A scene matrix-based automatic driving vehicle decision control method

Abstract

The application relates to a kind of automatic driving car decision control methods based on scene matrix, comprising: obtaining road parameter, self-vehicle position information, background car's position and behavior information, constructs scene matrix, for depicting the relative position of self-vehicle and background car under dynamic change in selected scene;According to scene matrix, respectively calculate decision safety matrix, decision efficiency matrix and behavior preference matrix;Calculate comprehensive decision matrix, complete self-vehicle decision through the mapping of behavior by comprehensive decision matrix, to control the driving action of vehicle accordingly.Compared with prior art, the application simplifies real road scene through scene matrix, filters redundant miscellaneous information, reduces the complexity of calculation;At the same time, safety, efficiency and driver behavior preference and other factors are considered comprehensively, which can realize horizontal and vertical comprehensive decision, ensure the real-time and diversity of decision.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving technology, and in particular to a decision-making control method for an autonomous driving vehicle based on a scenario matrix. Background Art

[0002] Autonomous driving is a technology that enables vehicles to drive autonomously without human intervention. It uses various sensors, computer vision, artificial intelligence, and machine learning technologies to perceive, analyze, and make decisions about the road environment, enabling autonomous navigation and control of the vehicle.

[0003] The development of autonomous driving technology aims to improve traffic safety, reduce traffic accidents, and provide more efficient traffic flow. It can eliminate traffic accidents caused by factors such as human error, fatigue, and distraction. Furthermore, autonomous driving can improve road utilization, reduce congestion, and lower energy consumption and emissions. Autonomous driving is undoubtedly a key direction for future automotive development. During operation, autonomous vehicles must continuously and in real time make reasonable decisions and plans based on dynamic changes in the external environment to complete driving tasks safely and efficiently.

[0004] Decision-making and planning in autonomous driving can be divided into upper-level decision-making and lower-level planning. Upper-level decision-making refers to the behavioral choices made by the autonomous vehicle based on the current operating environment. Lower-level planning involves the calculation of the vehicle's movement path, speed trajectory, and other information based on the upper-level decision-making results. Upper-level decision-making plays a leading role in the entire decision-making and planning process, significantly impacting the safety and efficiency of autonomous vehicles.

[0005] For the upper-level decision-making of autonomous vehicles, most existing methods adopt a horizontal and vertical decoupling approach, and require the construction of a specific judgment method for each decision behavior, lacking a unified and comprehensive decision-making method. At the same time, existing decision-making methods start from real-world scene information and often consider a large amount of specific details, which increases the computational complexity of the method and lacks real-time performance. They also rarely incorporate driver behavioral preferences or driving styles into the decision-making process, failing to reflect the diversity of vehicle decisions in real traffic environments. Summary of the Invention

[0006] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and provide an autonomous driving vehicle decision-making control method based on a scenario matrix, which can realize horizontal and vertical integrated decision-making and ensure the real-time and diversity of decision-making.

[0007] The purpose of the present invention can be achieved by the following technical solution: A decision-making control method for an autonomous driving vehicle based on a scene matrix, comprising the following steps:

[0008] S1. Obtain road parameters, vehicle location information, and background vehicle location and behavior information;

[0009] S2. Based on the data information obtained in step S1, a scene matrix is ​​constructed to describe the relative positions of the ego vehicle and background vehicles under dynamic changes in the selected scene;

[0010] S3. Calculate the decision safety matrix, decision efficiency matrix, and behavior preference matrix based on the scenario matrix.

[0011] S4. Calculate the comprehensive decision matrix and complete the vehicle's decision by mapping the comprehensive decision matrix to the behavior to control the vehicle's driving action accordingly.

[0012] Furthermore, the road parameters in step S1 include the number of lanes and the road length;

[0013] The vehicle position information includes the lane in which the vehicle is located;

[0014] The position information of the background vehicle includes the lane where the background vehicle is located and the longitudinal distance Δx of the background vehicle relative to the vehicle. i , where i represents the serial number of the background vehicle and specifies the longitudinal distance Δx in front of the vehicle i is positive, and vice versa is negative;

[0015] The behavior information of the background vehicle includes five behaviors: acceleration, deceleration, lane change left, lane change right, and remaining unchanged.

