A vehicle collision risk model suitable for extreme bend scenario obstacle avoidance

Abstract

This invention relates to a collision risk model based on model predictive control (MMC) applicable to obstacle avoidance in extreme cornering scenarios for automobiles. It can be used in both hierarchical obstacle avoidance architectures and integrated obstacle avoidance and control architectures. First, a collision risk model centered on the driver vehicle (VV) and based on the relative poses of the VV and obstacle vehicles is constructed in the Flyner coordinate system, incorporating road alignment features and road surface conditions. Next, based on the relative poses of the VV and obstacle vehicles, the expression for the collision risk model is determined as part of the cost function of the obstacle avoidance optimization algorithm. The poses of the obstacle vehicles are estimated using a kinematic model. When the model is used in a hierarchical architecture, the VV's pose is based on the optimal VV pose estimate obtained by the optimization algorithm in the previous obstacle avoidance process. When the model is used in an integrated control architecture, the VV's pose is determined by combining the optimal control quantity from the previous obstacle avoidance process with a multi-curvature augmented discrete dynamics model, further incorporating road alignment features through multi-curvature augmentation. This model can achieve collision risk prediction in extreme automotive scenarios, promoting the development of automotive-related research.

Description

Technical Field

[0001] This invention belongs to the field of vehicle control technology, specifically relating to a modeling method for collision risk models in extreme automotive scenarios. Background Technology

[0002] Obstacle avoidance in intelligent vehicles is closely related to the traffic environment. One of the key focuses of obstacle avoidance research is determining the collision risk between the driver vehicle and obstacle vehicles, and optimizing the driver vehicle's pose (obstacle avoidance hierarchical architecture) or vehicle input control variables (obstacle avoidance and decision control integrated architecture) based on the principle of minimizing collision risk. Factors influencing collision risk include the motion characteristics of obstacles, road surface conditions, and road alignment. Currently, two main modeling methods are used: one is the safety envelope method, which determines the driver vehicle's drivable area based on the obstacle's position and transforms it into a constraint for the optimization algorithm. This method is only applicable to static obstacles and does not consider road conditions. The other is the APF method, which guides the driver vehicle to avoid obstacles through the combined action of the gravitational field of the target point and the repulsive field of the obstacle. Commonly used methods include the obstacle's virtual rectangular repulsive field and the safety ellipse theory. These methods achieve good results on straight roads, but under curved conditions, they cannot reflect the influence of road geometry on the relative position between the driver vehicle and obstacles, and the road surface conditions are relatively simple. Furthermore, the above collision risk models are all constructed with the obstacle as the center, increasing modeling complexity in multi-obstacle vehicle scenarios. Summary of the Invention

[0003] The purpose of this invention is to design a collision risk model based on model predictive control, suitable for obstacle avoidance in extreme cornering scenarios for automobiles, and to use it in conjunction with an obstacle avoidance optimization control algorithm. This model can be used in hierarchical obstacle avoidance architectures or integrated decision-making and control architectures. The collision risk model provides the optimization target for the obstacle avoidance algorithm, which optimizes the control variables to minimize the collision risk between the main vehicle and the obstacle vehicle.

[0004] The modeling process of the collision risk model includes, firstly, constructing a collision risk model centered on the main vehicle and based on the relative pose relationship between the main vehicle and the obstacle vehicle in the Flexner coordinate system, incorporating road alignment features and road surface conditions. Next, based on the relative poses of the main vehicle and the obstacle vehicle, the expression of the collision risk model is determined for use in the obstacle avoidance optimization algorithm. The pose of the obstacle vehicle is estimated based on the kinematic model. When the model is used in a hierarchical architecture, the main vehicle pose is estimated using the optimal main vehicle pose obtained by the optimization algorithm in the previous obstacle avoidance process. When the model is used in a decision-making and control integrated architecture, since the optimal quantity obtained by the obstacle avoidance optimization control algorithm in the previous obstacle avoidance process is the vehicle control input (in this case, the front wheel steering angle), determining the main vehicle pose in the previous obstacle avoidance process requires two steps: first, designing a multi-curvature augmented discrete dynamics model based on a single-track two-degree-of-freedom dynamics model of the vehicle using model predictive control; second, determining the main vehicle pose by combining the optimal control quantity from the previous obstacle avoidance process and the multi-curvature augmented discrete dynamics model of the vehicle. Through algorithmic innovation, this model can predict collision risks in obstacle avoidance scenarios for automobiles in extreme curves.

[0005] Furthermore, the control quantity optimized by the integrated decision-making and control algorithm is the vehicle control input, not the vehicle pose. Therefore, the risk model expression needs to be transformed into a function of the vehicle control input to adapt to the optimization algorithm.

[0006] Furthermore, the extreme curve scenario has the following three characteristics: curves with large curvature or abrupt curvature changes, low tire-to-ground friction coefficient or abrupt changes, and the presence of multiple obstacles, including dynamic and static ones.

[0007] Furthermore, the hierarchical architecture decomposes the autonomous decision-making, trajectory planning, and trajectory tracking tasks of the obstacle avoidance process.

[0008] Furthermore, the integrated obstacle avoidance and control architecture integrates the above three sequential tasks into a constrained optimal control problem, realizing a direct mapping between traffic environment information and vehicle control.

