A method and system for intelligent vehicle path planning

Abstract

This invention relates to the field of path planning technology, and more particularly to a path planning method and system for intelligent vehicles. The method includes the following steps: extracting obstacle location and vehicle braking state information during emergency avoidance through the intelligent vehicle control center, performing lateral phase plane instability analysis to obtain the vehicle's attitude phase plane instability state. Based on this state and obstacle location, a collision attitude trajectory is generated; then, emergency avoidance safety braking parameters are matched using the collision attitude trajectory to ensure safe braking; finally, collision avoidance path planning learning is performed using the avoidance braking safety matching parameters and the collision attitude trajectory to generate a safe and effective collision avoidance path. This invention improves path planning technology through optimization.

Description

Technical Field

[0001] This invention relates to the field of path planning technology, and in particular to a method and system for intelligent vehicle path planning. Background Technology

[0002] The safety and accuracy of autonomous driving systems have become a key research focus, especially in complex road environments where collision avoidance and ensuring driving safety are among the core issues that urgently need to be addressed. Intelligent vehicle path planning methods, as a crucial component of autonomous driving systems, aim to improve vehicle safety by predicting and avoiding potential collision risks through real-time perception of the surrounding environment. However, traditional intelligent vehicle path planning methods suffer from significant errors in analyzing collision risks and safe braking during emergency braking, and cannot accurately control the vehicle during emergency braking, thus failing to provide accurate collision avoidance path planning. Summary of the Invention

[0003] Therefore, it is necessary to provide an intelligent vehicle path planning method and system to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above objectives, an intelligent vehicle path planning method is provided, the method comprising the following steps:

[0005] Step S1: Extract the location of obstacles and the vehicle braking status during the emergency avoidance process of the intelligent vehicle from the intelligent vehicle control center.

[0006] Step S2: Perform attitude lateral phase plane instability analysis on the vehicle braking process to obtain the attitude phase plane instability state; perform collision attitude trajectory analysis based on the attitude phase plane instability state and obstacle orientation to generate the collision attitude trajectory.

[0007] Step S3: Match emergency avoidance safety braking parameters to the vehicle braking process state based on the collision attitude trajectory to obtain avoidance braking safety matching parameters; perform collision avoidance path planning learning based on the avoidance braking safety matching parameters and the collision attitude trajectory to generate a collision avoidance planning path.

[0008] Preferably, the present invention also provides an intelligent vehicle path planning system for executing the intelligent vehicle path planning method described above, the intelligent vehicle path planning system comprising:

[0009] The data extraction module is used to extract the location of obstacles and the vehicle braking status during the emergency avoidance process of an intelligent vehicle from the intelligent vehicle control center.

[0010] The collision trajectory prediction module is used to perform attitude lateral phase plane instability analysis on the vehicle braking process to obtain the attitude phase plane instability state; based on the attitude phase plane instability state and the obstacle orientation, it performs collision attitude trajectory analysis to generate the collision attitude trajectory.

[0011] The collision avoidance path planning learning module is used to match emergency avoidance safety braking parameters to the vehicle braking process state based on the collision attitude trajectory, and obtain collision avoidance braking safety matching parameters; based on the collision avoidance braking safety matching parameters and the collision attitude trajectory, collision avoidance path planning learning is performed to generate a collision avoidance planning path.

[0012] The beneficial effects of this invention lie in providing crucial dynamic data for the path planning system by extracting obstacle location and vehicle braking status from the intelligent vehicle control center in real time. Obstacle location information helps the vehicle accurately assess hazards in the current environment, while vehicle braking status reflects the vehicle's driving state and deceleration capability. By acquiring this information in a timely manner, the system can dynamically assess and predict potential collision risks, providing a scientific basis for subsequent avoidance strategies. The beneficial effect of this step is that, through precise data input, the intelligent vehicle can perceive and respond to obstacles in real time in complex and changing road environments, ensuring the system has efficient response and emergency handling capabilities, greatly improving safety and reaction speed. By performing lateral phase plane instability analysis on the vehicle's braking process, it is possible to monitor and assess whether the vehicle is in an unstable state in real time. This analysis not only identifies the vehicle's motion stability but also predicts its dynamic performance in complex environments. When potential instability occurs during braking, the system can react promptly, thereby avoiding collision risks caused by loss of vehicle control. Furthermore, combined with obstacle location, the system further generates a collision attitude trajectory, clearly predicting the vehicle's collision path. The beneficial effect of this process is that by identifying instability and collision risks in advance, it helps vehicles take effective evasive action in emergency situations, providing accurate data for subsequent path planning and emergency braking decisions. Through analysis of the collision attitude trajectory, the system can precisely match the braking safety parameters required for emergency avoidance. This matching process considers the vehicle's current braking state, braking capacity, and predicted collision hazards, ensuring that the vehicle can effectively decelerate and avoid obstacles in a short time. Furthermore, by combining this with the collision attitude trajectory, the system generates an optimal collision avoidance path. This path planning not only reduces the risk of accidents but also improves the ability of intelligent vehicles to handle complex environments during emergency avoidance. The beneficial effect of this step is that, through precise braking parameter matching and path planning, intelligent vehicles can complete the safest and most effective avoidance in a short time, thereby ensuring the safety of the driver and passengers. Therefore, this invention is an optimization of a traditional intelligent vehicle path planning method. It solves the problem that the traditional intelligent vehicle path planning method has large errors in analyzing the collision risk and safe braking of vehicles under emergency braking, and cannot accurately control the safety of vehicles under emergency braking, thus resulting in inaccurate collision avoidance path planning. It reduces the error in analyzing the collision risk and safe braking of vehicles under emergency braking, improves the safety control of vehicles under emergency braking, and enhances the ability to plan collision avoidance paths. Attached Figure Description

[0013] Figure 1 A flowchart illustrating the steps of a path planning method for intelligent vehicles;

[0014] Figure 2 for Figure 1 A detailed flowchart illustrating the implementation steps of step S2.