[0016] Furthermore, the process of constructing the scene matrix in step S2 is as follows:

[0017] S21, determine the direction and size of the scene matrix;

[0018] S22, determining the positions of the ego vehicle and background vehicles in the scene matrix;

[0019] S23. Determine the corresponding relationship between the vehicle behavior and the position change of the vehicle in the scene matrix.

[0020] Furthermore, in step S21, the row vector direction of the scene matrix M is horizontal, the column vector direction is vertical, and the number of rows of the scene matrix M is M. row and the number of columns M col The calculation formula is:

[0021]

[0022] M col =N Lane

[0023] L unit =v Ego ×Δt

[0024] wherein R f , R b are adjustable parameters, respectively representing the front and rear determination ranges of the ego vehicle; N Lane is the number of lanes; ceil(·) is the ceiling function; L unit is the actual distance represented by each row of the matrix, v Ego is the speed of the ego vehicle; and Δt is the unit time length.

[0025] Further, the column coordinates C ego,t , C i,t of the ego vehicle and the background vehicle i at time t in the step S22 are represented by:

[0026]

[0027] wherein C ego,t , C i,t are the column coordinates of the ego vehicle and the background vehicle i at time t, respectively, and R ego,t , R i,t are the row coordinates of the ego vehicle and the background vehicle i at time t, respectively.

[0028] Further, the position of the vehicle after performing the behavior b in the step S23 is represented by:

[0029]

[0030] wherein R t is the row position of the vehicle in the scene matrix at time t, and C t is the column position of the vehicle in the scene matrix at time t.

[0031] Further, the decision safety matrix in the step S3 is represented by:

[0032]

[0033] wherein S b represents the safety score corresponding to the behavior b after the ego vehicle performs the behavior b.

[0034] b∈{acc, dec, left, right, stay}, respectively corresponding to the five vehicle behaviors of acceleration, deceleration, left lane change, right lane change, and staying unchanged.

[0035]

[0036] wherein w jk corresponds to each element in the safety weight matrix W safety ; and m f(j)g(k)is the corresponding element in the scene matrix M; f(j), g(k) represent the corresponding relationship between the multiplied elements in the two matrices; State(·) is the matrix algebra function;

[0037]

[0038] Among them, w is the weight coefficient of the lane area where the vehicle is located.

[0039] Furthermore, the decision efficiency matrix in step S3 is expressed as:

[0040]

[0041] Among them, the behavioral efficiency score E in the efficiency matrix b It represents the efficiency score of the behavior after the vehicle completes behavior b, b∈{acc,dec,left,right,stay}, which corresponds to the five vehicle behaviors of acceleration, deceleration, left lane change, right lane change and staying the same.

[0042] When b∈{acc,dec}, E b Calculated by the following formula:

[0043]

[0044] When b∈{left,right,stay}, E b Calculated by the following formula:

[0045]

[0046] Where m and n are the speed change efficiency factor and lane change efficiency factor, respectively; invalid behavior means that the vehicle collides or exceeds the lane after executing behavior b.

[0047] Furthermore, the behavior preference matrix in step S3 is used to describe the decision preferences of the vehicle for the five decision behaviors, and to characterize the preferences of different driving styles for different decisions. The behavior preference matrix D tendency Expressed as:

[0048]

[0049] Among them, T b represents the tendency of the ego vehicle to complete behavior b, b∈{acc,dec,left,right,stay}, which corresponds to the five vehicle behaviors of acceleration, deceleration, left lane change, right lane change and staying the same.

[0050] Furthermore, the calculation formula of the comprehensive decision matrix in step S4 is:

[0051] D decision =Dsafety +D efficiency +D tendency

[0052] The corresponding behavior with the highest score in the comprehensive decision matrix is ​​the behavioral decision made by the vehicle in response to the behavior of other background vehicles.

