A rolling optimization decision method for unmanned vehicles considering randomness of a roundabout

Abstract

The application discloses a kind of unmanned vehicle rolling optimization decision-making methods considering surrounding vehicle randomness, establishes unmanned vehicle decision-making model, is further established the collision probability model of car and surrounding vehicle based on the two-dimensional Gaussian distribution of the future position of surrounding vehicle output by surrounding vehicle trajectory prediction module, design unmanned vehicle rolling optimization decision-making optimization problem considering surrounding vehicle randomness, and output driving decision, realize the lateral and longitudinal rolling optimization decision-making of unmanned vehicle considering surrounding vehicle randomness;This method can consider the characteristics of vehicle longitudinal system model, longitudinal control model, vehicle lateral system model, lateral control model, lateral planning model when predicting future dynamics;The collision probability of car and surrounding vehicle can be evaluated according to the future position of surrounding vehicle output by surrounding vehicle trajectory prediction module;When the collision probability of car and surrounding vehicle is higher, only minimize the collision probability of car and surrounding vehicle, realize the unmanned vehicle decision-making of safety priority.

Description

A Rolling Optimization Decision-Making Method for Autonomous Vehicles Considering Weekly Vehicle Randomness Technical Field

[0001] This invention belongs to the field of autonomous vehicles and relates to a rolling optimization decision-making method for intelligent vehicles that takes into account the randomness of weekly vehicle cycles. Background Technology

[0002] Autonomous vehicles can reduce traffic accidents caused by driver error. Autonomous driving generally includes four parts: perception, decision-making, planning, and control. The decision-making part is responsible for integrating environmental and vehicle information to generate safe driving behavior. For autonomous vehicle decision-making, not only the vehicle's current position but also its future position should be considered. In practice, the vehicle trajectory prediction module cannot accurately predict the vehicle's future position, which is a random variable following a two-dimensional Gaussian distribution. Current decision-making methods typically use the mean of the two-dimensional Gaussian distribution of the vehicle's future position as a deterministic variable, failing to fully consider the uncertainty of the vehicle's future position. Therefore, when the uncertainty of the vehicle's future position is high, decisions that do not consider its randomness may increase the collision risk of the autonomous vehicle. Summary of the Invention

[0003] To address the problems in decision-making for autonomous vehicles, this invention proposes a rolling optimization decision-making method for autonomous vehicles that considers the randomness of weekly vehicle cycles.

[0004] This invention is achieved using the following technical solution:

[0005] An autonomous vehicle rolling optimization decision-making method considering the randomness of surrounding vehicles is proposed. The autonomous vehicle's decision-making needs to avoid increasing the risk of traffic accidents by getting too close to the future trajectories of surrounding vehicles. Here, the autonomous vehicle is the autonomous vehicle in this method, and the surrounding vehicles are the set of all vehicles sensed by the autonomous vehicle through its sensors. The future trajectories of the surrounding vehicles are random variables that cannot be accurately predicted. This method, based on the future positions of the surrounding vehicles output by the surrounding vehicle trajectory prediction module, enables autonomous vehicle decision-making under the condition of randomness in the future trajectories of the surrounding vehicles. When the randomness of the future trajectories of the surrounding vehicles is high, compared to when the randomness is low, the autonomous vehicle adopts a more conservative behavior, increasing the distance between the autonomous vehicle and the surrounding vehicles to improve the safety of the autonomous vehicle. This method is based on adding a [device / system] to the autonomous vehicle. Using inertial navigation, lidar, and cameras, the system acquires the vehicle's longitudinal velocity, longitudinal displacement, lateral displacement, yaw angle, lateral velocity, yaw rate, centerline equation of the current lane, centerline equation of the target lane, and future positions of surrounding vehicles output by the surrounding vehicle trajectory prediction module. This data forms the state of the autonomous vehicle's rolling optimization decision-making system. An autonomous vehicle decision-making model is established. Based on the two-dimensional Gaussian distribution of the future positions of surrounding vehicles output by the surrounding vehicle trajectory prediction module, a collision probability model between the vehicle and surrounding vehicles is further established. An autonomous vehicle rolling optimization decision-making problem considering the randomness of surrounding vehicles is designed, and driving decisions are output. This method realizes the lateral and longitudinal rolling optimization decision-making of the autonomous vehicle considering the randomness of surrounding vehicles. The specific steps are as follows:

[0006] Step 1: Establish a decision-making model for autonomous vehicles

[0007] Establish a decision coordinate system for autonomous vehicles. The X-axis of the coordinate system is the direction from the center of mass of the autonomous vehicle to the front of the vehicle on the horizontal plane, and the Y-axis is the direction from the center of mass of the autonomous vehicle to the left side of the vehicle on the horizontal plane, perpendicular to the X-axis. The origin of the coordinate system is the position of the center of mass of the autonomous vehicle. The X-axis is the longitudinal direction of the vehicle, and the Y-axis is the lateral direction of the vehicle.

[0008] A longitudinal decision-making model for autonomous vehicles is established, which includes two parts: a longitudinal system model and a longitudinal control model. The longitudinal system model is as follows:

[0009]

[0010] Where, x x For the longitudinal system state of the vehicle, x x =[x o v x ]′, ′ denotes matrix transpose; x o v represents the longitudinal displacement of the vehicle, in meters (m). x The longitudinal speed of the vehicle is expressed in m / s. u x For the longitudinal control of the vehicle, u x =a x ;a x The longitudinal acceleration of the vehicle, in m / s² 2 a x This is derived from the longitudinal control model:

[0011] u x =f x (v,v x (2)

[0012] Among them, u x v is the longitudinal control variable for the vehicle; v is the vehicle speed decision value, in m / s; v x f is the longitudinal speed of the vehicle, in m / s. x (v,v x ) represents the longitudinal control law, f x (v,v x )=P(vv x ), where P is the proportional coefficient of the longitudinal control law, and the proportional coefficient P of the longitudinal control law is determined by the engineering tuning method of the PID controller based on the vehicle longitudinal system model;

[0013] A lateral decision-making model for autonomous vehicles is established, comprising three parts: a vehicle lateral system model, a lateral control model, and a lateral planning model. The vehicle lateral system model is as follows:

[0014]

[0015] Where y o v represents the lateral displacement of the vehicle, in meters (m). x V is the vehicle's longitudinal velocity, in m / s; ψ is the vehicle's yaw angle, in rad; v y ω is the vehicle's lateral velocity, in m / s; r is the vehicle's yaw rate, in rad / s; w f The distance from the center of the front axle to the center of gravity of the vehicle, in meters (m); w r The distance from the center of the rear axle to the center of mass of the vehicle, in meters (m); z The moment of inertia of the vehicle about a plane perpendicular to the X-axis and Y-axis, expressed in kg·m. 2 C f C represents the front wheel lateral stiffness, in N / rad. r Rear wheel lateral stiffness (N / rad); δ is the front wheel steering angle (rad); m is the vehicle mass (kg); The resulting state-space model of the vehicle's lateral system is:

[0016]

[0017] in x y For the vehicle's lateral system state, x y =[y o ψ rv y ]′;u y For vehicle lateral control, u y =δ; δ is the front wheel steering angle of the vehicle, in rad, derived from the lateral control model:

[0018] δ=f y (R(l,T,x o ),x y )=R(l,T,x o )-Kx y (5)

[0019] Where δ is the front wheel steering angle of the vehicle, in rad; f y (R(l,T,x o ),x y ) represents the lateral control law; x y Let R(l,T,x) represent the vehicle lateral system state; K is the lateral control gain vector, with a dimension of 1 row and 4 columns, obtained using the pole placement method for linear system controller design based on the vehicle lateral system state-space model; o The lateral displacement (in meters) is the reference displacement, derived from the lateral planning model.