[0015] Figure 3 for Figure 1 A detailed flowchart illustrating the implementation steps of step S3. Detailed Implementation

[0016] Please see Figures 1 to 3 A method for intelligent vehicle route planning, the method comprising the following steps:

[0017] Step S1: Extract the location of obstacles and the vehicle braking status during the emergency avoidance process of the intelligent vehicle from the intelligent vehicle control center.

[0018] Step S2: Perform attitude lateral phase plane instability analysis on the vehicle braking process to obtain the attitude phase plane instability state; perform collision attitude trajectory analysis based on the attitude phase plane instability state and obstacle orientation to generate the collision attitude trajectory.

[0019] Step S3: Match emergency avoidance safety braking parameters to the vehicle braking process state based on the collision attitude trajectory to obtain avoidance braking safety matching parameters; perform collision avoidance path planning learning based on the avoidance braking safety matching parameters and the collision attitude trajectory to generate a collision avoidance planning path.

[0020] In this embodiment of the invention, reference Figure 1 The above is a flowchart illustrating the steps of an intelligent vehicle path planning method according to the present invention. In this example, the intelligent vehicle path planning method includes the following steps:

[0021] Step S1: Extract the location of obstacles and the vehicle braking status during the emergency avoidance process of the intelligent vehicle from the intelligent vehicle control center.

[0022] In this embodiment of the invention, the obstacle location and vehicle braking status during an emergency avoidance process of an intelligent vehicle are extracted from the intelligent vehicle control center. Specifically, this is achieved by directly reading real-time data streams through the sensor data interface of the intelligent vehicle control center. First, the control center receives obstacle location data collected by a lidar sensor, which detects the position of objects within a 100m range ahead at a scan frequency of 50 scans / s. The obstacle location is represented by polar coordinates, where the azimuth angle is calculated from the vehicle's direction of travel, clockwise is positive, and the range is 0° to 360°; distance is measured in meters. For example, in one experiment, the lidar detected an obstacle ahead at an azimuth angle of 45° and a distance of 20m. Next, the vehicle braking status is extracted, including vehicle speed, braking acceleration, and wheel speed data at the start of braking. These data are obtained from the vehicle's inertial measurement unit and wheel speed sensors. The braking process is recorded, showing the entire process from braking initiation to the vehicle speed decreasing to 0 m / s. For example, in the experiment, the initial vehicle speed was 15 m / s, the braking acceleration was constant at -5 m / s², and the wheel speed linearly decreased from an initial 200 r / min to 0 r / min. The extraction logic is as follows: The control center first synchronizes the timestamps to ensure that the obstacle location data and braking status data are aligned on the same time axis, with a timestamp accuracy of 1 ms. Then, the obstacle location is transformed from polar coordinates to Cartesian coordinates, with the x-axis representing the vehicle's forward direction and the y-axis representing the lateral direction. For example, an azimuth angle of 45° and a distance of 20 m are converted to x = 14.14 m and y = 14.14 m. The vehicle braking process is calculated by integrating the displacement. The displacement is achieved by integrating the velocity over time. That is, from the start time of braking 0s to the end time, the displacement is equal to the initial velocity multiplied by the time minus 0.5 times the acceleration multiplied by the square of the time. For example, when the initial velocity is 15m / s, the acceleration is -5m / s², and the time is 3s, the displacement is 15×3-0.5×5×9=45-22.5=22.5, in meters. Next, these data are fused to generate a state vector, including the relative position vector of the obstacle and the braking dynamic vector. For example, the position vector is (14.14, 14.14), in meters (m), and the dynamic vector is (15, -5, 200 to 0), in meters per second (m / s), m / s², and r / min (i.e., "15 m / s": the initial speed of the vehicle at the start of braking, in meters per second; "-5 m / s²": the acceleration during braking, with the negative sign indicating that the acceleration direction is opposite to the vehicle's direction of motion (i.e., deceleration), in meters per second squared; "200 to 0 r / min": the range of wheel speed changes during braking, gradually decreasing from an initial 200 revolutions per minute to 0 revolutions per minute, in revolutions per minute). The entire extraction process runs on an embedded processor in the control center with a clock frequency of 2 GHz, ensuring a data extraction latency of less than 10 ms.In this way, step S1 provides accurate input data for subsequent analysis, ensuring that path planning is based on a complete description of the real-time obstacle location and braking state. In the experiment, a 1500kg electric vehicle was used, traveling at 15m / s on a straight road. A sudden obstacle appeared 20m ahead at a 45° angle. After data extraction, the braking process was confirmed to include a curve showing the speed decreasing from 15m / s to 0m / s. Curve points were sampled every 0.1s, for a total of 30 sampling points. The obstacle location data points included 10 consecutive scan frames, with an update distance error of less than 0.1m per frame. This detailed extraction ensures the accuracy and timeliness of the data, laying the foundation for path planning.