[0009] Specifically, a collision risk model for obstacle avoidance in extreme cornering scenarios for automobiles, when used in a layered architecture, includes the following steps:

[0010] Step S1: In the Flyner coordinate system, construct a collision risk model centered on the main vehicle and based on the relative pose relationship between the main vehicle and the obstacle vehicle, incorporating road alignment features and road surface conditions.

[0011] Step S2: Estimate the obstacle vehicle pose and the driver vehicle pose in the prediction time domain to obtain the collision risk model expression;

[0012] Furthermore, it is assumed that the relative tangential velocities of the main vehicle and the obstacle vehicle are predicted in the time domain. Keeping the parameters constant, at time k, construct a collision risk model in the prediction time domain centered on the main vehicle in the Fleiner coordinate system, based on the relative poses of the two vehicles, as follows:

[0013]

[0014] In equation (1), The collision risk between the main vehicle and the i-th surrounding obstacle vehicle in the prediction time domain, i = 1, ..., N Obs j = 1, 2, ..., N P N Obs The number of obstructing vehicles around the main vehicle, N P η is the prediction step size for the model predictive control algorithm. CA s is the adjustment coefficient. rel,i (.), l rel,i (.), d represents the relative tangential position, normal position, and distance between the main vehicle and the i-th obstacle vehicle in the Flyner coordinate system. l,i The lateral safety distance between the main vehicle and the i-th obstacle vehicle. d B,i These are two longitudinal safety distance limits for the main vehicle after the i-th obstacle, where it avoids the obstacle by braking or changing lanes. Considering the obstacle avoidance safety after the main vehicle passes the i-th obstacle, a collision risk zone is set after overtaking, resulting in a longitudinal safety distance limit d. A,i .

[0015] Furthermore, by constructing the relative pose in the Flyner coordinate system, the road alignment features were incorporated.

[0016] Furthermore, d A,i Set to a fixed value of 5m, d B,i , d l,i The expression is as follows:

[0017]

[0018] In equation (2), λ B , λ s , λ l This is the magnification factor, which can be set to a fixed value, l CA The lateral displacement required for the main vehicle to complete obstacle avoidance is taken here as one lane width, i.e., 3.75m. These are the average relative tangential acceleration and normal acceleration for longitudinal braking obstacle avoidance and lateral lane-changing obstacle avoidance, respectively. d0 is the basic safety distance, and w is the vehicle width.

[0019] Furthermore, s rel,i l rel,i , and It can be represented as:

[0020]

[0021] In equation (3), s H l H , These represent the tangential and normal positions of the main vehicle and the i-th obstacle vehicle, respectively. Let be the tangential velocities of the main vehicle and the i-th obstacle vehicle, respectively, and assume that they remain constant in the prediction time domain.

[0022] Furthermore, the vehicle was traveling on a curve. It is mainly constrained by the tire's coefficient of friction with the ground, the friction circle, and the braking force of the braking system. The selection of the friction circle mainly considers the requirements of vehicle stability and the limitations of the friction circle. Therefore, it can be done through... and Incorporating the route table conditions, the expression is:

[0023]

[0024] In equation (4), For safety adjustment factors, they are set to 0.75 and 0.5 respectively, μ is the tire-to-ground friction coefficient, and g is the acceleration due to gravity, with a value of 9.8 m / s². 2 .

[0025] Furthermore, the determination of the collision risk model expression in equation (1) needs to be based on the driver vehicle's position pose s H l H Meanwhile, the collision risk model expression also includes the position and orientation of the main vehicle, resulting in an algebraic loop contradiction.

[0026] Furthermore, when the collision risk model is used in a layered architecture, it is used respectively... Substitution formula (1) judgment condition s rel,i l rel,i This solves the algebraic ring problem. That is:

[0027]

[0028] In equation (5), i = 1, ..., N Obs j = 1, 2, ..., N P , s H,pre (.) and l H,pre (.) represents the estimated vehicle pose in the prediction time domain, distinct from the vehicle pose s to be optimized. H and l H .

[0029] Furthermore, step S2 can be subdivided into the following steps:

[0030] Step S2.1: Construct the kinematic models of the main vehicle and the obstacle vehicle at time k in the prediction time domain, i.e.:

[0031]

[0032] In equation (6), Let j = 1, 2, ..., N, be the normal velocities of the main vehicle and the i-th obstacle vehicle at time k, and assume they remain constant in the prediction time domain. P ;

[0033] Step S2.2: Determine the pose of the obstacle vehicle in the prediction time domain according to the kinematic model of equation (6).

[0034] Step S2.3: Determine the position and pose of the main vehicle. H,Pre (.), l H,Pre (.). When the obstacle avoidance optimization algorithm is run for the first time, s is determined according to the kinematic model of equation (6). H,Pre (.), l H,Pre (.); When the optimization algorithm is not being run for the first time, the planning layer using the obstacle avoidance hierarchical architecture uses the vehicle pose obtained from the previous time step as s. H,Pre (.), l H,Pre (.);

[0035] Step S2.4: s H,Pre (.), l H,Pre (.), s i Obs (.), l i Obs (.), Substituting into equation (5), we can obtain the expression for the collision risk model, which can be used to solve the obstacle avoidance optimization problem.