[0053] Compared with the prior art, the present invention has the following advantages:

[0054] This method constructs a scenario matrix by acquiring road parameters, the vehicle's position, and the positions and behaviors of background vehicles. This matrix depicts the dynamic relative positions of the vehicle and background vehicles in a selected scenario. Based on the scenario matrix, a comprehensive decision matrix is ​​then calculated. Finally, by mapping the comprehensive decision matrix to the behaviors, the vehicle's decision-making is completed and the driving action is controlled accordingly. This scenario matrix simplifies the real-world road scene, filters out redundant information, and reduces computational complexity. It also comprehensively considers factors such as safety, efficiency, and driver preferences, enabling the autonomous vehicle to make rapid and appropriate decisions regarding the behavior of other background vehicles, demonstrating the diversity of vehicle decision-making in real-world traffic environments.

[0055] In the scenario matrix constructed by the present invention, the row vector direction is horizontal and the column vector direction is vertical, which can consider horizontal and vertical decision-making behaviors under the same behavior value judgment standard, thereby making horizontal and vertical comprehensive decisions.

[0056] After obtaining road parameters, vehicle position information, and background vehicle position and behavior information, the present invention converts real-scene information into a simplified mathematical matrix model, filtering out redundant information that is not helpful for decision-making, thereby reducing computational complexity and ensuring real-time decision-making.

[0057] The present invention calculates a comprehensive decision matrix including a decision safety matrix, a decision efficiency matrix and a behavior preference matrix. When the behavior of other traffic participants in the outside world changes, the decision of the vehicle itself takes into account both safety and the impact of the decision on efficiency. At the same time, the driver's behavior preferences are also integrated into the decision-making process, which can effectively realize diverse and comprehensive decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 Schematic diagram of the method flow of the present invention;

[0059] Figure 2 Schematic diagram of an actual road scene in the embodiment;

[0060] Figure 3 Schematic diagram of the construction process of the scene matrix in the embodiment. DETAILED DESCRIPTION

[0061] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0062] Embodiment

[0063] As Figure 1 shown, a scene matrix-based automatic driving vehicle decision control method comprises the following steps:

[0064] S1, obtaining road parameters, self-vehicle position information, and background vehicle position and behavior information;

[0065] S2, constructing a scene matrix based on the data information obtained in step S1, for depicting the relative positions of the self-vehicle and the background vehicles under dynamic changes in the selected scene;

[0066] S3, calculating a decision safety matrix, a decision efficiency matrix, and a behavior preference matrix respectively according to the scene matrix;

[0067] S4, calculating a comprehensive decision matrix, and completing self-vehicle decision through the mapping of the behavior by the comprehensive decision matrix, to control the driving actions of the vehicle accordingly.

[0068] In step S1, the road parameters include the number of lanes and the length of the road; the self-vehicle position information includes the lane where the self-vehicle is located; and the background vehicle position information includes the lane where the background vehicle is located and the longitudinal distance Δx i of the background vehicle relative to the self-vehicle, where i represents the serial number of the background vehicle, and it is stipulated that the longitudinal distance Δx i of the background vehicle relative to the self-vehicle is positive in front of the self-vehicle and negative behind the self-vehicle; and the behavior information of the background vehicle includes five possible behavior actions of acceleration, deceleration, left lane change, right lane change, and keeping unchanged.

[0069] In this embodiment, a scene as shown in Figure 2 is set, the number of lanes N Lane = 3; the speed of the self-vehicle v Ego = 20 m / s; the unit time length Δt = 1 s; the front and rear judgment ranges of the self-vehicle Ego are R f = 80 m and R b = 20 m respectively; the serial number of the lane where the self-vehicle Ego is located is 2; the longitudinal relative distances of two background vehicles BV1 and BV2 relative to the self-vehicle are Δx1 = 40 m and Δx2 = 20 m respectively, and the serial numbers of the lanes are 1 and 3 respectively; and the vehicle behaviors are right lane change and deceleration respectively.