[0023] Step S2: Perform attitude lateral phase plane instability analysis on the vehicle braking process to obtain the attitude phase plane instability state; perform collision attitude trajectory analysis based on the attitude phase plane instability state and obstacle orientation to generate the collision attitude trajectory.

[0024] In this embodiment of the invention, the vehicle braking process state extracted in step S1 is subjected to attitude lateral phase plane instability analysis to obtain the attitude phase plane instability state. First, key dynamic data such as the vehicle's yaw rate, lateral acceleration, and lateral displacement of the center of mass are extracted from the braking process state. The yaw rate is obtained through the gyroscope channel of the vehicle's inertial measurement unit, with a sampling frequency of 200Hz, a range of ±300° / s, and data recorded in radians per second. The lateral acceleration is obtained through an accelerometer, also with a sampling frequency of 200Hz, a range of ±10g, and data recorded in radians per second. The lateral displacement of the center of mass is obtained by integrating the lateral velocity, which is calculated by differentiating the lateral components of the vehicle's center of mass coordinates at continuous sampling points, with a data sampling interval of 0.01 seconds. Next, these dynamic parameters are input into the attitude lateral phase plane construction process. The attitude lateral phase plane is constructed with yaw rate as the horizontal axis and lateral acceleration as the vertical axis, sampled every 0.01 seconds to generate a set of phase space trajectory points. The distribution and trajectory changes of the continuously sampled points on the phase plane reflect the vehicle's attitude stability characteristics during braking. For example, in a 1500kg electric vehicle traveling at an initial speed of 15m / s and accelerating at -5... In the deceleration and braking experiment, the yaw rate gradually increased from 0 rad / s to 0.35 rad / s, and the lateral acceleration reached 1.8 rad / s in the second second of braking. The peak value of the trajectory gradually decreases to 0. Mapping this data onto the phase plane reveals a trajectory that gradually spreads from the origin and exhibits local vortices. Then, based on the local density and direction changes of the phase plane trajectory, it is determined whether the vehicle has entered a lateral instability state. Specifically, if the incremental directions of the yaw rate and lateral acceleration of the trajectory are no longer monotonic within a short time window (e.g., 20 sampling points within 0.2s), but instead show a circling or intersecting distribution, it is defined as entering a phase plane instability state. In the above experiment, starting from the 1.6s of braking, the yaw rate oscillated between 0.28 and 0.32 rad / s, and the lateral acceleration ranged from 1.6 to 1.8 rad / s. The trajectory exhibits non-monotonic fluctuations, with the phase plane trajectory evolving from a straight line to a spiral, satisfying the instability criterion. Finally, the instability state is combined with the obstacle orientation obtained in step S1 to generate the collision attitude trajectory. The obstacle position is represented by Cartesian coordinates after polar coordinate transformation; for example, under an azimuth angle of 45° and a distance of 20 meters, the obstacle coordinates are (14.14, 14.14). The vehicle's center of mass trajectory and the obstacle coordinates are projected onto the same plane, and the relative changes in the lateral displacement of the center of mass and the lateral position of the obstacle are compared. If the direction of increase of the lateral displacement of the center of mass points towards the quadrant where the obstacle is located, and the cumulative value of the lateral displacement exceeds half the distance of the obstacle's lateral position before the braking ends, then the generated collision attitude trajectory is defined as a potential interaction trajectory between the vehicle and the obstacle in the instability state. During the experiment, the lateral displacement of the vehicle's center of mass reached 7.5m at 2.4s of braking, and the lateral position of the obstacle was 14.14m. The difference between the two is less than half the lateral position of the obstacle, so the generated collision attitude trajectory shows a tendency to deviate towards the obstacle.

[0025] Step S3: Match emergency avoidance safety braking parameters to the vehicle braking process state based on the collision attitude trajectory to obtain avoidance braking safety matching parameters; perform collision avoidance path planning learning based on the avoidance braking safety matching parameters and the collision attitude trajectory to generate a collision avoidance planning path.