[0036] Furthermore, when the collision risk model is used in an integrated obstacle avoidance and control system, the control quantity obtained by the obstacle avoidance algorithm optimization is not the main vehicle's pose, but the control quantity input by the vehicle.

[0037] Furthermore, the collision risk model is used in the obstacle avoidance and control integrated architecture. When the obstacle avoidance optimization algorithm is run for the first time, the main vehicle's position s is determined according to the kinematic model of equation (6). H,Pre (.), l H,Pre (.).

[0038] Furthermore, the collision risk model is used in the integrated obstacle avoidance and control architecture. When the optimization algorithm is not being run for the first time, the optimal control quantity obtained from the optimization is converted into the main vehicle pose s based on the vehicle dynamics model in the prediction time domain.H,Pre (.), l H,Pre (.).

[0039] Furthermore, the front wheel steering angle was selected as the optimized control variable for the study of lateral obstacle avoidance of the vehicle.

[0040] Furthermore, in the prediction time domain, based on the vehicle dynamics model, the optimized control quantity is converted into the main vehicle pose s. H,Pre (.), l H,Pre (.), needs to be broken down into the following two steps:

[0041] Step S2.3.1: Based on the two-degree-of-freedom dynamic model of the vehicle single track based on the tracking deviation, a multi-curvature augmented discrete dynamic model is designed using the model predictive control method;

[0042] Step S2.3.2: Substitute the optimal control quantity obtained from the previous obstacle avoidance and control integrated system into the multi-curvature augmented discrete dynamics model to obtain the vehicle's pose s in the predicted time domain. H,Pre (.), l H,Pre (.);

[0043] Furthermore, the designed multi-curvature augmented discrete dynamics model is specifically as follows:

[0044]

[0045] In equation (7), e(k+i|k) is the output of the prediction at the k-th step, where i = 1, 2, ..., N P u(k+j|k) is the control variable calculated in the k-th step, j = 0, 1, ..., N C -1, N C Let N be the control step size for model predictive control, and N be the control step size for model predictive control. C ≤N P , Let Ψ be the state variable at step k. The expressions for Ψ and Θ are as follows:

[0046]

[0047] Furthermore, step S2.3.1 is further subdivided into the following steps:

[0048] Step S2.3.1.1: Construct a single-track dynamic model of the vehicle, that is, construct a model in the Flyner coordinate system with the vehicle's normal position l. H Normal velocity Yaw angle deviation and the rate of change of yaw angle deviation As a state variable, the vehicle's front wheel steering angle δ f The single-track two-degree-of-freedom dynamic model, which serves as the control variable, is also known as the continuous state-space equation:

[0049]

[0050] In equation (9), κ is the curvature of the reference trajectory. C αf C αr Let m be the lateral stiffness of a single tire on the front and rear wheels of the vehicle, and v be the vehicle mass. x Let be the longitudinal velocity of the vehicle, a and b be the distances from the front and rear axles to the center of mass of the vehicle, and I be the moment of inertia of the vehicle about the z-axis.

[0051] Step S2.3.1.2: Discretization of the continuous state-space equation. The continuous state-space equation of equation (9) is converted into a discrete equation with a fixed sampling frequency:

[0052]

[0053] In equation (10), T is the sampling time;

[0054] Step S2.3.1.3: Design a discrete state-space equation with reference trajectory curvature augmentation, incorporating road alignment features. Augment the reference trajectory curvature of equation (10) to the state variables, using the front wheel steering angle as the control variable, and construct the discrete state-space equation. Augment the future N in the k-th step. P The curvature of the step, and the state variables at this time are:

[0055]

[0056] The corresponding state-space equation is:

[0057]

[0058] In equation (12),

[0059] Step S2.3.1.4: Construct a prediction model. Using the model recursion of equation (12), predict the future output based on the current state variables of the controlled object and the future control variables, and obtain equation (7).

[0060] Furthermore, step S2.3.2 is further subdivided into the following steps:

[0061] Step S2.3.2.1: Substitute the optimal front wheel steering angle obtained in the previous round of optimization of the obstacle avoidance and control integrated system into equation (7) to obtain... In the collision risk model of equation (5), l H,Pre (.);

[0062] Step S2.3.2.2: [The sentence is incomplete and requires more context to be translated accurately.] Substitute (13) Obtain s in the collision risk model of equation (5) H,Pre (.).

[0063]

[0064] Let J represent the initial tangential position and velocity vector of the main vehicle at time k, where j = 1, 2, ..., N. P .

[0065] Furthermore, when the model is used in an integrated obstacle avoidance and control architecture, equations (7) and (13) need to be substituted into the model expression of equation (5) to convert the expression into a function of the vehicle control quantity to be optimized, so as to adapt to the optimization algorithm.

[0066] Furthermore, by directly incorporating road alignment features and road surface conditions into Equation (1), a multi-curvature augmented dynamic model of Equation (7) is designed to indirectly incorporate road alignment features, ensuring that the collision risk model is applicable to extreme curve scenarios.