[0020] R(l,T,x o )=(1-w(t,T))y0(x o )+w(t,T)y l (x o (6)

[0021] Where y0(x o (xo) represents the lateral displacement of the current lane's centerline at the longitudinal displacement point xo, and the equation of the current lane's centerline is... a0 is the cubic coefficient of the current lane's centerline equation, b0 is the quadratic coefficient of the current lane's centerline equation, c0 is the linear coefficient of the current lane's centerline equation, and d0 is the zeroth-order coefficient of the current lane's centerline equation; y l (x o Let xo be the lateral displacement of the centerline of the target lane l at the longitudinal displacement xo, and let xo be the equation of the centerline of the target lane l. a l Let b be the cubic coefficient of the centerline equation of the target lane l. l c represents the quadratic coefficient of the centerline equation of the target lane l. l Let d be the first-order coefficient of the centerline equation of the target lane l.l Let l be the zeroth-order coefficient of the centerline equation of the target lane l, where l is the target lane number decision, and 1 ≤ l ≤ L. n And L is an integer. n The number of drivable lanes is an integer greater than 1; w(t,T) is the lane-changing weight, derived from the following formula:

[0022]

[0023] Among them, t s The execution cycle of the lateral planning model is denoted by T, in seconds; T represents the lane change time decision, in seconds; and t represents the current lane change time, derived from the following formula:

[0024]

[0025] Among them, y o d1 represents the lateral displacement of the vehicle, in meters; d2 represents the zero-order coefficient of the centerline equation of the current lane; d3 represents the zero-order coefficient of the centerline equation of the target lane l.

[0026] Combining the vehicle's longitudinal system model and longitudinal control model, the decision-making model for autonomous vehicles is obtained:

[0027]

[0028] in, x represents the state of the autonomous vehicle decision-making model, x = [x x x y ]′; y is the output of the autonomous vehicle decision model; ψ is the vehicle yaw angle, in rad; f x (v,v x ) represents the longitudinal control law; f y (R(l,T,x o ),x y This is the lateral control law;

[0029] Discretize the autonomous vehicle decision model to obtain a discretized autonomous vehicle decision model:

[0030]

[0031] Where x(k) is the state x of the autonomous vehicle decision model at time k; x(k+1) is the state x of the autonomous vehicle decision model at time k+1; Let T be the derivative of the state x of the autonomous vehicle decision-making model at time k with respect to time; sy(k+1) represents the discrete sampling time of the model; y(k+1) represents the output of the autonomous vehicle decision model at time k+1.

[0032] Step 2: Establish a collision probability model between this vehicle and surrounding vehicles.

[0033] The future position of the vehicle output by the vehicle trajectory prediction module follows a two-dimensional Gaussian distribution, with the following probability density:

[0034]

[0035] Where, x i (k) represents the predicted longitudinal position of the i-th vehicle at time k, and is a random variable in meters; y i (k) represents the predicted lateral position of the i-th vehicle at time k, and is a random variable in meters (m); μ xi (k) represents the mean longitudinal predicted position of the i-th vehicle at time k; μ yi (k) represents the mean lateral predicted position of the i-th vehicle at time k; σ xi (k) represents the longitudinal predicted position variance of the i-th vehicle at time k; σ yi (k) represents the variance of the lateral predicted position of the i-th vehicle at time k; ρ i (k) is the longitudinal predicted position correlation coefficient of the i-th vehicle at time k;

[0036] Simplifying the circumference rhombus as a rectangle, the coordinates of its four corners are calculated as follows:

[0037]

[0038] in, Let be the X-axis coordinate of the first angle of the circumferential vehicle i at time k, in meters. Let be the Y-axis coordinate of the first angle of vehicle i at time k, in meters; Let x be the X-axis coordinate of the second angle of the circumferential vehicle i at time k, in meters. Let be the Y-axis coordinate of the second angle of vehicle i at time k, in meters. Let be the X-axis coordinate of the first triangle of vehicle i at time k, in meters. Let be the Y-axis coordinate of the first triangle of vehicle i at time k, in meters. Let x be the X-axis coordinate of the fourth angle of the circumferential vehicle i at time k, in meters. Let μ be the Y-axis coordinate of the fourth angle of vehicle i at time k, in meters. xi (k) represents the mean longitudinal predicted position of the i-th vehicle at time k; μ yi (k) represents the mean of the lateral predicted position of the i-th vehicle at time k; vleni The length of the diagonal of the circumference i is half the length, in meters. And ζ i Let be the angle between the line connecting the centroid of rook i and its first angle and the left side of rook i, in rad. len i Let wid be the length of vehicle i in meters. i Let θ be the width of vehicle i in meters; i (k) is the heading angle of vehicle i, in rad.

[0039] The coordinates of the four corners of this vehicle are calculated as follows:

[0040]

[0041] Where, x 1 (k) represents the X-axis coordinate of the first angle of the vehicle at time k, in meters; y 1 (k) represents the Y-axis coordinate of the first angle of the vehicle at time k, in meters; x 2 (k) represents the X-axis coordinate of the second angle of the vehicle at time k, in meters; y 2 (k) represents the Y-axis coordinate of the second angle of the vehicle at time k, in meters; x 3 (k) represents the X-axis coordinate of the vehicle's first triangle at time k, in meters; y 3 (k) represents the Y-axis coordinate of the vehicle's first triangle at time k, in meters; x 4 (k) represents the X-axis coordinate of the fourth angle of the vehicle at time k, in meters; y 4 (k) represents the Y-axis coordinate of the fourth corner of the vehicle at time k, in meters; x o The longitudinal displacement of the vehicle is expressed in meters (m); y o Vlen represents the lateral displacement of the vehicle, in meters (m); vlen represents the half-length of the vehicle's diagonal, in meters (m). And ζ is the angle between the line connecting the centroid of the vehicle and its first corner, and the left side of the vehicle, in rad. len is the vehicle length in meters (m), and wid is the vehicle width in meters (m).

[0042] The collision probability model between this vehicle and the surrounding vehicles is as follows:

[0043]

[0044] Where, p i (k) represents the collision probability between this vehicle and surrounding vehicle i; x i (k) represents the predicted longitudinal position of the i-th vehicle at time k, and is a random variable in meters; y i(k) is the predicted lateral position of the i-th vehicle at time k, which is a random variable in m; Let be the lower bound of the collision probability between this vehicle and surrounding vehicle i over the X-axis of the coordinate system, in meters. Let be the upper limit of the integral of the collision probability between this vehicle and surrounding vehicle i along the X-axis of the coordinate system, in meters; Let be the lower bound of the collision probability between this vehicle and surrounding vehicle i over the Y-axis of the coordinate system, in meters. The upper bound of the integral of the collision probability between this vehicle and surrounding vehicle i along the Y-axis of the coordinate system, in meters:

[0045]

[0046] in, j1 = 1, 2, 3, 4 are the X-axis coordinates of the first, second, third, and fourth angles of the vehicle at time k, respectively. j1 = 1, 2, 3, 4 are the Y-axis coordinates of the first, second, third, and fourth angles of the vehicle at time k, respectively. j2 = 1, 2, 3, 4 are the X-axis coordinates of the first, second, third, and fourth angles of the circumferential vehicle i at time k, respectively. j2 = 1, 2, 3, 4 are the Y-axis coordinates of the first, second, third, and fourth angles of the circumferential car i at time k, respectively.

[0047] in, The symbol {} represents a set;

[0048] Step 3: Design the rolling optimization decision-making problem for autonomous vehicles that consider the randomness of weekly vehicle movement.

