Autonomous driving takeover risk assessment model considering driving risk field and its construction method

By constructing an autonomous driving takeover risk assessment model that considers driving risk fields, using vehicle performance probability factors and risk distribution functions, combined with Kalman filtering method and genetic algorithm, the problem that the existing technology cannot effectively evaluate the risk of autonomous driving vehicles is solved, and the problem of improving the risk of autonomous driving vehicles has been improved. Risk assessment accuracy and traffic system safety.

CN115620530BActive Publication Date: 2024-11-08HARBIN INST OF TECH
View PDF 0 Cites 4 Cited by

Patent Information

Application Number
CN202211137773.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-09-19
Publication Date
2024-11-08
Estimated Expiration
2042-09-19

AI Technical Summary

Technical Problem

The existing risk assessment model for autonomous driving takeover is mainly aimed at non-autonomous vehicles and traditional follow-up and swing scenes, and cannot effectively evaluate driving risks in different driving environments, resulting in the extended response time of autonomous driving vehicles in emergency situations, increasing traffic conflicts and Accident risk.

Method used

Construct an autonomous driving takeover risk assessment model that considers driving risk fields. By calculating vehicle performance probability factors, constructing a risk distribution function based on distance and motion state, introducing Kalman filtering method and genetic algorithm, calculate driving risk field strength and takeover risk index , provides quantitative assessment and risk assessment model calibration.

Benefits of technology

It improves the accuracy of risk assessment of autonomous driving vehicles in different driving environments, reduces the accident rate and casualty rate, and provides a theoretical basis for the safety evaluation of intelligent transportation systems and the design of autonomous driving vehicle takeover strategy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115620530B_ABST
    Figure CN115620530B_ABST
Patent Text Reader

Abstract

The present invention relates to a method for constructing an automatic driving takeover risk assessment model considering a driving risk field, comprising the following steps: step 1, calculating the vehicle performance probability factor according to the vehicle's response speed and response degree to the risk; step 2, constructing a static risk distribution function based on distance; step 3, constructing a dynamic risk distribution function based on motion state; step 4, considering the influence of takeover reaction time, calculating the driving risk field strength; and extracting data information from the vehicle and the interactive vehicle; step 5, pre-processing the vehicle's speed and acceleration data based on the Kalman filter method; step 6, substituting the calibrated equivalent mass and various undetermined constant coefficients into the automatic driving takeover risk assessment model, that is, obtaining the automatic driving takeover risk assessment model based on the driving risk field. The present invention can improve the safety of the entire road network system and reduce the accident rate and casualty rate caused by automatic driving vehicles.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of traffic safety, and specifically relates to an automatic driving takeover risk assessment model taking into account a driving risk field and a construction method thereof. Background Art

[0002] When a vehicle in an autonomous driving state faces a dangerous emergency road condition that exceeds its capabilities or is in a state of failure or malfunction, the driver needs to handle the emergency and take over the vehicle. Since the driver in the autonomous driving state often performs secondary tasks unrelated to driving, their cognition and reaction to the traffic environment will be reduced to varying degrees, resulting in a longer risk response time for the driver in the autonomous driving state than in manual driving, which can easily lead to traffic conflicts or accidents.

[0003] Existing research on autonomous driving takeover mainly focuses on the influencing indicators and changing rules of takeover performance. The quantitative assessment of takeover risk needs to be further deepened. In addition, the current mainstream risk assessment model is mainly aimed at non-autonomous driving vehicles and traditional lane-changing scenarios, and the applicable driving scenarios are relatively single. Therefore, from the perspective of vehicle dynamic parameters and considering the characteristics of driving risk fields, there is an urgent need for an autonomous driving takeover risk assessment model that can provide methods and technical support for takeover risk assessment and driver assistance system improvement in different driving environments. Summary of the invention

[0004] The present invention aims to solve the above-mentioned technical problems and further provides an automatic driving takeover risk assessment model and a construction method thereof that takes into account the driving risk field.