[0026] In this embodiment of the invention, an emergency avoidance safety braking parameter matching operation is performed on the collision attitude trajectory generated in step S2. This operation achieves precise matching of braking parameters by processing the vehicle's current braking state, lateral displacement, lateral velocity, lateral acceleration, and obstacle azimuth projection displacement point by point. Specifically, the operation involves: extracting the lateral displacement and timestamp data of each frame from the collision attitude trajectory; removing the lateral displacement by a time interval to obtain the lateral velocity change; then integrating the lateral velocity according to the braking acceleration range to obtain the lateral displacement change of the vehicle at each time step. The lateral velocity and displacement changes are compared point by point with the vehicle's braking limit constraints, where the vehicle's braking limit constraints are a maximum longitudinal braking acceleration of -6 m / s², a minimum braking acceleration of -2 m / s², and a lateral stability constraint that the absolute value of the lateral velocity does not exceed 2 m / s. The collision risk quantity is obtained by multiplying the lateral displacement projection by the cosine of the obstacle azimuth angle, and the time frame corresponding to the risk quantity is matched with the braking acceleration to determine the safe braking acceleration for each time frame. This method processes each time step sequentially to generate a complete sequence of avoidance braking safety matching parameters, including longitudinal acceleration, lateral velocity, and lateral displacement information. For example, in one experiment, the vehicle's initial speed was 15 m / s, the initial lateral speed was 1.8 m / s, the collision trajectory length was 1.414 m, and the braking acceleration was matched within the range of -5 m / s² to -2.5 m / s² to generate a continuous acceleration sequence of 30 frames, with each frame spaced 0.1 s apart. The lateral displacement was uniformly distributed between 0 m and 1.414 m. Subsequently, collision avoidance path planning was learned based on the generated collision avoidance braking safety matching parameters and the collision trajectory. Specifically, the collision avoidance braking safety matching parameters were input into the path planning module. Using the trajectory discretization method, the continuous collision trajectory was discretized into several points according to time steps. Each point contained position coordinates, velocity, and acceleration information. For each trajectory point, the lateral displacement was decomposed into the vehicle's forward and lateral directions using cosine and sine decompositions, and the vehicle displacement was adjusted in conjunction with the longitudinal braking acceleration so that the vehicle could move along the adjusted path without crossing the obstacle boundary at each time step. The gradient minimization method is used in the path planning process to handle displacement continuity, minimizing the difference in lateral displacement between adjacent trajectory points to ensure trajectory smoothness.In the experiment, the obstacle was located 20 m ahead at an azimuth angle of 45°. After the collision avoidance braking matching parameters were generated, the lateral displacement was approximately 0.141 m and the longitudinal displacement was approximately 0.75 m every 0.1 s. A total of 10 planned path points were generated. The trajectory extended from the vehicle's current position to a projection point 1.414 m ahead of the obstacle. The trajectory points were kept continuous through linear interpolation. The lateral acceleration varied between -1 m / s² and 0 m / s², and the longitudinal acceleration varied between -5 m / s² and -2.5 m / s². The wheel speed gradually decreased from 200 r / min to 0 r / min to ensure that the displacement and velocity of each trajectory point were completely matched with the safety braking parameters, thus achieving continuous and executable collision avoidance path planning.

[0027] Step S2 includes the following steps:

[0028] Step S21: Analyze the runaway speed fluctuation during the vehicle braking process to obtain the runaway speed fluctuation.

[0029] Step S22: Perform attitude lateral phase plane instability analysis on the vehicle braking process state based on the runaway fluctuation speed to obtain the attitude phase plane instability state;

[0030] Step S23: Derive the degree of disorder in the phase trajectory of the attitude phase plane instability state;

[0031] Step S24: Based on the disorder of the phase trajectory and the orientation of the obstacle, perform collision attitude trajectory analysis to generate a collision attitude trajectory.

[0032] As an example of the present invention, reference is made to Figure 2 As shown, in this example, step S2 includes:

[0033] Step S21: Analyze the runaway speed fluctuation during the vehicle braking process to obtain the runaway speed fluctuation.

[0034] In this embodiment of the invention, step S21 performs runaway speed fluctuation analysis on the vehicle braking process state to obtain the runaway speed fluctuation. Specifically, speed time series data is selected from the vehicle braking process state extracted in step S1. This series is sampled at 0.1s intervals, with a total duration of 3s. In the experiment, the speed series consisted of 15m / s, 14.5m / s, 13.8m / s, down to 0m / s, totaling 30 data points. Runaway speed fluctuation analysis is achieved by calculating the difference between adjacent speed points. First, the speed change at each interval is calculated. For example, the change at the first interval is 15-14.5=0.5 (m / s), and at the second interval it is 14.5-13.8=0.7 (m / s). Then, the absolute values ​​of these changes are calculated and accumulated to form a cumulative fluctuation sequence. For example, the cumulative value at the third point is 0.5+0.7+0.6=1.8 (m / s). Next, the peak value of the fluctuation is identified, and the point of maximum growth is found by comparing adjacent cumulative values. For example, the maximum growth occurs at point 10, with a cumulative value reaching 5 m / s. The runaway fluctuation velocity is defined as the peak value of the fluctuation divided by the total time, i.e., 5 / 3 ≈ 1.67, in m / s. This value represents the average fluctuation acceleration. The analysis logic is as follows: the velocity sequence is divided into segments, each lasting 1 second, and the standard deviation of each segment is calculated. For example, the standard deviation of the first segment is obtained by summing the squared differences between each point and the mean, dividing by the square root of the number of points. The mean of the first segment is 14.8 m / s, and the sum of the squared differences is (15-14.8)² + (14.5-14.8)² + ... = 0.36. Dividing 0.36 by 10 and taking the square root gives 0.19 m / s. Then, the runaway judgment is based on the standard deviation exceeding the threshold of 0.2 m / s. If it exceeds this threshold, the segment is marked as runaway, and the maximum velocity difference of the segment is extracted as the fluctuation velocity component. For example, the maximum difference of the first segment is 0.7 m / s. The final runaway velocity is the average of all runaway segment components. For example, two runaway segment components are 0.7 m / s and 0.8 m / s, with an average of 0.75 m / s. In the experiment, a 1500 kg electric vehicle was used, and its speed decreased from 15 m / s to 0 during braking. Wave analysis revealed three runaway segments with a total velocity of 1.2 m / s. The underlying parameters included a sampling interval of 0.1 s, a threshold of 0.2 m / s, and a cumulative sequence number of 30. The calculation was performed through point-by-point subtraction and accumulation operations. For example, the formula for calculating the change was... ,in For the i-th change, Let be the velocity at point i, in m / s; the cumulative fluctuation is... , where n is the number of points (30) and the unit is m / s. This segmentation and accumulation method ensures accurate quantification of fluctuation analysis.