[0067] Instruction manual illustrations

[0068] Figure 1 This is a schematic diagram of the process steps for using the layered architecture of the present invention;

[0069] Figure 2 This is a schematic diagram of the process steps for determining the pose of the main vehicle (step S2.3) when the obstacle avoidance optimization algorithm is run for the first time in the integrated decision-making and control architecture of the present invention.

[0070] Figure 3 This is a schematic diagram of a scenario according to an embodiment of the present invention. Detailed Implementation

[0071] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0072] Please see Figure 1 and Figure 2The purpose of this invention is to design a collision risk model based on model predictive control (MMC) suitable for obstacle avoidance in extreme cornering scenarios, and to combine it with an obstacle avoidance optimization control algorithm. This model can be used in a hierarchical architecture or an integrated obstacle avoidance and control architecture. The collision risk model provides the optimization target for the obstacle avoidance algorithm, which optimizes the control variables to minimize the collision risk between the vehicle and the obstacle vehicle. The extreme cornering scenario has the following three characteristics: large curvature or abrupt curvature curves, low or abrupt tire-to-ground friction coefficients, and the presence of multiple obstacles, including both dynamic and static ones. The hierarchical architecture decomposes the autonomous decision-making, trajectory planning, and trajectory tracking tasks of the obstacle avoidance process. The integrated obstacle avoidance and control architecture integrates these three sequential tasks into a constrained optimal control problem, achieving a direct mapping between traffic environment information and vehicle control.

[0073] The modeling process of the collision risk model includes, firstly, constructing a collision risk model centered on the driver vehicle and based on the relative pose relationship between the driver vehicle and the obstacle vehicle in the FRINAL coordinate system, and incorporating road alignment features and road surface conditions into the model; nextly, estimating the poses of the obstacle vehicle and the driver vehicle in the prediction time domain to obtain the expression of the collision risk model.

[0074] The obstacle vehicle pose is estimated based on a kinematic model. When used in a hierarchical architecture, the main vehicle pose is estimated based on the optimal main vehicle pose obtained by the obstacle avoidance optimization control algorithm in the previous obstacle avoidance process. When the model is used in a decision-making and control integrated architecture, since the optimal quantity obtained by the obstacle avoidance optimization control algorithm in the previous obstacle avoidance process is the vehicle control input (in this case, the front wheel steering angle), determining the main vehicle pose in the previous obstacle avoidance process requires two steps: first, based on the vehicle's single-track two-degree-of-freedom dynamics model with tracking deviation, a multi-curvature augmented discrete dynamics model is designed using model predictive control; second, the optimal control quantity obtained by the decision-making and control integrated system in the previous obstacle avoidance process is input into the multi-curvature augmented discrete dynamics model to obtain the main vehicle pose in the prediction time domain. Furthermore, since the optimization quantity of the decision-making and control integrated algorithm is the vehicle control input, the risk model expression needs to be transformed into a function of the vehicle control input to adapt to the optimization algorithm.

[0075] Through algorithmic innovation, this model can predict collision risks in obstacle avoidance scenarios for automobiles in extreme curves.

[0076] Specifically, a collision risk model for obstacle avoidance in extreme cornering scenarios for automobiles, when used in a layered architecture, includes the following steps:

[0077] Step S1: In the Flyner coordinate system, construct a collision risk model centered on the main vehicle and based on the relative pose relationship between the main vehicle and the obstacle vehicle, incorporating road alignment features and road surface conditions.

[0078] Step S2: Estimate the obstacle vehicle pose and the driver vehicle pose in the prediction time domain to obtain the collision risk model expression;

[0079] Assuming the relative tangential velocities of the driver vehicle and the obstacle vehicle in the prediction time domain... Keeping the parameters constant, at time k, construct a collision risk model in the prediction time domain centered on the main vehicle in the Fleiner coordinate system, based on the relative poses of the two vehicles, as follows:

[0080]

[0081] In equation (1), The collision risk between the main vehicle and the i-th surrounding obstacle vehicle in the prediction time domain, i = 1, ..., N Obs j = 1, 2, ..., N P N Obs The number of obstructing vehicles around the main vehicle, N P η is the prediction step size for the model predictive control algorithm. CA s is the adjustment coefficient. rel,i (.), l rel,i (.), d represents the relative tangential position, normal position, and distance between the main vehicle and the i-th obstacle vehicle in the Flyner coordinate system. l,i The lateral safety distance between the main vehicle and the i-th obstacle vehicle. d B,i These are two longitudinal safety distance limits for the main vehicle after the i-th obstacle, where it avoids the obstacle by braking or changing lanes. Considering the obstacle avoidance safety after the main vehicle passes the i-th obstacle, a collision risk zone is set after overtaking, resulting in a longitudinal safety distance limit d. A,i .

[0082] Furthermore, by constructing the relative pose in the Flyner coordinate system, the road alignment features were incorporated.

[0083] Furthermore, d A,i Set to a fixed value of 5m, d B,i , d l,i The expression is as follows:

[0084]

[0085] In equation (2), λ B , λ s , λ l This is the magnification factor, which can be set to a fixed value, l CA The lateral displacement required for the main vehicle to complete obstacle avoidance is taken here as one lane width, i.e., 3.75m. These are the average relative tangential acceleration and normal acceleration for longitudinal braking obstacle avoidance and lateral lane-changing obstacle avoidance, respectively; d0 is the basic safety distance; and w is the vehicle width.