[0005] The present invention relates to a method for constructing an automatic driving takeover risk assessment model considering a driving risk field, comprising the following steps:

[0006] Step 1: Calculate the vehicle performance probability factor based on the vehicle's response speed and degree to the risk;

[0007] Step 2: Construct a distance-based static risk distribution function;

[0008] Step 3: Construct a dynamic risk distribution function based on motion state;

[0009] Step 4: Considering the impact of takeover reaction time, calculate the driving risk field strength; and extract data information from the vehicle and the interacting vehicle;

[0010] Step 5: Preprocess the vehicle speed and acceleration data based on the Kalman filter method;

[0011] Step 6. Calibrate the equivalent mass of the takeover risk assessment model; calibrate the various undetermined constant coefficients of the takeover risk assessment model; substitute the calibrated equivalent mass and various undetermined constant coefficients into the autonomous driving takeover risk assessment model to obtain the autonomous driving takeover risk assessment model based on the driving risk field.

[0012] In step 1, the calculation formula of the vehicle performance probability factor is as follows:

[0013]

[0014] Where, ζ is the state coefficient of the vehicle. For a normal vehicle, ζ=1, and for a faulty vehicle, 0<ζ<1; C is the amplitude of the vehicle's brake pedal being pressed down during the takeover process (N); C max ——The maximum value of the brake pedal depression amplitude of the vehicle during the takeover process (N); Z——The steering wheel angle of the vehicle during the takeover process (°); Z max ——The maximum steering wheel angle of the vehicle during the takeover process (°); α1, α2——constant coefficients to be determined, α1>0, α2>0.

[0015] In step 2, the calculation formula of the static risk distribution function based on distance is as follows:

[0016]

[0017] Where M i ——equivalent mass of vehicle i (kg); κ——constant coefficient to be determined, κ>0; d ji ——The distance between a point j and the vehicle i in the road plane (m); γ——Decline factor, which can be 1.

[0018] 3. The method for constructing an automatic driving takeover risk assessment model considering driving risk fields according to claim 1 is characterized in that, in step 3, the dynamic risk distribution function calculation formula based on the motion state is as follows:

[0019]

[0020] Where M i ——Equivalent mass of vehicle i (kg); a——Composite deceleration of vehicle i during the takeover process (m / s 2 );δ——composite deceleration reduction factor, δ>0, the value can be 0.15;θ ij ——The angle between the line connecting the vehicle i and other traffic participants j in the driving risk field and the positive direction of the x-axis during the takeover process, θ ij ∈[0°,180°].

[0021] In step 4, the calculation formula of the driving risk field strength is as follows:

[0022]

[0023] Where: S i ——the driving risk field intensity of vehicle i; ——dynamic risk distribution function generated by traffic unit j at ego vehicle i during the takeover process; ——The static risk distribution function generated by traffic unit j at vehicle i during the takeover process; P c ——Vehicle performance probability factor; t i ——the takeover reaction time of vehicle i in the takeover process (s); λ1——the weight of the dynamic risk distribution function; λ2——the weight of the static risk distribution function; Φ——the constant coefficient to be determined;

[0024] The calculation formula of the takeover risk index is as follows:

[0025]

[0026] In the formula, RISK i ——the takeover risk index of vehicle i; ——The average value of the driving risk field intensity;

[0027] Substituting equations (1)-(3) into equation (4) and combining them with equation (5), we get the risk assessment model for autonomous driving takeover considering the driving risk field:

[0028]

[0029] 5. The method for constructing an automatic driving takeover risk assessment model considering driving risk fields according to claim 1 is characterized in that, in step 4, the real-time coordinate position, linear acceleration, driving speed, steering wheel angle, brake pedal opening, and takeover reaction time of the self-vehicle driver are extracted; data information of the self-vehicle and the interacting vehicle, including the type of the interacting vehicle, the real-time position coordinates, and the speed of the vehicle, are extracted.

[0030] In step 5, the specific algorithm flow is:

[0031] (1) Calculate the state prediction value of the Kalman filter algorithm:

[0032]

[0033] In the formula, — observed variables of vehicle speed and acceleration, v r ——Vehicle filtering speed (m / s); a r ——Vehicle filter acceleration (m / s -2); E——coefficient matrix of system prediction state equation; F——input control item matrix; G——control quantity of current state; κ r ξ — process noise in the a priori prediction process; r-1 ——Time rate of change of acceleration (m / s -3 );

[0034] The calculation formula of the coefficient matrix E of the system prediction state equation, the input control item matrix F and the control quantity G of the current state is as shown in the formula:

[0035]

[0036] Where, T is the reciprocal of the data sampling frequency (s).