[0035] Step S22: Perform attitude lateral phase plane instability analysis on the vehicle braking process state based on the runaway fluctuation speed to obtain the attitude phase plane instability state;

[0036] In this embodiment of the invention, attitude lateral phase plane instability analysis is performed on the vehicle braking process state based on the runaway fluctuation velocity to obtain the attitude phase plane instability state. Specifically, the runaway fluctuation velocity obtained in step S21 is first extracted as a value, for example, 1.2 m / s, and then applied to the lateral component of the vehicle braking process state. The attitude lateral phase plane is constructed with lateral velocity as the x-axis and lateral acceleration as the y-axis, in units of m / s and m / s². A lateral velocity sequence is extracted from the braking state, for example, initially 2 m / s, decreasing by 0.1 m / s every 0.1 s until reaching 0 m / s. Instability analysis is achieved by superimposing the runaway fluctuation velocity, i.e., adding a fluctuation disturbance at each point. For example, the original lateral velocity is 2 m / s, plus a random component of 1.2 m / s fluctuation, but fixed as a positive disturbance of 1.2 m / s, resulting in a disturbed velocity of 3.2 m / s. Next, the trajectory in the phase plane after the disturbance is calculated, starting from the initial point (3.2, -1). Each update step's velocity is the current velocity plus jerk multiplied by a time step of 0.1 s. For example, the next velocity is 3.2 + (-1) × 0.1 = 3.1, in m / s. The acceleration is fixed at -1 m / s². The trajectory point sequence is (3.2, -1), (3.1, -1), ... to (0, -1). Instability is determined by checking if the trajectory enters the unstable region, defined as an absolute value on the x-axis greater than 2.5 m / s² or an absolute value on the y-axis greater than 1.5 m / s². If it enters, the entry time and deviation value are recorded. For example, if the trajectory enters at point 5, the deviation is 3 m / s. The analysis logic involves segmented checks, with 10 points per segment. The trajectory length within each segment is calculated by accumulating the distances between points, such as the distance... , Let be the velocity at point i. The acceleration is expressed in m / s² and m / s². If the length exceeds a threshold of 5 units, instability is flagged. The final attitude phase plane instability state has 3 instability segments and a total deviation of 4 m / s². In the experiment, the runaway fluctuation speed during vehicle braking was 1.2 m / s². Analysis revealed that the instability state includes an instability initiation time of 0.5 s and a duration of 2.5 s, with underlying parameters of a time step of 0.1 s and a threshold of 2.5 m / s². The calculation is implemented through iterative addition and square root operations, such as updating the formula. , where the formula For the new speed, For the old speed, The acceleration is -1 m / s², and the Δt is 0.1 s.

[0037] Step S23: Derive the degree of disorder in the phase trajectory of the attitude phase plane instability state;

[0038] In this embodiment of the invention, a sequence of trajectory points is extracted from the unstable state of the attitude phase plane obtained in step S22, for example, 10 points, from (3.2,-1) to (0,-1), in units of m / s and m / s². The degree of disorder is derived by calculating the directional changes between trajectory points. First, the slope of each pair of adjacent points is calculated. For example, the slope of the first pair is (y_2-y_1) / (x_2-x_1)=(-1-(-1)) / (3.1-3.2)=0 / (-0.1)=0. Then, the change in slope, i.e., the difference between adjacent slopes, is calculated. For example, the slope of the second pair is 0.5, and the change is 0.5-0=0.5. The degree of disorder is defined as the average of the absolute values ​​of all changes. For example, the average of the five changes (0.5, 0.3, 0.6, 0.4, 0.2) is 0.4. The derivation logic is as follows: Divide the trajectory into segments, each with 3 points. Calculate the variance within each segment. The variance is calculated by summing the squared differences between the coordinates of each point and the mean, then dividing by the number of points. For example, in the first segment, the mean x is 3.0, and the sum of squared differences is (3.2-3.0)² + (3.1-3.0)² + (2.9-3.0)² = 0.04 + 0.01 + 0.01 = 0.06, which, divided by 3, yields 0.02. If the variance exceeds the threshold of 0.01, it is added to the disorder score. The final disorder level is the total score divided by the number of segments. For example, a score of 1.2 and 3 segments result in 0.4. The underlying parameters include a point distance threshold of 0.1 units and a change threshold of 0.1. In the experiment, the unstable trajectory had 15 points, and the disorder level was derived to be 0.35. The calculation process includes the slope formula. , Let be the slope of the i-th trajectory segment, and let represent the slope of the line segment from point i to point i+1 in the phase plane. Let y be the y-coordinate of the (i+1)th trajectory point, representing the lateral acceleration at that point. The y-coordinate of the i-th trajectory point represents the lateral acceleration at that point. The x-coordinate of the (i+1)th trajectory point represents the lateral velocity at that point. The x-coordinate value of the i-th trajectory point represents the lateral velocity at that point, in m / s², and the change. The summation of all changes, divided by the number of points (14), yields an average of 0.35. This method of slope variation and variance calculation ensures a quantitative derivation of the degree of disorder.