[0049] Establish collision avoidance target J r1 :

[0050]

[0051] Among them, J r1 Indicates the target to be avoided; p i (k) represents the collision probability between this vehicle and surrounding vehicle i; P represents the prediction step size, which is an integer greater than 1; Σ represents the addition symbol; Π represents the multiplication symbol; n represents the number of surrounding vehicles;

[0052] Establish rapid target J s1 :

[0053]

[0054] Among them, J s1 Let v(k) represent the rapid target, where v(k) is the vehicle speed decision at time k, in m / s; N is the control step size, 1 ≤ N ≤ P and is an integer, where P represents the prediction step size; v ref The desired vehicle speed, in m / s, v ref =τv roadmax +(1-τ)v roadmin ;v roadmax The maximum speed limit for the road, in m / s; v roadmin τ is the minimum speed limit for the road, in m / s; τ is the speed factor, 0≤τ≤1;

[0055] Establish lane keeping target J s2 :

[0056]

[0057] Among them, J s2 The lane keeping target is represented by N; the control step size is N; Δl(k) is the lane change amount, Δl(k) = l(k) - l(k-1), and l(k) is the target lane number decision at time k;

[0058] Establish a comfortable lane-changing target J s3 :

[0059]

[0060] Among them, J s3 The target lane change is a comfortable lane change; N is the control step size; T(k) is the lane change time decision at time k.

[0061] Establish lane constraints:

[0062] 1≤l(k)≤L n ,l∈Z (20)

[0063] l(k) represents the target lane numbering decision at time k; L n Z represents the number of drivable lanes; Z represents the set of integers.

[0064] Establish lane-changing time constraints:

[0065] T min ≤T(k)≤T max (twenty one)

[0066] Where T(k) is the lane-changing time decision at time k; T min This is the lower limit for lane change time; T max This is the maximum time limit for lane changing;

[0067] Establish constraints on velocity change:

[0068] Δv min ≤Δv(k)≤Δv max (twenty two)

[0069] Δv(k) is the change in velocity, in m / s, Δv(k) = v(k) - Δv(k-1), where v(k) is the vehicle speed decision at time k; Δv min The lower limit of the change in velocity, in m / s, Δv max This represents the upper limit of the velocity change, in m / s.

[0070] Establish speed constraints:

[0071] 0≤v(k)≤v roadmax (twenty three)

[0072] Where v(k) is the vehicle speed decision at time k; v roadmax This is the maximum speed limit for the road;

[0073] The model predictive control method is adopted, which predicts the future state of the system based on a discretized decision model of the autonomous vehicle, and defines the vehicle speed decision control sequence v. V =[v(1) v(1) … v(N)], target lane numbering control sequence l V =[l(1) … l(N)]、Lane change time control sequence T V = [T(1) … T(N)], where N is the control time domain of the model predictive control method, v(1) is the vehicle speed decision at time 1, v(2) is the vehicle speed decision at time 2, v(N) is the vehicle speed decision at time N, l(1) is the target lane number decision at time 1, l(2) is the target lane number decision at time 2, l(N) is the target lane number decision at time N, T(1) is the lane change time decision at time 1, T(2) is the lane change time decision at time 2, and T(N) is the lane change time decision at time N; By weighting the collision avoidance target, the speed target, the lane keeping target, and the comfortable lane change target, and combining the lane constraints, lane change time constraints, speed change constraints, and speed constraints, we obtain the unmanned vehicle rolling optimization decision optimization problem considering the randomness of the surrounding vehicles:

[0074]

[0075] Among them, v V =[v(1) v(1) … v(N)] represents the vehicle speed decision control sequence, l V =[l(1) … l(N)] is the target lane numbering control sequence, T V=[T(1) … T(N)] is the lane-changing time control sequence, and J is the objective function for the rolling optimization decision of the unmanned vehicle considering the randomness of the weekly vehicle; J r1 To avoid collision with the target; C is the danger deviation constant, C > 1; p i (k) represents the collision probability between this vehicle and surrounding vehicle i; ε is the collision risk tolerance factor, 0 < ε < 1; J s1 For the rapid objective; J s2 To maintain lane target; J s3 For the goal of comfortable lane changing; Γ s1 For fast target weights; Γ s2 To maintain the target weight for lane keeping; Γ s3 Weighting of comfort lane changing objectives; Γ r1 To avoid collision target weights; J smin For safety objective J s The minimum value of J smax For safety objective J s The maximum value of J s For safety objectives:

[0076] J s =Γ s1 J s1 +Γ s2 J s2 +Γ s3 J s3 +Γ r1 J r1 (twenty four)

[0077] Among them, J r1 To avoid colliding with the target; J s1 For the rapid objective; J s2 To maintain lane target; J s3 For the goal of comfortable lane changing; Γ s1 For fast target weights; Γ s2 To maintain the target weight for lane keeping; Γ s3 Weighting of the target for comfortable lane changing; Γ r1 To avoid collisions, target weights are used.

[0078] The vehicle speed decision control sequence v V =[v(1) v(1) … v(N)], target lane numbering control sequence l V = [l(1) … l(N)] and lane change time control sequence T V =[T(1) … T(N)] are used together as variables to be optimized. A genetic algorithm is used to solve the rolling optimization decision problem of an autonomous vehicle that considers the randomness of the vehicle cycle. The vehicle speed decision control sequence v that minimizes the objective function J of the rolling optimization decision of the autonomous vehicle is obtained. V=[v(1) v(1) … v(N)], target lane numbering control sequence l V = [l(1) … l(N)] and lane change time control sequence T V =[T(1) … T(N)] is defined as the optimal control sequence in The optimal vehicle speed control sequence; The optimal target lane numbering control sequence; This is the optimal lane-changing time control sequence. * This indicates that the variable is optimal;

[0079] When the state of the autonomous vehicle's rolling optimization decision-making system is collected and updated in real time, the optimal control sequence is obtained by solving the autonomous vehicle rolling optimization decision-making problem that considers the randomness of the vehicle cycle. And the optimal vehicle speed control sequence The first element Through the vertical control law f x (v,v x Converted to vehicle longitudinal acceleration a x Optimal target lane number decision vector The first element Optimal lane-changing time decision vector The first element Through the lateral control law f y (R(l,T,x o ),x y This is converted into the vehicle's front wheel steering angle δ to achieve longitudinal and lateral control of the vehicle.

[0080] The beneficial effects of this invention are as follows:

[0081] This invention provides a rolling optimization decision-making method for autonomous vehicles that considers the randomness of surrounding vehicles. It establishes an autonomous vehicle decision-making model, allowing the prediction of future dynamics to consider the characteristics of the vehicle's longitudinal system model, longitudinal control model, lateral system model, lateral control model, and lateral planning model. It also establishes a collision probability model between the vehicle and surrounding vehicles, enabling the assessment of collision probabilities based on the future positions of surrounding vehicles output by the surrounding vehicle trajectory prediction module. Furthermore, it designs an autonomous vehicle rolling optimization decision-making problem that considers the randomness of surrounding vehicles, allowing the minimization of collision probabilities between the vehicle and surrounding vehicles when the collision probability is high, thus achieving safety-first autonomous vehicle decision-making. Attached Figure Description

[0082] Figure 1 is a simplified flowchart of the rolling optimization decision-making method for unmanned vehicles that considers the randomness of weekly vehicles according to the present invention;

[0083] Figure 2 is a schematic diagram of the integration region of the collision probability model between the vehicle and surrounding vehicles in this method;

[0084] Figure 3 is a control block diagram of this method; Detailed Implementation

[0085] The present invention will now be described in detail with reference to the accompanying drawings:

[0086] An autonomous vehicle rolling optimization decision-making method considering the randomness of surrounding vehicles is proposed. The autonomous vehicle's decision-making needs to avoid increasing the risk of traffic accidents by getting too close to the future trajectories of surrounding vehicles. Here, the autonomous vehicle is the autonomous vehicle in this method, and the surrounding vehicles are the set of all vehicles sensed by the autonomous vehicle through its sensors. The future trajectories of the surrounding vehicles are random variables that cannot be accurately predicted. This method, based on the future positions of the surrounding vehicles output by the surrounding vehicle trajectory prediction module, enables autonomous vehicle decision-making under the condition of randomness in the future trajectories of the surrounding vehicles. When the randomness of the future trajectories of the surrounding vehicles is high, compared to when the randomness is low, the autonomous vehicle adopts a more conservative behavior, increasing the distance between the autonomous vehicle and the surrounding vehicles to improve the safety of the autonomous vehicle. This method is based on the autonomous vehicle's... The added inertial navigation, lidar, and cameras acquire the vehicle's longitudinal velocity, longitudinal displacement, lateral displacement, yaw angle, lateral velocity, yaw rate, centerline equation of the current lane, centerline equation of the target lane l, and the future positions of surrounding vehicles output by the surrounding vehicle trajectory prediction module. These are the states of the autonomous vehicle's rolling optimization decision-making system. An autonomous vehicle decision-making model is established. Based on the two-dimensional Gaussian distribution of the future positions of surrounding vehicles output by the surrounding vehicle trajectory prediction module, a collision probability model between the vehicle and surrounding vehicles is further established. An autonomous vehicle rolling optimization decision-making problem considering the randomness of surrounding vehicles is designed, and driving decisions are output. The process of realizing the autonomous vehicle's lateral and longitudinal rolling optimization decision-making considering the randomness of surrounding vehicles is as follows:

[0087] Step 1: Establish a decision-making model for autonomous vehicles

[0088] In the model predictive control method, it is necessary to establish a model of the controlled object. As shown in the first step of Figure 1, the decision model of the unmanned vehicle describing the displacement of the vehicle is established in the first step.

[0089] Establish a decision coordinate system for autonomous vehicles. The X-axis of the coordinate system is the direction from the center of mass of the autonomous vehicle to the front of the vehicle on the horizontal plane, and the Y-axis is the direction from the center of mass of the autonomous vehicle to the left side of the vehicle on the horizontal plane, perpendicular to the X-axis. The origin of the coordinate system is the position of the center of mass of the autonomous vehicle. The X-axis is the longitudinal direction of the vehicle, and the Y-axis is the lateral direction of the vehicle.

[0090] A longitudinal decision-making model for autonomous vehicles is established, which includes two parts: a longitudinal system model and a longitudinal control model. The longitudinal system model is as follows:

[0091]

[0092] Where, x x For the longitudinal system state of the vehicle, x x =[x o v x ]′, ′ denotes matrix transpose; x o v represents the longitudinal displacement of the vehicle, in meters (m). x The longitudinal speed of the vehicle is expressed in m / s. u x For the longitudinal control of the vehicle, u x =a x ;a x The longitudinal acceleration of the vehicle, in m / s² 2 a x This is derived from the longitudinal control model:

[0093] u x =f x (v,v x (2)

[0094] Among them, u x v is the longitudinal control variable for the vehicle; v is the vehicle speed decision value, in m / s; v x f is the longitudinal speed of the vehicle, in m / s. x (v,v x ) represents the longitudinal control law, f x (v,v x )=P(vv x ), where P is the proportional coefficient of the longitudinal control law, and the proportional coefficient P of the longitudinal control law is determined by the engineering tuning method of the PID controller based on the vehicle longitudinal system model;

[0095] A lateral decision-making model for autonomous vehicles is established, comprising three parts: a vehicle lateral system model, a lateral control model, and a lateral planning model. The vehicle lateral system model is as follows:

[0096]

[0097] Where y o v represents the lateral displacement of the vehicle, in meters (m). x V is the vehicle's longitudinal velocity, in m / s; ψ is the vehicle's yaw angle, in rad; v y ω is the vehicle's lateral velocity, in m / s; r is the vehicle's yaw rate, in rad / s; wf The distance from the center of the front axle to the center of gravity of the vehicle, in meters (m); w r The distance from the center of the rear axle to the center of mass of the vehicle, in meters (m); z The moment of inertia of the vehicle about a plane perpendicular to the X-axis and Y-axis, expressed in kg·m. 2 C f C represents the front wheel lateral stiffness, in N / rad. r Rear wheel lateral stiffness (N / rad); δ is the front wheel steering angle (rad); m is the vehicle mass (kg); The resulting state-space model of the vehicle's lateral system is:

[0098]

[0099] in x y For the vehicle's lateral system state, x y =[y o ψ rv y ]′;u y For vehicle lateral control, u y =δ; δ is the front wheel steering angle of the vehicle, in rad, derived from the lateral control model:

[0100] δ=f y (R(l,T,x o ),x y )=R(l,T,x o )-Kx y (5)

[0101] Where δ is the front wheel steering angle of the vehicle, in rad; f y (R(l,T,x o ),x y ) represents the lateral control law; x y Let R(l,T,x) represent the vehicle lateral system state; K is the lateral control gain vector, with a dimension of 1 row and 4 columns, obtained using the pole placement method for linear system controller design based on the vehicle lateral system state-space model; o The lateral displacement (in meters) is the reference displacement, derived from the lateral planning model.

[0102] R(l,T,x o )=(1-w(t,T))y0(x o )+w(t,T)y l (x o (6)

[0103] Where y0(x o(xo) represents the lateral displacement of the current lane's centerline at the longitudinal displacement point xo, and the equation of the current lane's centerline is... a0 is the cubic coefficient of the current lane's centerline equation, b0 is the quadratic coefficient of the current lane's centerline equation, c0 is the linear coefficient of the current lane's centerline equation, and d0 is the zeroth-order coefficient of the current lane's centerline equation; y l (x o Let xo be the lateral displacement of the centerline of the target lane l at the longitudinal displacement xo, and let xo be the equation of the centerline of the target lane l. a l Let b be the cubic coefficient of the centerline equation of the target lane l. l c represents the quadratic coefficient of the centerline equation of the target lane l. l Let d be the first-order coefficient of the centerline equation of the target lane l. l Let l be the zeroth-order coefficient of the centerline equation of the target lane l, where l is the target lane number decision, and 1 ≤ l ≤ L. n And L is an integer. n The number of drivable lanes is an integer greater than 1; w(t,T) is the lane-changing weight, derived from the following formula:

[0104]

[0105] Among them, t s The execution cycle of the lateral planning model is denoted by T, in seconds; T represents the lane change time decision, in seconds; and t represents the current lane change time, derived from the following formula:

[0106]

[0107] Among them, y o d1 represents the lateral displacement of the vehicle, in meters; d2 represents the zero-order coefficient of the centerline equation of the current lane; d3 represents the zero-order coefficient of the centerline equation of the target lane l.

[0108] Combining the vehicle's longitudinal system model and longitudinal control model, the decision-making model for autonomous vehicles is obtained:

[0109]

[0110] in, x represents the state of the autonomous vehicle decision-making model, x = [x x x y ]′; y is the output of the autonomous vehicle decision model; ψ is the vehicle yaw angle, in rad; f x (v,v x ) represents the longitudinal control law; f y (R(l,T,xo ),x y This is the lateral control law;

[0111] Discretize the autonomous vehicle decision model to obtain a discretized autonomous vehicle decision model:

[0112]

[0113] Where x(k) is the state x of the autonomous vehicle decision model at time k; x(k+1) is the state x of the autonomous vehicle decision model at time k+1; Let T be the derivative of the state x of the autonomous vehicle decision-making model at time k with respect to time; s y(k+1) represents the discrete sampling time of the model; y(k+1) represents the output of the autonomous vehicle decision model at time k+1.