[0037] (2) Calculate the filter gain matrix

[0038] Assumption ξ r-1 and κ r All satisfy the normal distribution, and the covariances are J and C respectively, then the filter gain matrix is ​​formula (9):

[0039]

[0040] In the formula, ——covariance matrix of prior estimation error (observation value); F——input control term matrix; E——coefficient matrix of system prediction state equation; C——process noise κ in the prior prediction process r The covariance of .

[0041] (3) The covariance matrix of the prior estimation error (observation value) is formula (3-23):

[0042]

[0043] Where, F is the input control matrix; — observed variables of vehicle speed and acceleration, G——Control quantity of the current state; C——Time rate of change of acceleration ξ r-1 The covariance of .

[0044] (4) The covariance matrix of the posterior estimation error (prediction value) is formula (3-23):

[0045]

[0046] Where, K is the unit matrix; W is r ——Filter gain matrix; E——Coefficient matrix of system prediction state equation; ——The covariance matrix of the prior estimation errors (observations).

[0047] In step 6, the equivalent mass is calibrated based on the actual mass, type and speed of the vehicle, and the calculation formula is as follows:

[0048] M i =M(m i ,v i )=m i ·(ρ·v i u +χ) (12)

[0049] In the formula, m i ——actual mass of vehicle i (kg); v i ——The speed of vehicle i (m / s); ρ,u,χ——constant coefficients to be determined.

[0050] In step six, a genetic algorithm is used to calibrate the various undetermined constant coefficients of the takeover risk assessment model. The genetic parameters are configured as follows: the population size and scale are set to 200; the crossover probability is 0.8; the mutation probability is 0.2, and the number of iterations is 300.

[0051] The present invention also relates to a model of a method for constructing an automatic driving takeover risk assessment model taking into account a driving risk field.

[0052] Beneficial Effects

[0053] The present invention takes the quantitative assessment of the risks existing in the takeover process of L3 autonomous driving vehicles as its starting point, takes the idea of ​​driving risk field as its theoretical basis, establishes a static risk distribution function based on distance and a dynamic risk distribution function based on motion state, introduces vehicle performance probability factors and considers the influence of takeover reaction time, thus constructing an autonomous driving takeover risk assessment model considering the driving risk field.

[0054] The present invention can not only provide a theoretical basis and reference for the takeover risk research and safety evaluation of autonomous driving vehicles under intelligent transportation systems, but also provide new research ideas for the design of takeover strategies for autonomous driving vehicles, thereby improving the safety of the entire road network system and reducing the accident rate and casualty rate caused by autonomous driving vehicles. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 The present invention is a flow chart of a method for constructing an automatic driving takeover risk assessment model that takes into account a driving risk field. DETAILED DESCRIPTION

[0056] The following combination Figure 1 This embodiment will be described in detail.

[0057] The present invention provides a method for constructing an automatic driving takeover risk assessment model considering a driving risk field, comprising the following steps:

[0058] Step 1: Consider the vehicle's response speed and degree to risk and introduce vehicle performance probability factors;

[0059] During the process of taking over the driver, there are two main ways for the vehicle to avoid risks: braking and steering. The vehicle's brake pedal depression amplitude and steering wheel angle are used to measure the vehicle's response speed and degree to risks during the takeover process. The smaller the brake pedal depression amplitude, the lower the driver's control over the vehicle, and the more likely it is to have a collision risk; the larger the vehicle's steering wheel angle, the greater the driver's chance of successfully bypassing obstacles when taking over. The calculation formula for the vehicle performance probability factor is as follows:

[0060]

[0061] Where, ζ is the state coefficient of the vehicle. For a normal vehicle, ζ=1, and for a faulty vehicle, 0<ζ<1; C is the amplitude of the vehicle's brake pedal being pressed down during the takeover process (N); C max ——The maximum value of the brake pedal depression amplitude of the vehicle during the takeover process (N); Z——The steering wheel angle of the vehicle during the takeover process (°); Z max ——The maximum steering wheel angle of the vehicle during the takeover process (°); α1, α2——constant coefficients to be determined, α1>0, α2>0.