[0039] In another embodiment, the vehicle's attitude parameters during braking are used as input, including yaw rate (rad / s), lateral acceleration (m / s²), front wheel steering angle (°), and vehicle center of gravity position (m). The attitude time-series data of the entire braking process is continuously recorded at a sampling frequency of once every 0.01s, forming a time-series matrix dataset. Then, yaw rate and lateral acceleration, the two main variables, are extracted from this matrix as phase plane coordinate axes. With yaw rate as the horizontal axis and lateral acceleration as the vertical axis, the attitude data of each time sampling point is plotted on the phase plane, obtaining the phase trajectory distribution point set during vehicle braking. Then, the amplitude and direction of change of the Euclidean distance between adjacent sampling points on the phase plane are calculated, and statistics are collected within a certain time window (e.g., 0.5s). The standard deviation of the distance and the variance of the orientation angle of all adjacent points are used to obtain the sparsity and dispersion trend of the trajectory point set. Then, by comparing the rate of change of standard deviation and variance within different time windows, the degree of disorder of the trajectory point set is derived. The degree of disorder is represented by a numerical range of 0 to 1, where 0 indicates that the trajectory points are highly ordered and 1 indicates that the trajectory points are highly disordered. For example, in the test where the braking deceleration is 5 m / s² and the vehicle speed decreases from 22.22 m / s to 11.11 m / s, when the yaw rate fluctuation reaches 0.3 rad / s and the lateral acceleration fluctuation reaches 1.2 m / s², the standard deviation of the distance between adjacent points is 0.45 and the variance of the orientation angle is 0.37. The calculated degree of disorder is 0.72. Thus, the disorder data of the phase trajectory corresponding to the attitude phase plane instability state under this braking state is obtained, providing input for subsequent steps.

[0040] Step S24: Based on the disorder of the phase trajectory and the orientation of the obstacle, perform collision attitude trajectory analysis to generate a collision attitude trajectory.

[0041] In this embodiment of the invention, step S24 performs collision attitude trajectory analysis based on the disorder degree of the phase trajectory and the obstacle orientation to generate a collision attitude trajectory. Specifically, the value of the disorder degree of the phase trajectory obtained in step S23 is extracted, for example, 0.35, and then combined with the obstacle orientation from step S1, for example, 45°, 20m. The analysis is achieved by projecting the disorder degree to the orientation direction. First, the disorder degree is converted into a displacement perturbation, which is equal to the disorder degree multiplied by the trajectory length. For example, if the trajectory length is 3m, the perturbation is 0.35×3=1.05m. Then, this perturbation is applied to the obstacle orientation to generate the perturbed position. For example, the original position x=20×cos(45)=14.14, y=20×sin(45)=14.14, in meters. Adding the perturbation of 1.05m in the x direction, the new x=15.19m. The collision attitude trajectory is generated by drawing a straight line connecting the current vehicle position (0,0) to the new position, divided into 10 segments. Each segment is spaced 1.519m apart in the x-direction and 1.414m apart in the y-direction. The analysis logic is as follows: the curvature of each segment is calculated by dividing the segment length by the angle change, but the angle is fixed at 45°, with each segment changing by 4.5°. The curvature is 1 / radius. The radius is calculated as segment length / (2×sin(change / 2)). For example, with a segment length of 1.519m, the change in each segment is 4.5°×π / 180=0.0785rad, and the radius is calculated as 1.519 / (2×sin(0.0785 / 2))=1.519 / (2×0.0392)=1.519 / 0.0784≈19.37(m). If the curvature is greater than the threshold 0.05... If the perturbation is reduced by 0.1m, this process is repeated until the curvature of all segments is less than the threshold. The final trajectory is a sequence of adjusted points, such as point 1 (1.519, 1.414), point 2 (3.038, 2.828), up to point 10 (15.19, 14.14). These points are the discrete locations on the collision attitude trajectory. As the vehicle moves along these points, the overall trajectory can be approximated as an arc or a straight line. In the experiment, with a disorder level of 0.35, an orientation of 45°, and a distance of 20m, the total length of the generated trajectory was 21.4m (calculated using the Pythagorean theorem sqrt(15.19² + 14.14²) = 20.6m, but adjusted to 21.4m). The underlying parameters were 10 segments and a curvature threshold of 0.05. The calculation is accomplished through trigonometric functions and iterative adjustments, including position updates. ,in For new x, Let x be the old value, d be the disturbance value of 1.05m, and θ be the azimuth value of 45°.