[0086] Furthermore, s rel,i l rel,i , and It can be represented as:

[0087]

[0088] In equation (3), s H l H , These represent the tangential and normal positions of the main vehicle and the i-th obstacle vehicle, respectively. Let be the tangential velocities of the main vehicle and the i-th obstacle vehicle, respectively, and assume that they remain constant in the prediction time domain.

[0089] Furthermore, the vehicle was traveling on a curve. It is mainly constrained by the tire's coefficient of friction with the ground, the friction circle, and the braking force of the braking system. The selection of the friction circle mainly considers the requirements of vehicle stability and the limitations of the friction circle. Therefore, it can be done through... and Incorporating route table conditions, the expression is as follows:

[0090]

[0091] In equation (4), For safety adjustment factors, they are set to 0.75 and 0.5 respectively, μ is the tire-to-ground friction coefficient, and g is the acceleration due to gravity, with a value of 9.8 m / s². 2 .

[0092] Furthermore, the determination of the collision risk model expression in equation (1) needs to be based on the driver vehicle's position pose s H l H Meanwhile, the collision risk model expression also includes the position and orientation of the main vehicle, resulting in an algebraic loop contradiction.

[0093] Furthermore, respectively using Substitution formula (1) judgment condition s rel,i l rel,i This resolves the contradiction of the algebraic ring. That is:

[0094]

[0095] In equation (5), i = 1, ..., N Obs j = 1, 2, ..., N P , sH,pre (.) and l H,pre (.) represents the estimated vehicle pose in the prediction time domain, distinct from the vehicle pose s to be optimized. H and l H .

[0096] Furthermore, step S2 can be subdivided into the following steps:

[0097] Step S2.1: Construct the kinematic models of the main vehicle and the obstacle vehicle at time k in the prediction time domain, i.e.:

[0098]

[0099] In equation (6), Let j = 1, 2, ..., N, be the normal velocities of the main vehicle and the i-th obstacle vehicle at time k, and assume they remain constant in the prediction time domain. P ;

[0100] Step S2.2: Determine the pose of the obstacle vehicle in the prediction time domain according to the kinematic model of equation (6).

[0101] Step S2.3: Determine the position and pose of the main vehicle. H,Pre (.), l H,Pre (.). When the obstacle avoidance optimization algorithm is run for the first time, the main vehicle pose is determined according to the kinematic model of equation (6); when it is not the first time the optimization algorithm is run, the main vehicle pose obtained by the planning layer of the obstacle avoidance hierarchical architecture in the previous time step is used as s. H,Pre (.), l H,Pre (.);

[0102] Step S2.4: s H,Pre (.), l H,Pre (.), Substituting into equation (5), we can obtain the expression for the collision risk model, which can be used to solve the obstacle avoidance optimization problem.

[0103] Furthermore, when the collision risk model is used in an integrated obstacle avoidance and control system, the control quantity obtained by the obstacle avoidance algorithm optimization is not the main vehicle's pose, but the control quantity input by the vehicle.

[0104] Furthermore, the collision risk model is used in the obstacle avoidance and control integrated architecture. When the obstacle avoidance optimization algorithm is run for the first time, the main vehicle's position s is determined according to the kinematic model of equation (6). H,Pre (.), l H,Pre (.).

[0105] Furthermore, the collision risk model is used in the integrated obstacle avoidance and control architecture. When the optimization algorithm is not being run for the first time, the optimal control quantity obtained from the optimization is converted into the main vehicle pose s based on the vehicle dynamics model in the prediction time domain.H,Pre (.), l H,Pre (.).

[0106] Furthermore, the front wheel steering angle was selected as the optimized control variable for the study of lateral obstacle avoidance of the vehicle.

[0107] Furthermore, in the prediction time domain, based on the vehicle dynamics model, the optimized control quantity is converted into the main vehicle pose s. H,Pre (.), l H,Pre (.), needs to be broken down into the following two steps:

[0108] Step S2.3.1: Based on the two-degree-of-freedom dynamic model of the vehicle single track based on the tracking deviation, a multi-curvature augmented discrete dynamic model is designed using the model predictive control method;

[0109] Step S2.3.2: Substitute the optimal control quantity obtained from the previous obstacle avoidance and control integrated system into the multi-curvature augmented discrete dynamics model to obtain the vehicle's pose s in the predicted time domain. H,Pre (.), l H,Pre (.);

[0110] Furthermore, the designed multi-curvature augmented discrete dynamics model is specifically as follows:

[0111]

[0112] In equation (7), e(k+i|k) is the output of the prediction at the k-th step, where i = 1, 2, ..., N P u(k+j|k) is the control variable calculated in the k-th step, j = 0, 1, ..., N C -1, N C Let N be the control step size for model predictive control, and N be the control step size for model predictive control. C ≤N P , Let Ψ be the state variable at step k. The expressions for Ψ and Θ are as follows:

[0113]

[0114] Furthermore, step S2.3.1 is further subdivided into the following steps:

[0115] Step S2.3.1.1: Construct a single-track dynamic model of the vehicle, that is, construct a model in the Flyner coordinate system with the vehicle's normal position l. H Normal velocity Yaw angle deviation and the rate of change of yaw angle deviation As a state variable, the vehicle's front wheel steering angle δ f The single-track two-degree-of-freedom dynamic model, which serves as the control variable, is also known as the continuous state-space equation:

[0116]

[0117] In equation (9), u = δ f κ is the curvature of the reference trajectory.