[0070] The construction process of the scene matrix in step S2 is as follows:

[0071] S21, determining the direction and size of the scene matrix;

[0072] S22, determining the positions of the self-vehicle and the background vehicles in the scene matrix;

[0073] S23, determining the corresponding relationship between the vehicle behavior and the position change of the vehicle in the scene matrix.

[0074] In step S21, the row vector direction of the scene matrix M is horizontal, the column vector direction is vertical, and the number of rows in the matrix M is row and the number of columns M col The calculation method is:

[0075]

[0076] M col =N Lane

[0077] Among them, R f , R b are adjustable parameters, representing the range of determination in front of the vehicle and the range of determination behind the vehicle respectively; N Lane is the number of lanes; ceil(·) is the rounding function; L unit The actual distance represented by each row of the matrix is ​​determined by the following formula:

[0078] L unit =v Ego ×Δt

[0079] Among them, v Ego is the vehicle speed; Δt is the unit time.

[0080] Therefore, in this embodiment:

[0081] L unit =v Ego ×Δt=20×1=20m

[0082]

[0083] M col =N Lane =3

[0084] The column coordinates C of the ego vehicle and background vehicle i at time t in step S22 ego,t 、C i,t is the lane number of the vehicle, and the row coordinates are expressed as:

[0085]

[0086] In the embodiment, the vehicle position coordinates are calculated:

[0087]

[0088] C ego,t =2

[0089] Therefore, the vehicle's position coordinates are (4,2).

[0090] Calculate the position coordinates of the background vehicle BV1 as:

[0091]

[0092] C 1,t =1

[0093] Therefore, the coordinates of the background vehicle BV1 are (2,1).

[0094] Calculate the position coordinates of the background vehicle BV2 as:

[0095]

[0096] C 2,t =3

[0097] The coordinates of the background vehicle BV2 are (3,3).

[0098] The position of the vehicle after executing action b in step S23 is expressed as:

[0099]

[0100] Therefore, the positions of the background vehicles BV1 and BV2 after completing the behavior are (2,2) and (4,3) respectively, and the position of the ego vehicle Ego is (4,2). The construction of the scene matrix is ​​as follows: Figure 3 shown.

[0101] The decision safety matrix in step S3 is expressed as:

[0102]

[0103] Among them, S b It represents the safety score corresponding to the behavior b after the vehicle completes the behavior b. Behavior b∈{acc,dec,left,right,stay} corresponds to the five vehicle behaviors of acceleration, deceleration, left lane change, right lane change and maintaining the original state.

[0104] The safety score of the decision safety matrix in step S3 is calculated as follows:

[0105]

[0106] Among them, w jk Corresponding security weight matrix W safety Each element in m f(j)g(k) is the corresponding element in the scene matrix M; f(j) and g(k) represent the corresponding relationship between the multiplied elements in the two matrices; State(·) is the matrix algebra function. The corresponding calculation methods of the above parameters are as follows:

[0107]

[0108] Among them, w is the weight coefficient of the lane area where the vehicle is located;

[0109]

[0110] In this embodiment, the weight coefficient w of the lane area where the vehicle is located is set to 0.7, and the scene matrix after algebraic transformation is:

[0111]

[0112] (1) When the ego vehicle performs the maintain-status behavior, it is determined that the ego vehicle is in a non-boundary lane, and the safety weight matrix is:

[0113]

[0114] at this time, Then we can further calculate:

[0115]

[0116] Finally, we can calculate:

[0117]

[0118] Where ⊙ represents the multiplication of corresponding elements of two matrices.

[0119] (2) When the ego vehicle is accelerating or decelerating, it is determined that the ego vehicle is in a non-boundary lane, and the safety weight matrix is:

[0120]

[0121] Using the same calculation process, we can get S acc =1.15, S dec =1.85.

[0122] (3) When the ego vehicle performs a left lane change, it is determined that the ego vehicle will be in the left boundary lane, and the safety weight matrix is:

[0123]

[0124] Using the same calculation process, calculate S left =2.0.