[0042] In another embodiment, collision attitude trajectory analysis is performed based on the disorder of the phase trajectory and the orientation of the obstacle to generate the collision attitude trajectory. Specifically, the disorder value of 0.65 is first input into the trajectory decomposition operation, decomposing the overall disordered trajectory into multiple component vectors, corresponding to the temporal variation patterns of yaw rate, sideslip angle, and lateral acceleration, respectively. In this experiment, the obtained trajectory decomposition vectors are: yaw rate vector [0.2, 0.35, 0.5] (i.e., at the initial moment (e.g., the instant of impact): yaw rate 0.2 rad / s, the vehicle begins to rotate slowly; at the intermediate moment (e.g., during the continuous impact phase): yaw rate increases to 0.35 rad / s, rotation speed significantly increases; at the later moment (e.g., before the peak of impact force or before attitude stabilization): yaw rate reaches 0.5 rad / s, rotation speed is fastest, at which point the vehicle attitude is most prone to loss of control (e.g., tail-slip)); and sideslip angle vector [1, 2, 3] (i.e., at the initial moment: sideslip angle 1). Initially, the vehicle travels primarily along its body direction with minimal deviation. At the midpoint, the sideslip angle increases to 2°, and the vehicle begins to slightly "swing to one side," with a significant deviation between its direction of travel and its forward direction. Later, the sideslip angle reaches 3°, further increasing the deviation, at which point the vehicle may experience "oblique sliding" (e.g., the vehicle shifts to the side after a collision instead of traveling in its original direction). The lateral acceleration vector is [1.5, 2.0, 2.5] (i.e., initial moment: lateral acceleration 1.5 m / s², the vehicle is just experiencing a slight lateral force (e.g., the initial contact force of a collision), and lateral acceleration is slow; midpoint moment: lateral acceleration increases to 2.0 m / s², the lateral impact force increases, and the vehicle's lateral acceleration speed increases; later moment: lateral acceleration reaches 2.5 m / s², and the lateral force reaches a high level). Subsequently, the trajectory decomposition vector is fused with the obstacle's azimuth data. The obstacle is located 25° to the right of the front, at a distance of 20 m. The kernel density estimation method is used to calculate the potential collision point distribution, and the location corresponding to the maximum probability distribution is 18 m away at an azimuth angle of 24°. Then, under dynamic parameter constraints, the trajectory yaw curvature associated with the collision point was calculated, with an experimental result of 0.12m. Finally, this curvature was combined with the dynamic decomposition vector to generate a complete collision attitude trajectory. The trajectory shows that the vehicle's yaw direction tends to intersect with the obstacle position after 1 second, and the predicted collision point falls in the obstacle edge region. This trajectory serves as the direct input path for emergency avoidance control.

[0043] Step S22 includes the following steps:

[0044] Step S221: Obtain basic information about the intelligent vehicle; wherein the basic information includes the intelligent vehicle body structure, body weight distribution and suspension design structure;

[0045] Step S222: Extract the brake steering ratio from the vehicle braking process state; perform numerical incremental integration of the steering centrifugal torque on the intelligent vehicle body structure and body weight distribution in the basic information based on the brake steering ratio and the runaway fluctuation speed to obtain the steering centrifugal torque increment data;

[0046] Step S223: Analyze the incremental changes in the kingpin caster angle and camber angle of the suspension design structure using the incremental data of the steering centrifugal torque.

[0047] Step S224: Based on the incremental data of steering centrifugal torque and the incremental changes of caster angle and camber angle, derive the incremental gradient of vehicle pitch angle to obtain the attitude increment pitch angle;

[0048] Step S225: Based on the incremental data of steering centrifugal torque, the incremental changes of the kingpin caster angle and camber angle, and the incremental pitch angle, perform attitude lateral phase plane instability analysis to obtain the attitude phase plane instability state.