[0118] C αf C αr Let m be the lateral stiffness of a single tire on the front and rear wheels of the vehicle, and v be the vehicle mass. x Let be the longitudinal velocity of the vehicle, a and b be the distances from the front and rear axles to the center of mass of the vehicle, and I be the moment of inertia of the vehicle about the z-axis.

[0119] Step S2.3.1.2: Discretization of the continuous state-space equation. The continuous state-space equation of equation (9) is converted into a discrete equation with a fixed sampling frequency:

[0120]

[0121] In equation (10), T is the sampling time;

[0122] Step S2.3.1.3: Design a discrete state-space equation with reference trajectory curvature augmentation, incorporating road alignment features. Augment the reference trajectory curvature of equation (10) to the state variables, using the front wheel steering angle as the control variable, and construct the discrete state-space equation. Augment the future N in the k-th step. P The curvature of the step, and the state variables at this time are:

[0123]

[0124] The corresponding state-space equation is:

[0125]

[0126] In equation (12),

[0127] Step S2.3.1.4: Construct a prediction model. Using the model recursion of equation (12), predict the future output based on the current state variables of the controlled object and the future control variables, and obtain equation (7).

[0128] Furthermore, step S2.3.2 is further subdivided into the following steps:

[0129] Step S2.3.2.1: Substitute the optimal front wheel steering angle obtained in the previous round of optimization of the obstacle avoidance and control integrated system into equation (7) to obtain... In the collision risk model of equation (5), l H,Pre (.);

[0130] Step S2.3.2.2: [The sentence is incomplete and requires more context to be translated accurately.] Substitute (13) Obtain s in the collision risk model of equation (5) H,Pre (.).

[0131]

[0132] Let J represent the initial tangential position and velocity vector of the main vehicle at time k, where j = 1, 2, ..., N. P .

[0133] Furthermore, when the model is used in an integrated obstacle avoidance and control architecture, equations (7) and (13) need to be substituted into the model expression of equation (5) to convert the expression into a function of the vehicle control quantity to be optimized, so as to adapt to the optimization algorithm.

[0134] Furthermore, by directly incorporating road alignment features and road surface conditions into Equation (1), a multi-curvature augmented dynamic model of Equation (7) is designed to indirectly incorporate road alignment features, ensuring that the collision risk model is applicable to extreme curve scenarios.

[0135] Example

[0136] Determining the collision risk model expression for an integrated obstacle avoidance and decision-making control architecture is complex, especially when the expression is not determined for the first time. Therefore, this study focuses on determining the model expression under the integrated decision-making control architecture, selecting one obstacle vehicle around the main vehicle. The risk model expression is determined by estimating the poses of both the main vehicle and the obstacle vehicle. The relevant parameters of the main vehicle used in the calculation are: vehicle mass m = 2020 kg, distance a from the front axle to the center of mass = 1.265 m, distance b from the rear axle to the center of mass = 1.682 m, and moment of inertia I about the z-axis = 4095 kg / m. 2 Single front wheel lateral stiffness C αf The single rear wheel lateral stiffness C is -66900 N / rad. αr The value is -62700 N / rad, the ground adhesion coefficient μ is 0.8, and the sampling time T is 0.2 s, as shown in Table 1.

[0137] Please refer to the traffic scene layout. Figure 3Both the main vehicle and the obstacle vehicle are traveling on an S-curve. Using the center line of the right lane as the s-coordinate in the Frenet coordinate system, the initial position of the main vehicle is at the beginning of the S-curve, with its tangential and normal coordinates at (125m, 0m). The initial position of the obstacle vehicle is in front of the main vehicle, with its tangential and normal coordinates at (180m, 0m). Assume k = 0. To reduce computational load, assume the model predictive control algorithm has a control step size of 2 and a prediction step size of 3. Furthermore, assume that the speeds of both the main vehicle and the obstacle vehicle remain constant during the prediction time domain, with tangential speeds of 30m / s and 20m / s respectively, and normal speeds of 0m / s for both.

[0138] Table 1 Vehicle Parameters

[0139]

[0140] The detailed calculation steps are as follows:

[0141] 1. Calculate using equation (4) and

[0142]

[0143] 2. Calculate using equation (3)

[0144]

[0145] 3. Determine the three safety distances using equation (2).

[0146] At this point, d0 = 5.625m, λ B =λ s =λ l =1.3, w=2.5m.

[0147]

[0148] d l,i =λ l w = 3.25m

[0149] 4. Determine the obstacle vehicle position and pose in the prediction time domain according to equation (6).