[0062] Step 2: Considering the distance between the vehicle and other traffic units during the takeover process, the changing trend of the distance, and the equivalent mass of the vehicle, a distance-based static risk distribution function is constructed;

[0063] Due to the limitation of the driver's perception ability, another factor affecting the takeover risk is the distance between the traffic unit and other traffic units during the takeover process and the changing trend of the distance. In road traffic, the greater the distance from the source of the driving risk field, the smaller the probability of a collision accident, that is, the weaker the driving risk field generated by the field source. In the driving risk field, the calculation formula of the static risk distribution function based on distance is as follows:

[0064]

[0065] Where M i ——equivalent mass of vehicle i (kg); κ——constant coefficient to be determined, κ>0; d ji ——The distance between a point j and the vehicle i in the road plane (m); γ——Decline factor, which can be 1.

[0066] Step 3: Considering the motion state of the vehicle and other surrounding traffic participants during the takeover process and the equivalent mass of the vehicle, a dynamic risk distribution function based on the motion state is constructed;

[0067] The greater the speed of the vehicle, the greater the field strength generated by the traffic units in motion around it. When the vehicle encounters an obstacle and decelerates, the smaller the deceleration value, the slower the speed change, the worse the braking effect, and the higher the probability of a collision accident. In addition, the smaller the angle between the direction vector of the vehicle pointing to the surrounding traffic participants and the positive direction of the x-axis (defined as the road direction and consistent with the direction of the vehicle's advance), the greater the driving risk. The dynamic risk distribution function calculation formula of the traffic unit based on the motion state during the takeover process is as follows:

[0068]

[0069] Where M i ——Equivalent mass of vehicle i (kg); a——Composite deceleration of vehicle i during the takeover process (m / s 2 );δ——composite deceleration reduction factor, δ>0, the value can be 0.15;θ ij ——The angle between the line connecting the vehicle i and other traffic participants j in the driving risk field and the positive direction of the x-axis during the takeover process, θ ij ∈[0°,180°].

[0070] Step 4: Introduce the vehicle performance probability factor to represent the probability of potential traffic accidents caused by abnormal vehicle trajectory during the takeover process. Use the risk distribution function to describe the distribution characteristics of the potential risk of the vehicle in the traffic scene during the driving takeover process. Considering the influence of the takeover reaction time, the calculation formula of the driving risk field strength is as follows:

[0071]

[0072] Where: S i ——the driving risk field intensity of vehicle i; ——dynamic risk distribution function generated by traffic unit j at ego vehicle i during the takeover process; ——The static risk distribution function generated by traffic unit j at vehicle i during the takeover process; P c ——Vehicle performance probability factor; t i ——The takeover reaction time of vehicle i during the takeover process (s); λ1——The weight of the dynamic risk distribution function; λ2——The weight of the static risk distribution function; Φ——Undetermined constant coefficient.

[0073] The driving risk field strength is an absolute indicator with a large range of numerical variations, making it difficult to effectively assess the risk during the takeover process. Therefore, it is necessary to process this indicator and define its relative indicator as the "takeover risk index" to assess the risk during the takeover process. The larger the takeover risk index, the more likely the vehicle is to incur risks during the takeover process. The calculation formula for the takeover risk index is as follows:

[0074]

[0075] In the formula, RISK i ——the takeover risk index of vehicle i;

[0076] ——The average value of the driving risk field intensity.

[0077] Substituting equations (1)-(3) into equation (4) and combining them with equation (5), we get the risk assessment model for autonomous driving takeover considering the driving risk field:

[0078]

[0079] Step 5: Extract data information from the vehicle and the interacting vehicle;

[0080] Extract the real-time coordinate position, linear acceleration (x / y / z direction), driving speed, steering wheel angle, brake pedal opening, and driver's takeover reaction time of the self-vehicle, so as to calculate the subsequent synthetic deceleration, collision time countdown, and driving risk field strength;

[0081] The type of interacting vehicle, real-time location coordinates, and vehicle speed are extracted for subsequent calculation of dynamic and static risk distribution functions.