[0114] Step 2: Establish a collision probability model between this vehicle and surrounding vehicles.

[0115] In the rolling optimization decision of autonomous vehicles that considers the randomness of the surrounding vehicles, the surrounding vehicles are important to the safety of the vehicle itself, because the displacement of the vehicle and the displacement of the surrounding vehicles together determine whether a collision will occur. That is, the closer the displacement of the vehicle and the displacement of the surrounding vehicles are, the more likely a collision is to occur. The collision probability model between the vehicle and the surrounding vehicles is established in step two, as shown in the second step of Figure 1.

[0116] The future position of the vehicle output by the vehicle trajectory prediction module follows a two-dimensional Gaussian distribution, with the following probability density:

[0117]

[0118] Where, x i (k) represents the predicted longitudinal position of the i-th vehicle at time k, and is a random variable in meters; y i (k) represents the predicted lateral position of the i-th vehicle at time k, and is a random variable in meters (m); μ xi (k) represents the mean longitudinal predicted position of the i-th vehicle at time k; μ yi (k) represents the mean lateral predicted position of the i-th vehicle at time k; σ xi (k) represents the longitudinal predicted position variance of the i-th vehicle at time k; σ yi (k) represents the variance of the lateral predicted position of the i-th vehicle at time k; ρ i (k) is the longitudinal predicted position correlation coefficient of the i-th vehicle at time k;

[0119] Simplifying the circumference rhombus as a rectangle, the coordinates of its four corners are calculated as follows:

[0120]

[0121] in, Let be the X-axis coordinate of the first angle of the circumferential vehicle i at time k, in meters. Let be the Y-axis coordinate of the first angle of vehicle i at time k, in meters; Let x be the X-axis coordinate of the second angle of the circumferential vehicle i at time k, in meters. Let be the Y-axis coordinate of the second angle of vehicle i at time k, in meters. Let be the X-axis coordinate of the first triangle of vehicle i at time k, in meters. Let be the Y-axis coordinate of the first triangle of vehicle i at time k, in meters. Let x be the X-axis coordinate of the fourth angle of the circumferential vehicle i at time k, in meters. Let μ be the Y-axis coordinate of the fourth angle of vehicle i at time k, in meters. xi (k) represents the mean longitudinal predicted position of the i-th vehicle at time k; μ yi (k) represents the mean of the lateral predicted position of the i-th vehicle at time k; vlen i The length of the diagonal of the circumference i is half the length, in meters. And ζ i Let be the angle between the line connecting the centroid of rook i and its first angle and the left side of rook i, in rad. len i Let wid be the length of vehicle i in meters. i Let θ be the width of vehicle i in meters; i (k) is the heading angle of vehicle i, in rad.

[0122] The coordinates of the four corners of this vehicle are calculated as follows:

[0123]

[0124] Where, x 1 (k) represents the X-axis coordinate of the first angle of the vehicle at time k, in meters; y 1 (k) represents the Y-axis coordinate of the first angle of the vehicle at time k, in meters; x 2 (k) represents the X-axis coordinate of the second angle of the vehicle at time k, in meters; y 2 (k) represents the Y-axis coordinate of the second angle of the vehicle at time k, in meters; x 3 (k) represents the X-axis coordinate of the vehicle's first triangle at time k, in meters; y 3 (k) represents the Y-axis coordinate of the vehicle's first triangle at time k, in meters; x 4(k) represents the X-axis coordinate of the fourth angle of the vehicle at time k, in meters; y 4 (k) represents the Y-axis coordinate of the fourth corner of the vehicle at time k, in meters; x o The longitudinal displacement of the vehicle is expressed in meters (m); y o Vlen represents the lateral displacement of the vehicle, in meters (m); vlen represents the half-length of the vehicle's diagonal, in meters (m). And ζ is the angle between the line connecting the centroid of the vehicle and its first corner, and the left side of the vehicle, in rad. len is the vehicle length in meters (m), and wid is the vehicle width in meters (m).

[0125] The collision probability model between this vehicle and the surrounding vehicles is as follows:

[0126]

[0127] Where, p i (k) represents the collision probability between this vehicle and surrounding vehicle i; x i (k) represents the predicted longitudinal position of the i-th vehicle at time k, and is a random variable in meters; y i (k) represents the predicted lateral position of the i-th surrounding vehicle at time k, a random variable in units of m. Figure 2 shows a schematic diagram of the integration region in the collision probability model between the vehicle and the surrounding vehicles. The integration region is centered on the centroid of the autonomous vehicle and includes the minimum bounding rectangle of the first, second, third, and fourth corners of the vehicle, and the minimum bounding rectangle of the first, second, third, and fourth corners of the translated surrounding vehicle i. Let be the lower bound of the collision probability between this vehicle and surrounding vehicle i over the X-axis of the coordinate system, in meters. Let be the upper limit of the integral of the collision probability between this vehicle and surrounding vehicle i along the X-axis of the coordinate system, in meters; Let be the lower bound of the collision probability between this vehicle and surrounding vehicle i over the Y-axis of the coordinate system, in meters. The upper bound of the integral of the collision probability between this vehicle and surrounding vehicle i along the Y-axis of the coordinate system, in meters:

[0128]

[0129] in, j1 = 1, 2, 3, 4 are the X-axis coordinates of the first, second, third, and fourth angles of the vehicle at time k, respectively. j1 = 1, 2, 3, 4 are the Y-axis coordinates of the first, second, third, and fourth angles of the vehicle at time k, respectively. j2 = 1, 2, 3, 4 are the X-axis coordinates of the first, second, third, and fourth angles of the circumferential vehicle i at time k, respectively. j2 = 1, 2, 3, 4 are the Y-axis coordinates of the first, second, third, and fourth angles of the circumferential car i at time k, respectively.

[0130] in, The symbol {} represents a set;

[0131] Step 3: Design the rolling optimization decision-making problem for autonomous vehicles that consider the randomness of weekly vehicle movement.

[0132] Based on the autonomous vehicle decision-making model and the collision probability model between the vehicle and surrounding vehicles, the output y of the autonomous vehicle decision-making model and the collision probability p between the vehicle and surrounding vehicle i can be predicted. i (k), but further design of objective function and constraints is needed to measure and compare the performance of autonomous vehicle decision-making so that intelligent vehicle can make optimal autonomous vehicle decision-making, as shown in the third step of Figure 1.

[0133] Establish collision avoidance target J r1 :

[0134]

[0135] Among them, J r1 Indicates the target to be avoided; p i (k) represents the collision probability between this vehicle and surrounding vehicle i; P represents the prediction step size, which is an integer greater than 1; Σ represents the addition symbol; Π represents the multiplication symbol; n represents the number of surrounding vehicles;

[0136] Establish rapid target J s1 :

[0137]

[0138] Among them, J s1 Let v(k) represent the rapid target, where v(k) is the vehicle speed decision at time k, in m / s; N is the control step size, 1 ≤ N ≤ P and is an integer, where P represents the prediction step size; v ref The desired vehicle speed, in m / s, v ref =τv roadmax +(1-τ)v roadmin ;v roadmax The maximum speed limit for the road, in m / s; v roadmin τ is the minimum speed limit for the road, in m / s; τ is the speed factor, 0≤τ≤1;

[0139] Establish lane keeping target J s2 :

[0140]

[0141] Among them, J s2 The lane keeping target is represented by N; the control step size is N; Δl(k) is the lane change amount, Δl(k) = l(k) - l(k-1), and l(k) is the target lane number decision at time k;

[0142] Establish a comfortable lane-changing target J s3 :

[0143]

[0144] Among them, J s3 The target lane change is a comfortable lane change; N is the control step size; T(k) is the lane change time decision at time k.