[0150]

[0151] 5. Determine the vehicle pose and collision risk model expression in the prediction time domain; 1) Determine the collision risk model expression for the first time;

[0152] ① Determine s according to equation (6) H,pre (.), l H,pre (.);

[0153]

[0154] l H,pre (1) = l H,pre (2) = l H,pre (3)=0m

[0155] ② Calculate according to formula (3)

[0156]

[0157] ③ Determine the expression for the collision risk model based on equation (5);

[0158]

[0159] If the prediction time domain is extended to 9, then at the end of the prediction time domain we have:

[0160]

[0161] The expression for the collision risk model is:

[0162]

[0163] 2) If this is not the first time the collision risk model expression has been determined;

[0164] ① Calculate Ψ and Θ according to equation (8);

[0165]

[0166] ② Determine the initial state variables;

[0167] For simplicity of calculation, we assume here that at the initial moment, the main vehicle does not detect any obstacle vehicles and actively performs trajectory tracking, then:

[0168]

[0169] ③ Combining the curvature of the road alignment at s = 125, the augmented state quantity is obtained according to equation (11).

[0170]

[0171] ④ Determine the optimal set of control variables for the previous round of obstacle avoidance and integrated control algorithm;

[0172] The optimal control quantity needs to be determined in conjunction with the optimization control algorithm. Here, we assume that:

[0173] U = [0.0013 0.0032] T

[0174] ⑤ Ψ, Θ, U is substituted into equation (7) to determine and l H,Pre (.);

[0175]

[0176] [l H,pre (1) l H,pre (2) l H,pre [(3)] = [0.0205 0.0931 0.0241]

[0177] ⑥ Substitute into equation (13) to determine s H,Pre (.);

[0178] [s H,pre (1) s H,pre (2) s H,pre (3)]=[131 137 143]

[0179] ⑦ Calculate according to formula (3)

[0180]

[0181] ⑧ Determine the expression for the collision risk model based on equation (5).

[0182]

[0183] The present invention has been described in detail with reference to the foregoing embodiments. Those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A collision risk model for obstacle avoidance in extreme cornering scenarios of automobiles, characterized in that: The collision risk model is used in conjunction with the obstacle avoidance optimization control algorithm, and is suitable for layered architectures. The collision risk model provides the optimization target for the obstacle avoidance algorithm. The obstacle avoidance algorithm obtains the optimal pose of the main vehicle through optimization, ensuring that the collision risk between the main vehicle and the obstacle vehicle is minimized. The modeling includes the following steps: Step S1: In the Flyner coordinate system, construct a collision risk model centered on the main vehicle and based on the relative pose relationship between the main vehicle and the obstacle vehicle, incorporating road alignment features and road surface conditions. Step S2: Estimate the obstacle vehicle pose and the driver vehicle pose in the prediction time domain to obtain the collision risk model expression; Assuming the relative tangential velocities of the driver vehicle and the obstacle vehicle in the prediction time domain... Remain unchanged, in At any given time, a collision risk model is constructed in the prediction time domain in the Flyner coordinate system, centered on the main vehicle and based on the relative poses of the two vehicles, incorporating road alignment features, specifically: (1) In equation (1), The main car and the surrounding area Collision risk between obstacle vehicles in the predicted time domain , , The number of obstructing vehicles around the main vehicle. This refers to the prediction step size of the model predictive control algorithm. This is the adjustment coefficient; , , The main vehicle and the first vehicle in the Freyner coordinate system respectively The relative tangential position, normal position, and distance between the obstacle vehicles; Main vehicle and the first Lateral safety distance for each obstacle vehicle , The main vehicle is in the first After the first obstacle vehicle, the main vehicle avoids the two longitudinal safety distance limits created by the obstacle vehicle by braking or changing lanes, taking into account that the main vehicle exceeds the first obstacle vehicle's distance. The obstacle avoidance safety features behind the obstacle vehicle include a collision risk zone after overtaking, creating a longitudinal safety distance limit. .

2. The collision risk model according to claim 1, characterized in that: Set to a fixed value of 5m. , , The expression is as follows: (2) In equation (2), , , This is the magnification factor, set to a fixed value. The lateral displacement required for the main vehicle to complete obstacle avoidance is taken here as one lane width, i.e., 3.75m. , These are the average relative tangential acceleration and normal acceleration for longitudinal braking obstacle avoidance and lateral lane-changing obstacle avoidance, respectively. Based on the safe distance, For vehicle width; , , and Represented as: (3) In equation (3), , , , These represent the tangential and normal positions of the main vehicle and the i-th obstacle vehicle, respectively. , Let be the tangential velocities of the main vehicle and the i-th obstacle vehicle, respectively, and assume that they remain constant in the prediction time domain; The vehicle was driving on the curve. It is mainly constrained by the tire's coefficient of friction with the ground, the friction circle, and the braking force of the braking system. The selection of [the appropriate component] mainly considers the requirements for vehicle stability and the limitations of the friction circle. Therefore, [the following is done]: and Incorporating the route table conditions, the expression is: (4) In equation (4), , The safety adjustment factors are set to 0.75 and 0.5 respectively. The coefficient of friction between the tire and the ground. The acceleration due to gravity is taken as 9.8 m / s². 2 .

3. The collision risk model according to claim 2, characterized in that: Use respectively , Substitution (1) judgment condition , To resolve the contradiction of the algebraic ring, namely: (5) In equation (5), , , , , and It is the estimated vehicle pose in the prediction time domain, which is different from the vehicle pose to be optimized. and .