[0125] (4) When the ego vehicle changes lanes right, it will collide with BV2, so S right =-∞.

[0126] In summary, the final decision safety matrix is

[0127]

[0128] The decision efficiency matrix in step S3 is expressed as:

[0129]

[0130] Among them, the behavioral efficiency score E in the efficiency matrix b It represents the efficiency score of the behavior after the vehicle completes behavior b, b∈{acc,dec,left,right,stay}, which corresponds to the five vehicle behaviors of acceleration, deceleration, left lane change, right lane change, and staying the same. The specific calculation method of efficiency score is as follows:

[0131] When b∈{acc,dec}, E b Calculated by the following formula:

[0132]

[0133] When b∈{left,right,stay}, E b Calculated by the following formula:

[0134]

[0135] Where m and n are the speed change efficiency factor and lane change efficiency factor, respectively; invalid behavior means that the vehicle collides or exceeds the lane after executing behavior b.

[0136] In this embodiment, the speed change efficiency factor m is set to 1, and the lane change efficiency factor n is set to 0.5. Therefore, when the vehicle accelerates or decelerates, E acc =1, E dec = -1; When the vehicle maintains its original state, the state of whether there is a car in front of the behavior remains unchanged, so E stay = 0; When the vehicle changes lanes left, since there is no car ahead, E left = 0.5; When the ego vehicle changes lanes right, the action is invalid because it collides with BV2, so E right = 0. The final decision efficiency matrix is:

[0137]

[0138] The behavior preference matrix in step S3 is used to describe the decision preferences of the vehicle for the five decision behaviors and to characterize the preferences of different driving styles for different decisions. tendehcy for:

[0139]

[0140] Among them, T brepresents the tendency of the ego vehicle to complete behavior b, b∈{acc,dec,left,right,stay}, which corresponds to the five vehicle behaviors of acceleration, deceleration, left lane change, right lane change and staying the same.

[0141] In this embodiment, the decision preference is set to be aggressive (preferring acceleration and overtaking), and the behavior preference matrix is ​​as follows:

[0142]

[0143] The calculation formula of the comprehensive decision matrix in step S4 is:

[0144] D decision =D safety +D efficiency +D tendency

[0145] The corresponding behavior with the highest comprehensive score is the decision made by the ego vehicle on the behavior of the background vehicle.

[0146] In this embodiment, the final calculated comprehensive decision matrix is

[0147]

[0148] According to the final comprehensive decision matrix, it is determined that the comprehensive decision made by the vehicle Ego is to change lanes to the left.

[0149] This solution abstracts the real road scene through a scenario matrix, thereby filtering out redundant information and considering only the key parameters that affect vehicle decision-making, such as road parameters, the vehicle's position information, and the position and behavior information of other traffic participants (i.e., background vehicles). It further constructs a decision matrix that integrates all driving behaviors, taking into account safety, efficiency, and driving behavior preferences. Ultimately, the comprehensive decision matrix is ​​used to enable real-time, rapid, and reasonable decision-making by autonomous vehicles, enabling comprehensive horizontal and vertical decision-making, real-time and rapid decision-making, and diverse and comprehensive decision-making.