[0145] Establish lane constraints:

[0146] 1≤l(k)≤L n ,l∈Z (20)

[0147] l(k) represents the target lane numbering decision at time k; L n Z represents the number of drivable lanes; Z represents the set of integers.

[0148] Establish lane-changing time constraints:

[0149] T min ≤T(k)≤T max (twenty one)

[0150] Where T(k) is the lane-changing time decision at time k; T min This is the lower limit for lane change time; T max This is the maximum time limit for lane changing;

[0151] Establish constraints on velocity change:

[0152] Δv min ≤Δv(k)≤Δv max (twenty two)

[0153] Δv(k) is the change in velocity, in m / s, Δv(k) = v(k) - Δv(k-1), where v(k) is the vehicle speed decision at time k; Δv min The lower limit of the change in velocity, in m / s, Δv max This represents the upper limit of the velocity change, in m / s.

[0154] Establish speed constraints:

[0155] 0≤v(k)≤v roadmax (twenty three)

[0156] Where v(k) is the vehicle speed decision at time k; v roadmax This is the maximum speed limit for the road;

[0157] The model predictive control method is adopted, which predicts the future state of the system based on a discretized decision model of the autonomous vehicle, and defines the vehicle speed decision control sequence v. V =[v(1) v(1) … v(N)], target lane numbering control sequence l V =[l(1) … l(N)]、Lane change time control sequence T V = [T(1) … T(N)], where N is the control time domain of the model predictive control method, v(1) is the vehicle speed decision at time 1, v(2) is the vehicle speed decision at time 2, v(N) is the vehicle speed decision at time N, l(1) is the target lane number decision at time 1, l(2) is the target lane number decision at time 2, l(N) is the target lane number decision at time N, T(1) is the lane change time decision at time 1, T(2) is the lane change time decision at time 2, and T(N) is the lane change time decision at time N; By weighting the collision avoidance target, the speed target, the lane keeping target, and the comfortable lane change target, and combining the lane constraints, lane change time constraints, speed change constraints, and speed constraints, we obtain the unmanned vehicle rolling optimization decision optimization problem considering the randomness of the surrounding vehicles:

[0158]

[0159] Among them, v V =[v(1) v(1) … v(N)] represents the vehicle speed decision control sequence, l V =[l(1) … l(N)] is the target lane numbering control sequence, T V =[T(1) … T(N)] is the lane-changing time control sequence, and J is the objective function for the rolling optimization decision of the unmanned vehicle considering the randomness of the weekly vehicle; J r1 To avoid collision with the target; C is the danger deviation constant, C > 1; p i (k) represents the collision probability between this vehicle and surrounding vehicle i; ε is the collision risk tolerance factor, 0 < ε < 1; J s1 For the rapid objective; J s2 To maintain lane target; J s3 For the goal of comfortable lane changing; Γ s1 For fast target weights; Γ s2 To maintain the target weight for lane keeping; Γ s3 Weighting of the target for comfortable lane changing; Γr1 To avoid collision target weights; J smin For safety objective J s The minimum value of J smax For safety objective J s The maximum value of J s For safety objectives:

[0160] J s =Γ s1 J s1 +Γ s2 J s2 +Γ s3 J s3 +Γ r1 J r1 (twenty four)

[0161] Among them, J r1 To avoid colliding with the target; J s1 For the rapid objective; J s2 To maintain lane target; J s3 For the goal of comfortable lane changing; Γ s1 For fast target weights; Γ s2 To maintain the target weight for lane keeping; Γ s3 Weighting of the target for comfortable lane changing; Γ r1 To avoid collisions, target weights are used.

[0162] The vehicle speed decision control sequence v V =[v(1) v(1) … v(N)], target lane numbering control sequence l V = [l(1) … l(N)] and lane change time control sequence T V =[T(1) … T(N)] are used together as variables to be optimized. A genetic algorithm is used to solve the rolling optimization decision problem of an autonomous vehicle that considers the randomness of the vehicle cycle. The vehicle speed decision control sequence v that minimizes the objective function J of the rolling optimization decision of the autonomous vehicle is obtained. V =[v(1) v(1) … v(N)], target lane numbering control sequence l V = [l(1) … l(N)] and lane change time control sequence T V =[T(1) … T(N)] is defined as the optimal control sequence in The optimal vehicle speed control sequence; The optimal target lane numbering control sequence; This is the optimal lane-changing time control sequence. * This indicates that the variable is optimal;

[0163] When the state of the autonomous vehicle's rolling optimization decision-making system is collected and updated in real time, the optimal control sequence is obtained by solving the autonomous vehicle rolling optimization decision-making problem that considers the randomness of the vehicle cycle. And the optimal vehicle speed control sequence The first element Through the vertical control law f x (v,v x Converted to vehicle longitudinal acceleration a x Optimal target lane number decision vector The first element Optimal lane-changing time decision vector The first element Through the lateral control law f y (R(l,T,x o ),x y The angle is converted into the front wheel steering angle δ of the vehicle, as shown in Figure 3, to achieve longitudinal and lateral control of the vehicle.