[0082] Step 6: Preprocess the vehicle speed and acceleration data based on the Kalman filter method;

[0083] The Kalman filter algorithm is used to process the speed and acceleration of the vehicle in a timely manner. The Kalman filter is an algorithm based on the state equation. Its principle is to consider the distribution of data at different time points and fuse the data. The standard Kalman filter is divided into two aspects: one is a priori prediction, and the other is prediction correction. The specific algorithm process is:

[0084] (1) Calculate the state prediction value of the Kalman filter algorithm:

[0085]

[0086] In the formula, — observed variables of vehicle speed and acceleration, v r ——Vehicle filtering speed (m / s); a r ——Vehicle filter acceleration (m / s -2 ); E——coefficient matrix of system prediction state equation; F——input control item matrix; G——control quantity of current state; κ r ξ — process noise in the a priori prediction process; r-1——Time rate of change of acceleration (m / s -3 ).

[0087] Among them, the calculation formula of the coefficient matrix E of the system prediction state equation, the input control item matrix F and the control quantity G of the current state is shown in formula (8):

[0088]

[0089] Where, T is the reciprocal of the data sampling frequency (s).

[0090] (2) Calculate the filter gain matrix

[0091] Assumption ξ r-1 and κ r All satisfy the normal distribution, and the covariances are J and C respectively, then the filter gain matrix is ​​formula (3-21):

[0092]

[0093] In the formula, ——covariance matrix of prior estimation error (observation value); F——input control term matrix; E——coefficient matrix of system prediction state equation; C——process noise κ in the prior prediction process r The covariance of .

[0094] (3) The covariance matrix of the prior estimation error (observation value) is formula (3-23):

[0095]

[0096] Where, F is the input control matrix; — observed variables of vehicle speed and acceleration, G——Control quantity of the current state; C——Time rate of change of acceleration ξ r-1 The covariance of .

[0097] (4) The covariance matrix of the posterior estimation error (prediction value) is formula (3-23):

[0098]

[0099] Where, K is the unit matrix; W is r ——Filter gain matrix; E——Coefficient matrix of system prediction state equation; ——The covariance matrix of the prior estimation errors (observations).

[0100] Step 7: Calibrate the equivalent quality of the takeover risk assessment model;

[0101] The equivalent mass represents the risk that the vehicle's properties pose to itself. The greater the equivalent mass, the greater the severity of the collision between the vehicle and other traffic participants. The equivalent mass is not only related to the actual mass of the vehicle, but also to its type and speed. Therefore, the equivalent mass is calibrated based on the actual mass, type and speed of the vehicle, and the calculation formula is as follows:

[0102] M i =M(m i ,v i )=m i ·(ρ·v i u +χ) (12)

[0103] In the formula, m i ——actual mass of vehicle i (kg); v i ——The speed of vehicle i (m / s); ρ,u,χ——constant coefficients to be determined.

[0104] Step 8: Calibrate various undetermined constant coefficients of the takeover risk assessment model;

[0105] The genetic algorithm is used to calibrate the various undetermined constant coefficients of the takeover risk assessment model. The genetic parameters are configured as follows: the population size and scale are set to 200; the crossover probability is 0.8; the mutation probability is 0.2, and the number of iterations is 300.

[0106] Step 9: Substitute the calibrated equivalent mass and various undetermined constant coefficients into formula (6) to obtain the autonomous driving takeover risk assessment model based on the driving risk field.

[0107] Example

[0108] The root mean square error is selected as the evaluation index, and the effectiveness of the autonomous driving takeover risk assessment model considering the driving risk field is verified by comparing the inverse of the collision time and the Risk value of the constructed model.

[0109] The inverse of collision time is a commonly used indicator for evaluating vehicle driving risk, and its calculation formula is as follows:

[0110]

[0111] In the formula, TD i ——The inverse of the collision time of vehicle i (s -1 );v i ——The speed of vehicle i during braking (m / s); v j ——During the braking process, the speed of the front vehicle j (m / s, if there is a stationary obstacle in front, then v j =0); d ij——The distance between vehicle i and vehicle j in front (m). Use driving simulation method to collect relevant data for validity verification.