4. The collision risk model according to claim 3, characterized in that: Step S2 is further divided into the following steps: Step S2.1: Construction The kinematic models of the main vehicle and the obstacle vehicle in the prediction time domain at each moment, namely: (6) In equation (6), , The main vehicle and the i-th obstacle vehicle are in The normal velocity at time t, and assumed to remain constant in the prediction time domain. Sampling time, ; Step S2.2: Determine the pose of the obstacle vehicle in the prediction time domain according to the kinematic model of equation (6). , ; Step S2.3: Determine the position and orientation of the main vehicle , ; When the obstacle avoidance optimization algorithm is run for the first time, the kinematic model of equation (6) is used to determine... , ; When the optimization algorithm is not being run for the first time, the planning layer using the obstacle avoidance hierarchical architecture uses the master vehicle pose obtained from the previous time step as... , ; Step S2.4: [The text appears to be incomplete and contains several grammatical errors. A more accurate translation would require the full context.] , , , , , Substituting into equation (5) yields the expression for the collision risk model, which is used to solve the obstacle avoidance optimization problem.

5. The collision risk model according to claim 4, characterized in that: The collision risk model is also used in the obstacle avoidance and decision-making integrated architecture; in this case, when the obstacle avoidance optimization algorithm is run for the first time, the main vehicle's position and posture are determined according to the kinematic model of equation (6). , When the optimization algorithm is not being run for the first time, the optimal control quantity obtained from the optimization needs to be converted into the main vehicle pose based on the vehicle dynamics model. , Here, the front wheel steering angle is selected as the optimized control variable to study the vehicle's lateral obstacle avoidance.

6. The collision risk model according to claim 5, characterized in that: When the optimization algorithm is not being run for the first time, the optimal control quantity obtained from the optimization is converted into the main vehicle pose based on the vehicle dynamics model in the prediction time domain. , This needs to be broken down into the following two steps: Step S2.3.1: Based on the two-degree-of-freedom dynamic model of the vehicle single track based on the tracking deviation, a multi-curvature augmented discrete dynamic model is designed using the model predictive control method; Step S2.3.2: Substitute the optimal control quantity obtained from the previous obstacle avoidance and control integrated system into the multi-curvature augmented discrete dynamics model to obtain the vehicle's pose in the predicted time domain. , ; The designed multi-curvature augmented discrete dynamics model is as follows: (7) In equation (7), , For the first Step-by-step predicted output , For the first The control quantity calculated step by step, , Let the control step size be the model predictive control step size, and , For the first The state variables of the step; and The expression is as follows: (8) In equation (8), , ; .

7. The collision risk model according to claim 6, characterized in that: Step S2.3.1 is further subdivided into the following steps: Step S2.3.1.1: Construct a single-track dynamic model of the vehicle, that is, construct a model in the Flyner coordinate system with the vehicle's normal position. Normal velocity Yaw angle deviation and the rate of change of yaw angle deviation As a state variable, the vehicle's front wheel steering angle The single-track two-degree-of-freedom dynamic model, which serves as the control variable, is also known as the continuous state-space equation: (9) In equation (9), , , The curvature of the reference trajectory, , , , This refers to the individual tire lateral stiffness of the front and rear wheels of the vehicle. For vehicle quality, For the longitudinal speed of the vehicle, , This is the distance between the centers of gravity of the front and rear axles of the vehicle. Let be the moment of inertia of the vehicle about the z-axis; Step S2.3.1.2: Discretization of the continuous state-space equation; converting the continuous state-space equation of equation (9) into a discrete equation with a fixed sampling frequency: (10) In equation (10), , , , Sampling time; Step S2.3.1.3: Design the discrete state-space equations with increased curvature of the reference trajectory, incorporating road alignment features; Extending the reference trajectory curvature of equation (10) to the state variables, and using the front wheel steering angle as the control variable, a discrete state-space equation is constructed. Step by step, we will broaden the future. The curvature of the step, and the state variables at this time are: (11) The corresponding state-space equation is: (12) In equation (12), , ; ; Step S2.3.1.4: Construct a prediction model; using the model recursion of equation (12), predict the future output based on the current state quantity and future control quantity of the controlled object, and obtain equation (7).

8. The collision risk model according to claim 7, characterized in that: Step S2.3.2 is further subdivided into the following steps: Step S2.3.2.1: Substitute the optimal front wheel steering angle obtained in the previous round of optimization of the obstacle avoidance and control integrated system into equation (7) to obtain... In the collision risk model of formula (5) ; Step S2.3.2.2: [The sentence is incomplete and requires more context to be translated accurately.] Substitution (13) To obtain the collision risk model in equation (5) ; (13) , They are respectively The initial tangential position and velocity vector of the main vehicle at each moment. Sampling time, .

9. The collision risk model according to claim 8, characterized in that, When the model is used in an integrated obstacle avoidance and control architecture, equations (7) and (13) need to be substituted into the model expression of equation (5) to convert the expression into a function of the vehicle control quantity to be optimized, so as to adapt to the optimization algorithm.

10. The collision risk model according to claim 9, characterized in that, By directly incorporating road alignment features and road surface conditions into Equation (1), the multi-curvature augmented dynamic model of Equation (7) is designed to indirectly incorporate road alignment features, ensuring that the collision risk model is applicable to extreme curve scenarios.

Images