Claims

1. A rolling optimization decision-making method for autonomous vehicles considering the randomness of surrounding vehicles. The autonomous vehicle's decision-making needs to avoid increasing the risk of traffic accidents by getting too close to the future trajectories of surrounding vehicles. Here, "autonomous vehicle" refers to the autonomous vehicle in this method, and "surrounding vehicles" is the set of all vehicles perceived by the autonomous vehicle through sensors. The future trajectories of surrounding vehicles are random variables that cannot be accurately predicted. This method, based on the future positions of surrounding vehicles output by the surrounding vehicle trajectory prediction module, enables autonomous vehicle decision-making under the condition that the future trajectories of surrounding vehicles are random. When the randomness of the future trajectories of surrounding vehicles is high, compared to when the randomness is low, the autonomous vehicle adopts a more conservative behavior, increasing the distance between itself and surrounding vehicles to improve the safety of the autonomous vehicle. This method is based on the autonomous vehicle... The system uses an inertial navigation system, lidar, and camera to acquire data on the autonomous vehicle's longitudinal velocity, longitudinal displacement, lateral displacement, yaw angle, lateral velocity, yaw rate, centerline equation of the current lane, centerline equation of the target lane, and the future positions of surrounding vehicles output by the surrounding vehicle trajectory prediction module. This data forms the state of the autonomous vehicle's rolling optimization decision-making system. An autonomous vehicle decision-making model is established. Based on the two-dimensional Gaussian distribution of the future positions of surrounding vehicles output by the surrounding vehicle trajectory prediction module, a collision probability model between the vehicle and surrounding vehicles is further established. An autonomous vehicle rolling optimization decision-making problem considering the randomness of surrounding vehicles is designed, and a driving decision is output. This achieves autonomous vehicle lateral and longitudinal rolling optimization decision-making considering the randomness of surrounding vehicles. Its key feature is... The specific steps of this method are as follows: Step 1: Establish an autonomous vehicle decision-making model. Establish an autonomous vehicle decision-making coordinate system. The X-axis of the coordinate system is the direction from the center of mass of the autonomous vehicle to the front of the vehicle on the horizontal plane. The Y-axis of the coordinate system is the direction from the center of mass of the autonomous vehicle to the left side of the vehicle on the horizontal plane, perpendicular to the X-axis. The origin of the coordinate system is the position of the center of mass of the autonomous vehicle. The X-axis direction is the longitudinal direction of the vehicle, and the Y-axis direction is the lateral direction of the vehicle. Establish an autonomous vehicle longitudinal decision-making model, which includes two parts: a longitudinal system model and a longitudinal control model. The longitudinal system model is as follows: Where, x x For the longitudinal system state of the vehicle, x x =[x o v x ]′, ′ denotes matrix transpose; x o v represents the longitudinal displacement of the vehicle, in meters (m). x The longitudinal speed of the vehicle is expressed in m / s. u x For the longitudinal control of the vehicle, u x =a x ;a x The longitudinal acceleration of the vehicle, in m / s² 2 a x It is derived from the longitudinal control model: u x =f x (v,v x (2) Where, u x v is the longitudinal control variable for the vehicle; v is the vehicle speed decision value, in m / s; v x f is the longitudinal speed of the vehicle, in m / s. x (v,v x ) represents the longitudinal control law, f x (v,v x )=P(vv x ), where P is the proportional coefficient of the longitudinal control law. The proportional coefficient P of the longitudinal control law is determined using the engineering tuning method of the PID controller based on the vehicle's longitudinal system model. A lateral decision-making model for the autonomous vehicle is established, comprising three parts: a vehicle lateral system model, a lateral control model, and a lateral planning model. The vehicle lateral system model is as follows: Where y o v represents the lateral displacement of the vehicle, in meters (m). x V is the vehicle's longitudinal velocity, in m / s; ψ is the vehicle's yaw angle, in rad; v y ω is the vehicle's lateral velocity, in m / s; r is the vehicle's yaw rate, in rad / s; w f The distance from the center of the front axle to the center of gravity of the vehicle, in meters (m); w r The distance from the center of the rear axle to the center of mass of the vehicle, in meters (m); z The moment of inertia of the vehicle about a plane perpendicular to the X-axis and Y-axis, expressed in kg·m. 2 C f C represents the front wheel lateral stiffness, in N / rad. r Rear wheel lateral stiffness (N / rad); δ is the front wheel steering angle (rad); m is the vehicle mass (kg); The resulting state-space model of the vehicle's lateral system is: in x y For the vehicle's lateral system state, x y =[y o ψ rv y ]′;u y For vehicle lateral control, u y =δ; δ is the front wheel steering angle of the vehicle, in rad, derived from the lateral control model: δ = f y (R(l,T,x o ),x y )=R(l,T,x o )-Kx y (5) Where δ is the front wheel steering angle of the vehicle, in rad; f y (R(l,T,x o ),x y ) represents the lateral control law; x y Let R(l,T,x) represent the vehicle lateral system state; K is the lateral control gain vector, with a dimension of 1 row and 4 columns, obtained using the pole placement method for linear system controller design based on the vehicle lateral system state-space model; o R(l,T,x) is the reference lateral displacement, in meters, derived from the lateral programming model: o )=(1-w(t,T))y0(x o )+w(t,T)y l (x o (6) where y0(x o ) represents the longitudinal displacement x o Lateral displacement of the centerline of the current lane, equation of the centerline of the current lane a0 is the cubic coefficient of the current lane's centerline equation, b0 is the quadratic coefficient of the current lane's centerline equation, c0 is the linear coefficient of the current lane's centerline equation, and d0 is the zeroth-order coefficient of the current lane's centerline equation; y l (x o Let xo be the lateral displacement of the centerline of the target lane l at the longitudinal displacement xo, and let xo be the equation of the centerline of the target lane l. a l Let b be the cubic coefficient of the centerline equation of the target lane l. l c represents the quadratic coefficient of the centerline equation of the target lane l. l Let d be the first-order coefficient of the centerline equation of the target lane l. l Let l be the zeroth-order coefficient of the centerline equation of the target lane l, where l is the target lane number decision, and 1 ≤ l ≤ L. n And L is an integer. n The number of drivable lanes is an integer greater than 1; w(t,T) is the lane-changing weight, derived from the following formula: Among them, t s The execution cycle of the lateral planning model is denoted by T, in seconds; T represents the lane change time decision, in seconds; and t represents the current lane change time, derived from the following formula: Among them, y o Let d1 be the lateral displacement of the vehicle, in meters; d2 be the zero-order coefficient of the centerline equation of the current lane; d3 be the zero-order coefficient of the centerline equation of the target lane l; combining the vehicle's longitudinal system model and longitudinal control model, the autonomous vehicle decision model is obtained: in, x represents the state of the autonomous vehicle decision-making model, x = [x x x y ]′; y is the output of the autonomous vehicle decision model; ψ is the vehicle yaw angle, in rad; f x (v,v x ) represents the longitudinal control law; f y (R(l,T,x o ),x y The lateral control law is used; the decision model of the autonomous vehicle is discretized to obtain the discretized decision model of the autonomous vehicle: Where x(k) is the state x of the autonomous vehicle decision model at time k; x(k+1) is the state x of the autonomous vehicle decision model at time k+1; Let T be the derivative of the state x of the autonomous vehicle decision-making model at time k with respect to time; s y(k+1) represents the discrete sampling time of the model; y(k+1) is the output of the autonomous vehicle decision model at time k+1; Step 2: Establish the collision probability model between the vehicle and surrounding vehicles. The future positions of the surrounding vehicles output by the surrounding vehicle trajectory prediction module follow a two-dimensional Gaussian distribution, with the following probability density: Where, x i (k) represents the predicted longitudinal position of the i-th vehicle at time k, and is a random variable in meters; y i (k) represents the predicted lateral position of the i-th vehicle at time k, and is a random variable in meters (m); μ xi (k) represents the mean longitudinal predicted position of the i-th vehicle at time k; μ yi (k) represents the mean lateral predicted position of the i-th vehicle at time k; σ xi (k) represents the longitudinal predicted position variance of the i-th vehicle at time k; σ yi (k) represents the variance of the lateral predicted position of the i-th vehicle at time k; ρ i (k) is the longitudinal predicted position correlation coefficient of the i-th vehicle at time k; simplifying the vehicle as a rectangle, the coordinates of the four corners of the vehicle are calculated as follows: in, Let be the X-axis coordinate of the first angle of the circumferential vehicle i at time k, in meters. Let be the Y-axis coordinate of the first angle of vehicle i at time k, in meters; Let x be the X-axis coordinate of the second angle of the circumferential vehicle i at time k, in meters. Let be the Y-axis coordinate of the second angle of vehicle i at time k, in meters. Let be the X-axis coordinate of the first triangle of vehicle i at time k, in meters. Let be the Y-axis coordinate of