Claims

1. A decision-making and control method for an autonomous vehicle based on a scenario matrix, characterized in that: The following steps are involved: S1. Obtain road parameters, vehicle location information, and background vehicle location and behavior information; S2. Based on the data information obtained in step S1, a scene matrix is ​​constructed to describe the relative positions of the ego vehicle and background vehicles under dynamic changes in the selected scene; S3. Calculate the decision safety matrix, decision efficiency matrix, and behavior preference matrix based on the scenario matrix. S4. Calculate a comprehensive decision matrix based on the decision safety matrix, decision efficiency matrix, and behavior preference matrix. Map the comprehensive decision matrix to the behavior to complete the ego vehicle's decision and control the vehicle's driving actions accordingly. The construction process of the scene matrix in step S2 is: S21. Determine the scene matrix M direction and size; S22, determine the scene matrix of the vehicle and background vehicles M Position in S23, determine the vehicle behavior and the vehicle in the scene matrix M The corresponding relationship of position changes in The decision safety matrix in step S3 is expressed as: ， in, Indicates that when the vehicle completes the behavior Then, the safety score corresponding to the behavior; , corresponding to the five vehicle behaviors of acceleration, deceleration, left lane change, right lane change and remaining unchanged; ， ， ， in, Corresponding security weight matrix The various elements in is the scene matrix The corresponding elements in ; , Indicates the correspondence between the multiplied elements in the two matrices; Executing behavior for the ego vehicle b back t +1 moment in the scene matrix M The row position in ; Executing behavior for the ego vehicle b back t +1 moment in the scene matrix M The column position in ; is the matrix algebraization function; ， in, is the weight coefficient of the lane area where the vehicle is located; The decision efficiency matrix in step S3 is expressed as: ， Among them, the behavioral efficiency score in the efficiency matrix Indicates that when the vehicle completes the behavior After that, the efficiency score of the behavior is , corresponding to the five vehicle behaviors of acceleration, deceleration, left lane change, right lane change and remaining unchanged, when hour, Calculated by the following formula: ， when hour, Calculated by the following formula: ， in, 、 They are speed change efficiency factor and lane change efficiency factor respectively; invalid behavior means that the vehicle performs the behavior After a collision or exceeding the lane; The behavior preference matrix in step S3 is used to describe the decision preferences of the vehicle for the five decision behaviors and to characterize the preferences of different driving styles for different decisions. Expressed as: ， in, Indicates that the vehicle completes the behavior tendency, , corresponding to the five vehicle behaviors of acceleration, deceleration, left lane change, right lane change and remaining unchanged.

2. The method for decision-making and control of an autonomous driving vehicle based on a scenario matrix according to claim 1, characterized in that: The road parameters in step S1 include the number of lanes and the road length; The ego vehicle position information includes the lane in which the ego vehicle is located; The position information of the background vehicle includes the lane where the background vehicle is located and the longitudinal distance of the background vehicle relative to the vehicle. ,in Represents the serial number of the background vehicle and specifies the longitudinal distance in front of the vehicle is positive, and vice versa is negative; The behavior information of the background vehicle includes five behaviors: acceleration, deceleration, lane change left, lane change right, and remaining unchanged.

3. The method for decision-making and control of an autonomous driving vehicle based on a scenario matrix according to claim 2, characterized in that: The scene matrix in step S21 The row vector direction is horizontal, the column vector direction is vertical, and the scene matrix number of rows and number of columns The calculation formula is: ， ， ， in, 、 are adjustable parameters, representing the judgment range in front of the vehicle and the judgment range behind the vehicle respectively; is the number of lanes; is the ceiling function; is the actual distance represented by each row of the matrix, is the vehicle speed; The unit duration.

4. The method for decision-making and control of an autonomous driving vehicle based on a scenario matrix according to claim 3, characterized in that: In step S22, the vehicle and the background vehicles exist Column coordinates of the moment 、 is the lane number of the vehicle, and the row coordinates are expressed as: ， in, 、 The ego car and background car respectively exist The column coordinates of the moment, 、 The ego car and background car respectively exist The row coordinates of the moment.

5. The method for decision-making and control of an autonomous driving vehicle based on a scenario matrix according to claim 4, characterized in that: In step S23, the vehicle performs the following actions: The position after is expressed as: ， in, for The vehicle is in the scene matrix at the moment M The row position in for The vehicle is in the scene matrix at the moment M The column position in .

6. The method for decision-making and control of an autonomous driving vehicle based on a scenario matrix according to claim 1, characterized in that: The calculation formula of the comprehensive decision matrix in step S4 is: ， The corresponding behavior with the highest score in the comprehensive decision matrix is ​​the behavioral decision made by the vehicle in response to the behavior of other background vehicles.

Images