the driving trajectory is planned based on a preset method to generate multiple candidate trajectories, and the multiple candidate trajectories are filtered according to a preset filtering strategy to obtain a candidate trajectory cluster. Based on the driving risk assessment model established by the potential energy risk field and the preset loss function, each candidate trajectory in the candidate trajectory cluster is evaluated to obtain the target trajectory whose evaluation result meets the preset requirements. The weight parameters of the loss function corresponding to the target trajectory are learned to perform adaptive learning for omnidirectional assisted driving of the vehicle. The step of learning the weight parameters of the loss function corresponding to the target trajectory for adaptive learning of omnidirectional assisted driving includes: Multiple historical trajectories and virtual trajectories are generated based on the target trajectory. The historical trajectories include historical straight-through trajectories and historical lane-change trajectories, and the virtual trajectories include virtual straight-through trajectories and virtual lane-change trajectories. The safety, efficiency, and comfort indicators of the historical and virtual trajectories are calculated according to the preset indicator calculation method to obtain a set of weight parameter inequalities. Driving risk weights are then learned based on the set of weight parameter inequalities and the preset learning method. The evaluation of each candidate trajectory in the candidate trajectory cluster based on the driving risk assessment model established by the potential energy risk field and the preset loss function includes: Based on vehicle kinetic energy and front wheel steering angle data, risk equivalence and front wheel steering angle probability are obtained, and roadside view risk is calculated based on the risk equivalence and front wheel steering angle probability. The driver's perspective risk is calculated by taking the vehicle's own speed, the speed of other vehicles, and the maximum risk perception distance in the lateral and longitudinal directions. The potential collision risk field is calculated using the longitudinal length and lateral width of the target vehicle. A comprehensive driving risk model is constructed based on the roadside view risk, driver view risk, and potential collision risk.

2. The method according to claim 1, characterized in that, The process of planning the driving trajectory based on a preset method generates multiple candidate trajectories, including: Collect high-speed driving data of the target vehicle, and analyze the lane change time and loop speed of different drivers based on the high-speed driving data. Based on the lane change time and loop speed of different drivers, obtain the driving style of the corresponding drivers. The vehicle's lateral and longitudinal motions are fitted using a polynomial model to plan the lane-changing trajectory. Based on the driver's driving style, the planned lane-changing trajectory, and preset trajectory constraints, multiple candidate trajectories are generated.

3. The method according to claim 2, characterized in that, The step of filtering the multiple candidate trajectories according to a preset filtering strategy to obtain a candidate trajectory cluster includes: Transform the multiple candidate trajectories from the Frenet coordinate system to the Cartesian coordinate system; Collision detection and curvature detection are performed on the multiple candidate trajectories in the Cartesian coordinate system to obtain the candidate trajectory cluster that meets the preset requirements.

4. The method according to claim 1, characterized in that, The mathematical expression for the preset loss function is as follows: in, For safety indicators, As an efficiency indicator, For comfort indicators, , and These represent the weighting coefficients for the safety, efficiency, and comfort indicators, respectively.

5. An adaptive omnidirectional driving assistance device for vehicles based on driving risk fields, characterized in that, include: The filtering module is used to collect the driving trajectory of the target vehicle, plan the driving trajectory based on a preset method, generate multiple candidate trajectories, and filter the multiple candidate trajectories according to a preset filtering strategy to obtain a candidate trajectory cluster. The evaluation module is used to evaluate each candidate trajectory in the candidate trajectory cluster based on a driving risk assessment model established by a potential energy risk field and a preset loss function, to obtain the target trajectory whose evaluation result meets preset requirements, and The learning module is used to learn the weight parameters of the loss function corresponding to the target trajectory in order to perform adaptive learning for omnidirectional assisted driving of the vehicle. The learning module includes: The generation unit is used to generate multiple historical trajectories and virtual trajectories based on the target trajectory, wherein the historical trajectories include historical straight-through trajectories and historical lane-change trajectories, and the virtual trajectories include virtual straight-through trajectories and virtual lane-change trajectories; The third calculation unit is used to calculate the safety index, efficiency index and comfort index of the historical trajectory and the virtual trajectory respectively according to the preset index calculation method to obtain a set of weight parameter inequalities, and to perform driving risk weight learning based on the set of weight parameter inequalities and the preset learning method. The evaluation module includes: The first calculation unit is used to obtain the risk equivalence and front wheel angle probability based on the vehicle kinetic energy and front wheel angle data, and to calculate the roadside view risk based on the risk equivalence and front wheel angle probability. The second calculation unit is used to calculate the driver's perspective risk by using the speed of the vehicle itself, the speed of other vehicles, and the maximum risk perception distance in the lateral and longitudinal directions, and to calculate the potential collision risk field using the longitudinal length and lateral width of the target vehicle. The modeling unit is used to construct a comprehensive driving risk model based on the roadside view risk, driver view risk, and potential collision risk.

6. The apparatus according to claim 5, characterized in that, The filtering module includes: The analysis unit is used to collect high-speed driving data of the target vehicle, and analyze the lane-changing time and loop speed of different drivers based on the high-speed driving data, and obtain the driving style of the corresponding driver based on the lane-changing time and loop speed of the different drivers. The fitting unit is used to fit the lateral and longitudinal motion of the vehicle based on a polynomial model to plan the lane-changing trajectory, and to generate the multiple candidate trajectories according to the driver's driving style, the planned lane-changing trajectory and preset trajectory constraints.

7. The apparatus according to claim 6, characterized in that, The filtering module also includes: A transformation unit is used to transform the plurality of candidate trajectories from the Frenet coordinate system to the Cartesian coordinate system; The detection unit is used to perform collision detection and curvature detection on the multiple candidate trajectories in the Cartesian coordinate system to obtain the candidate trajectory cluster that meets the preset requirements.

8. The apparatus according to claim 5, characterized in that, The mathematical expression for the preset loss function is as follows: in, For safety indicators, As an efficiency indicator, For comfort indicators, , and These represent the weighting coefficients for the safety, efficiency, and comfort indicators, respectively.

9. An electronic device, characterized in that, include: The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the program to implement the adaptive omnidirectional driving assistance method for vehicles based on driving risk fields as described in any one of claims 1-4.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement the adaptive omnidirectional driving assistance method for vehicles based on driving risk fields as described in any one of claims 1-4.

Images