[0112] The simulation scenario is built based on SCANeR studio software. The road type is a straight section of an urban expressway with a total length of 7.5 km and 6 lanes in both directions. The weather and road conditions are good. The road speed limit is 90 km / h, the traffic flow is set to a steady flow (15 pcu / km·ln) and is evenly distributed, and the initial speed is set to 80 km / h. At the beginning of the experiment, the vehicle is in automatic driving mode and is located in the middle lane. The takeover request time is set to 5s, and the takeover request method uses a combination of vision and hearing. 15 drivers were recruited as subjects, and a total of 15 groups of takeover data of the test drivers were obtained.

[0113] The collected velocity and acceleration data are filtered. Compared with the actual trajectory before filtering, the velocity and acceleration trajectory after filtering becomes smoother and more uniform, and has a higher degree of fit with the actual trajectory.

[0114] Based on the above data, the constant coefficients to be determined in the model are calibrated, and the calibration results are shown in Table 1.

[0115] Table 1 Calibration results of undetermined constant coefficients in the takeover risk assessment model

[0116]

[0117] The TD and RISK values ​​and their means of the 15 subjects during the takeover process were calculated, the means were normalized, and their root mean square errors were calculated. The mean root mean square error of RISK was 0.059, and the mean root mean square error of TD was 0.093, indicating that RISK data is better than TD data in terms of the accuracy of risk characterization, with a smaller error with the actual risk, more stable data fluctuations, and higher credibility. Therefore, the takeover risk assessment model is reliable in risk assessment and can effectively assess the takeover risk of the driver.

[0118] The above contents of the present invention are only preferred embodiments of the present invention and are not intended to limit the implementation scheme of the present invention. A person skilled in the art can easily make corresponding changes or modifications based on the main concept and spirit of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope required by the claims.

Claims

1. A method for constructing an automatic driving takeover risk assessment model considering driving risk fields, characterized in that: The steps include: Step 1: Calculate the vehicle performance probability factor based on the vehicle's response speed and degree to the risk; Step 2: Construct a distance-based static risk distribution function; Step 3: Construct a dynamic risk distribution function based on motion state; Step 4: Considering the impact of takeover reaction time, calculate the driving risk field strength; and extract data information from the vehicle and the interacting vehicle; Step 5: Preprocess the vehicle speed and acceleration data based on the Kalman filter method; Step 6: calibrate the equivalent quality of the takeover risk assessment model; calibrate various undetermined constant coefficients of the takeover risk assessment model; Substitute the calibrated equivalent mass and various undetermined constant coefficients into the autonomous driving takeover risk assessment model, and obtain the autonomous driving takeover risk assessment model based on the driving risk field; In step 1, the calculation formula of the vehicle performance probability factor is as follows: Where, ζ is the state coefficient of the vehicle. For a normal vehicle, ζ=1, and for a faulty vehicle, 0<ζ<1; C is the amplitude of the vehicle's brake pedal being pressed down during the takeover process (N); C max ——The maximum value of the brake pedal depression amplitude of the vehicle during the takeover process (N); Z——The steering wheel angle of the vehicle during the takeover process (°); Z max ——The maximum value of the vehicle's steering wheel angle during the takeover process (°); α1, α2——Undetermined constant coefficients, α1>0, α2>0; In step 2, the calculation formula of the static risk distribution function based on distance is as follows: Where M i ——equivalent mass of vehicle i (kg); κ——undetermined constant coefficient, κ>0; d ji ——the distance between a certain point j and the vehicle i in the road plane (m); γ——the descent speed factor, which can be 1; In step 3, the calculation formula of the dynamic risk distribution function based on the motion state is as follows: Where M i ——Equivalent mass of vehicle i (kg); a——Composite deceleration of vehicle i during the takeover process (m / s 2 );δ——composite deceleration reduction factor, δ>0, the value can be 0.15;θ ij ——The angle between the line connecting the vehicle i and other traffic participants j in the driving risk field and the positive direction of the x-axis during the takeover process, θ ij ∈[0°,180°]; In step 4, the calculation formula of the driving risk field strength is as follows: Where: S i ——the driving risk field intensity of vehicle i; ——dynamic risk distribution function generated by traffic unit j at ego vehicle i during the takeover process; ——The static risk distribution function generated by traffic unit j at vehicle i during the takeover process; P c ——Vehicle performance probability factor; t i ——the takeover reaction time of vehicle i during the takeover process (s); λ1——the weight of the dynamic risk distribution function; λ2——weight of static risk distribution function; Φ——undetermined constant coefficient; The calculation formula of the takeover risk index is as follows: In the formula, RISK i ——the takeover risk index of vehicle i; ——The average value of the driving risk field intensity; Substituting equations (1)-(3) into equation (4) and combining them with equation (5), we get the risk assessment model for autonomous driving takeover considering the driving risk field:

2. The method for constructing an automatic driving takeover risk assessment model considering driving risk fields according to claim 1, characterized in that: In step 4, the real-time coordinate position, linear acceleration, driving speed, steering wheel angle, brake pedal opening, and takeover reaction time of the driver of the self-vehicle are extracted; data information of the self-vehicle and the interacting vehicle, including the type of the interacting vehicle, the real-time position coordinates, and the speed of the vehicle.

3. The method for constructing an automatic driving takeover risk assessment model considering driving risk fields according to claim 1, characterized in that: In step 5, the specific algorithm flow is: (1) Calculate the state prediction value of the Kalman filter algorithm: In the formula, — observed variables of vehicle speed and acceleration, v r ——Vehicle filtering speed (m / s); a r ——Vehicle filter acceleration (m / s -2 ); E——coefficient matrix of system prediction state equation; F——input control item matrix; G——control quantity of current state; κ r ξ — process noise in the a priori prediction process; r-1 ——Time rate of change of acceleration (m / s -3 ); The calculation formula of the coefficient matrix E of the system prediction state equation, the input control item matrix F and the control quantity G of the current state is shown in formula (8): Where, T is the reciprocal of the data sampling frequency (s); (2) Calculate the filter gain matrix Assumption ξ r-1 and κ r All satisfy the normal distribution, and the covariances are J and C respectively, then the filter gain matrix is ​​formula (9): In the formula, ——covariance matrix of prior estimation error (observation value); F——input control term matrix; E——coefficient matrix of system prediction state equation; C——process noise κ in the prior prediction process r The covariance of (3) The covariance matrix of the prior estimation error (observation value) is formula (3-23): Where, F is the input control matrix; — observed variables of vehicle speed and acceleration, G——Control quantity of the current state; J——Time rate of change of acceleration ξ r-1 The covariance of (4) The covariance matrix of the posterior estimation error (prediction value) is formula (3-23): Where, K is the unit matrix; W is r ——Filter gain matrix; E——Coefficient matrix of system prediction state equation; ——The covariance matrix of the prior estimation errors (observations).

4. The method for constructing an automatic driving takeover risk assessment model considering driving risk fields according to claim 1, characterized in that: In step 6, the equivalent mass is calibrated based on the actual mass, type and speed of the vehicle, and the calculation formula is as follows: Where M i ——Equivalent mass of vehicle i (kg); m i ——actual mass of vehicle i (kg); v i ——The speed of vehicle i (m / s); ρ,u,χ——constant coefficients to be determined.

5. The method for constructing an automatic driving takeover risk assessment model considering driving risk fields according to claim 1, characterized in that: In step six, a genetic algorithm is used to calibrate the various undetermined constant coefficients of the takeover risk assessment model. The genetic parameters are configured as follows: the population size and scale are set to 200; the crossover probability is 0.8; the mutation probability is 0.2, and the number of iterations is 300.

6. A model for the method of constructing an automatic driving takeover risk assessment model taking into account driving risk fields as described in any one of claims 1 to 5.

Citation Information

Cited By

  • Multi-dimensional risk field enabled dynamic driving risk early warning and risk avoiding method and system

    CN120340308A

  • Dynamic driving risk warning and risk avoidance method and system enabled by multi-dimensional risk field

    CN120340308B

  • Driving risk assessment method for automatic driving vehicle in expressway interleaving area

    CN120472415A

  • A driving risk assessment method for autonomous vehicles in weaving areas on expressways

    CN120472415B