the first triangle of vehicle i at time k, in meters. Let x be the X-axis coordinate of the fourth angle of the circumferential vehicle i at time k, in meters. Let μ be the Y-axis coordinate of the fourth angle of vehicle i at time k, in meters. xi (k) represents the mean longitudinal predicted position of the i-th vehicle at time k; μ yi (k) represents the mean of the lateral predicted position of the i-th vehicle at time k; vlen i The length of the diagonal of the circumference i is half the length, in meters. And ζ i Let be the angle between the line connecting the centroid of rook i and its first angle and the left side of rook i, in rad. len i Let wid be the length of vehicle i in meters. i Let θ be the width of vehicle i in meters; i (k) is the heading angle of vehicle i, in rad. The coordinates of the four corners of this vehicle are calculated as follows: Where, x 1 (k) represents the X-axis coordinate of the first angle of the vehicle at time k, in meters; y 1 (k) represents the Y-axis coordinate of the first angle of the vehicle at time k, in meters; x 2 (k) represents the X-axis coordinate of the second angle of the vehicle at time k, in meters; y 2 (k) represents the Y-axis coordinate of the second angle of the vehicle at time k, in meters; x 3 (k) represents the X-axis coordinate of the vehicle's first triangle at time k, in meters; y 3 (k) represents the Y-axis coordinate of the vehicle's first triangle at time k, in meters; x 4 (k) represents the X-axis coordinate of the fourth angle of the vehicle at time k, in meters; y 4 (k) represents the Y-axis coordinate of the fourth corner of the vehicle at time k, in meters; x o The longitudinal displacement of the vehicle is expressed in meters (m); y o Vlen represents the lateral displacement of the vehicle, in meters (m); vlen represents the half-length of the vehicle's diagonal, in meters (m). And ζ is the angle between the line connecting the centroid of the vehicle and its first corner, and the left side of the vehicle, in rad. len represents the vehicle length in meters (m), and wid represents the vehicle width in meters (m). The collision probability model between this vehicle and surrounding vehicles is as follows: Where, p i (k) represents the collision probability between this vehicle and surrounding vehicle i; x i (k) represents the predicted longitudinal position of the i-th vehicle at time k, and is a random variable in meters; y i (k) is the predicted lateral position of the i-th vehicle at time k, which is a random variable in m; Let be the lower bound of the collision probability between this vehicle and surrounding vehicle i over the X-axis of the coordinate system, in meters. Let be the upper limit of the integral of the collision probability between this vehicle and surrounding vehicle i along the X-axis of the coordinate system, in meters; Let be the lower bound of the collision probability between this vehicle and surrounding vehicle i over the Y-axis of the coordinate system, in meters. The upper bound of the integral of the collision probability between this vehicle and surrounding vehicle i along the Y-axis of the coordinate system, in meters: in, j1 = 1, 2, 3, 4 are the X-axis coordinates of the first, second, third, and fourth angles of the vehicle at time k, respectively. Let Y be the Y-axis coordinates of the first, second, third, and fourth angles of the vehicle at time k. Let X represent the X-axis coordinates of the first, second, third, and fourth angles of vehicle i at time k. Let Y and Y represent the Y-axis coordinates of the first, second, third, and fourth angles of vehicle i at time k, respectively; where, The symbol {} represents a set; Step 3: Design the rolling optimization decision optimization problem for unmanned vehicles considering the randomness of the surrounding vehicles, and establish the collision avoidance objective J. r1 : Among them, J r1 Indicates the target to be avoided; p i (k) represents the collision probability between this vehicle and surrounding vehicle i; P represents the prediction step size, which is an integer greater than 1; Σ represents the addition symbol; Π represents the multiplication symbol; n represents the number of surrounding vehicles; establish a fast target J s1 : Among them, J s1 Let v(k) represent the rapid target, where v(k) is the vehicle speed decision at time k, in m / s; N is the control step size, 1 ≤ N ≤ P and is an integer, where P represents the prediction step size; v ref The desired vehicle speed, in m / s, v ref =τv roadmax +(1-τ)v roadmin ;v roadmax The maximum speed limit for the road, in m / s; v roadmin The minimum speed limit for the road is given in m / s; τ is the speed factor, 0 ≤ τ ≤ 1; a lane keeping target J is established. s2 : Among them, J s2 The target lane keeping is represented by N; the control step size is N; Δl(k) is the lane change amount, Δl(k) = l(k) - l(k-1), and l(k) is the target lane number decision at time k; a comfortable lane changing target J is established. s3 : Among them, J s3 Represents the comfortable lane-changing objective; N is the control step size; T(k) is the lane-changing time decision at time k; establish lane constraints: 1≤l(k)≤L n ,l∈Z (20)l(k) is the target lane numbering decision at time k; L n Z represents the number of drivable lanes; Z represents the set of integers; establish lane-changing time constraints: T min ≤T(k)≤T max (21) where T(k) is the lane-changing time decision at time k; T min This is the lower limit for lane change time; T max Establish an upper limit for lane-changing time; set a constraint on speed change: Δv min ≤Δv(k)≤Δv max (22) Δv(k) is the change in velocity, in m / s, Δv(k) = v(k) - Δv(k-1), where v(k) is the vehicle speed decision at time k; Δv min The lower limit of the change in velocity, in m / s, Δv max The upper limit of velocity change, in m / s; establish velocity constraint: 0 ≤ v(k) ≤ v roadmax (23) where v(k) is the vehicle speed decision at time k; v roadmax The maximum speed limit for the road is set; a model predictive control method is adopted, based on a discretized decision model of the autonomous vehicle to predict the future state of the system, and the vehicle speed decision control sequence v is defined. V =[v(1) v(1) … v(N)], target lane numbering control sequence l V =[l(1) …l(N)], Lane change time control sequence T V = [T(1) … T(N)], where N is the control time domain of the model predictive control method, v(1) is the vehicle speed decision at time 1, v(2) is the vehicle speed decision at time 2, v(N) is the vehicle speed decision at time N, l(1) is the target lane number decision at time 1, l(2) is the target lane number decision at time 2, l(N) is the target lane number decision at time N, T(1) is the lane change time decision at time 1, T(2) is the lane change time decision at time 2, and T(N) is the lane change time decision at time N; By weighting the collision avoidance target, the speed target, the lane keeping target, and the comfortable lane change target, and combining the lane constraints, lane change time constraints, speed change constraints, and speed constraints, we obtain the unmanned vehicle rolling optimization decision optimization problem considering the randomness of the surrounding vehicles: Among them, v V =[v(1) v(1) … v(N)] represents the vehicle speed decision control sequence, l V =[l(1) … l(N)] is the target lane numbering control sequence, T V =[T(1) … T(N)] is the lane-changing time control sequence, and J is the objective function for the rolling optimization decision of the unmanned vehicle considering the randomness of the weekly vehicle; J r1 To avoid collision with the target; C is the danger deviation constant, C > 1; p i (k) represents the collision probability between this vehicle and surrounding vehicle i; ε is the collision risk tolerance factor, 0 < ε < 1; J s1 For the rapid objective; J s2 To maintain lane target; J s3 For the goal of comfortable lane changing; Γ s1 For fast target weights; Γ s2 To maintain the target weight for lane keeping; Γ s3 Weighting of the target for comfortable lane changing; Γ r1 To avoid collision target weights; J smin For safety objective J s The minimum value of J smax For safety objective J s The maximum value of J s For safety objectives: J s =Γ s1 J s1 +Γ s2 J s2 +Γ s3 J s3 +Γ r1 J r1 (24) Among them, J r1 To avoid colliding with the target; J s1 For the rapid objective; J s2 To maintain lane target; J s3 For the goal of comfortable lane changing; Γ s1 For fast target weights; Γ s2 To maintain the target weight for lane keeping; Γ s3 Weighting of the target for comfortable lane changing; Γ r1 To determine the target weights for collision avoidance; the vehicle speed decision control sequence v V =[v(1) v(1) … v(N)], target lane numbering control sequence l V = [l(1) … l(N)] and lane change time control sequence T V =[T(1) … T(N)] are used together as variables to be optimized. A genetic algorithm is used to solve the rolling optimization decision problem of an autonomous vehicle that considers the randomness of the vehicle cycle. The vehicle speed decision control sequence v that minimizes the objective function J of the rolling optimization decision of the autonomous vehicle is obtained. V =[v(1) v(1) … v(N)], target lane numbering control sequence l V = [l(1) … l(N)] and lane change time control sequence T V =[T(1) … T(N)] is defined as the optimal control sequence in The optimal vehicle speed control sequence; The optimal target lane numbering control sequence; This is the optimal lane-changing time control sequence. * The variable represents the optimal state; when the state of the autonomous vehicle rolling optimization decision system is collected and updated in real time, the optimal control sequence is obtained by solving the autonomous vehicle rolling optimization decision optimization problem considering the randomness of the vehicle cycle. And the optimal vehicle speed control sequence The first element Through the vertical control law f x (v,v x Converted to vehicle longitudinal acceleration a x Optimal target lane number decision vector The first element Optimal lane-changing time decision vector The first element Through the lateral control law f y (R(l,T,x o ),x y This is converted into the vehicle's front wheel steering angle δ to achieve longitudinal and lateral control of the vehicle.

Images