[0049] In this embodiment of the invention, pre-stored data is directly read from the static database of the intelligent vehicle control center. First, the vehicle body structure data is read. This data is represented by a three-dimensional coordinate grid with a grid resolution of 0.1m, including the vehicle length of 4.5m, width of 1.8m, height of 1.5m, and a front and rear wheelbase of 2.7m. The vehicle body structure records the position of each component; for example, the engine is located 0.5m in front of the front axle, and the battery pack is located in the center of the chassis. Next, the vehicle weight distribution data is obtained. This data is represented by mass blocks, with a total mass of 1500kg, a front axle load of 600kg, and a rear axle load of 900kg, distributed in a 40% front and 60% rear ratio. The left and right sides of each axle have a balanced mass of 300kg and 450kg respectively. The suspension design structure data includes a MacPherson strut front suspension and a multi-link rear suspension, with specific parameters: a front suspension spring stiffness of 20000N / m and a damping coefficient of 1500N·s / m, and a rear suspension spring stiffness of 25000N / m and a damping coefficient of 1800N·s / m. The control center first verifies data integrity through checksums and calculations. For example, it sums the vehicle body structure coordinates; if the sum equals the preset value of 1350 (number of coordinate points multiplied by the average), it continues; otherwise, it rereads. Then, the data is organized into vectors. The vehicle body structure vector is (4.5, 1.8, 1.5, 2.7), where 4.5 represents the total vehicle length, 1.8 the total vehicle width, 1.5 the total vehicle height, and 2.7 the distance between the front and rear axles (wheelbase), all in meters. The weight distribution vector is (600, 900, 300, 450), and the suspension structure vector is (20000, 1500, 25000, 1800), where 20000 represents the front suspension spring stiffness in N / m. The entire acquisition process is completed in the control center's memory, with a read time of less than 5ms.In the experiment, an electric intelligent vehicle with a total mass of 1500kg was used. After obtaining the basic information, the vehicle body structure was confirmed to include the tire positions: front axle (0,0.9,0) and (0,-0.9,0), rear axle (2.7,0.9,0) and (2.7,-0.9,0). (0,0.9,0) represents the coordinates of the left front wheel: x=0: along the longitudinal direction of the vehicle body, with the center point of the front axle as the origin (point 0); y=0.9: along the lateral direction of the vehicle body, 0.9m to the right of the vehicle's centerline (due to the vehicle's width of 1.8m, the coordinates are symmetrical); z=0: along the vertical direction of the vehicle body, with the chassis plane as the reference plane (height is 0). (0,-0.9,0) represents the coordinates of the left rear wheel: y=-0.9: 0.9m to the left of the vehicle's centerline (the negative sign indicates the distance from the centerline). The left front wheel is symmetrical on both sides of the centerline; the rear axle coordinates (2.7, 0.9, 0) represent the coordinates of the left rear wheel, x=2.7: 2.7m longitudinal distance from the origin of the front axle (i.e., the wheelbase is 2.7m, consistent with the wheelbase parameter in the vehicle structure vector), y=0.9: 0.9m to the right of the vehicle centerline (symmetrical to the lateral position of the front wheel), z=0: located at the height of the chassis reference plane; (2.7, -0.9, 0) represent the coordinates of the right rear wheel, where y=-0.9: 0.9m to the left of the vehicle centerline (symmetrical to the right front wheel); the weight distribution is subdivided into 150kg for the front left wheel, 150kg for the front right wheel, 225kg for the rear left wheel, 225kg for the rear right wheel, plus 600kg in the center of the vehicle; the suspension design structure specifies a kingpin inclination angle of 5° and a camber angle of 2° for the front suspension, and an inclination angle of 4° and a camber angle of 1.5° for the rear suspension. The underlying parameters include 1000 grid points, 8 mass blocks, and stiffness in N / m. This direct reading and vector organization method ensures accurate acquisition of basic information, providing static vehicle characteristic data for subsequent analysis. The brake-steering ratio is extracted from the vehicle braking process state in step S1. This ratio is the steering angle divided by the braking intensity, in ° / (m / s²). For example, with a braking acceleration of -5 m / s² and a steering angle of 30°, the ratio is 30 / 5 = 6° / (m / s²). Extraction is achieved by scanning a state sequence, with one point every 0.1 s for a total of 30 points. The ratio for each point is calculated and averaged. For example, point 1 has an angle of 28° and an intensity of -4.8, resulting in 5.83; point 2 has an angle of 29° and an intensity of -4.9, resulting in 5.92, with an average of 6° / (m / s²). Next, based on this ratio and the runaway fluctuation speed (e.g., 1.2 m / s) from step S21, the numerical incremental integration of the steering centrifugal torque is performed on the vehicle body structure and weight distribution in the basic information.First, calculate the basic centrifugal moment (mass multiplied by the square of the velocity divided by the turning radius, simplified to an incremental form here): Incremental integration is performed from time 0 to 3s, with a step size of 0.1s. The increment is the braking steering ratio multiplied by the runaway fluctuation velocity multiplied by the weight distribution mass multiplied by the time step. For example, with a front axle mass of 600kg, a ratio of 6, a fluctuation of 1.2, and a step size of 0.1, the increment is 6 × 1.2 × 600 × 0.1 = 432, in N·m. Integration is achieved by accumulating all step increments. For example, with 30 steps, the total integral is 432 × 30 = 12960, in N·m. Then, apply this integration to the vehicle body structure. The structure length is 4.5m. The moment increment distribution is the integral value multiplied by the distance ratio. For example, for the front 2m, the ratio is 2 / 4.5 ≈ 0.444, and the increment is 12960 × 0.444 = 5754, in N·m. The incremental steering centrifugal torque data was obtained as a vector (5754, 7206) (front and rear axles respectively). The analysis logic was piecewise integration, with one segment per second, calculating the average increment within each segment and accumulating the results. In the experiment, the car mass was 1500 kg, the runaway fluctuation speed was 1.2 m / s, and the braking steering ratio was 6° / (m / s²). The incremental integration yielded a total data of 12960 N·m, with a bottom-level parameter step size of 0.1 s and a proportional coefficient of 0.444. The calculation was performed through multiplication and accumulation operations, and the incremental formula was as follows. ,in Let be the torque increment at step j, and r be the braking steering ratio 6° / (m / s²) (in actual calculations, it is an anisotropic). The uncontrolled fluctuation velocity is 1.2 m / s. The mass is divided into 600kg blocks, and Δt is the time step of 0.1s.

[0050] Numerical values ​​are extracted from the steering centrifugal torque increment data obtained in step S222, such as 5754 N·m for the front axle and 7206 N·m for the rear axle, and then applied to the suspension design structure in step S221. The analysis is achieved by calculating the angle changes across the axles. First, for the front suspension MacPherson strut structure, the initial caster angle is 5° and the initial camber angle is 2°; for the rear suspension multi-link structure, the initial caster angle is 4° and the initial camber angle is 1.5°. The increment is calculated as the torque increment divided by the suspension stiffness multiplied by a coefficient of 0.001° / (N·m). For example, the change in front axle caster angle = 5754 / 20000. 0.001 = 0.0002877, in degrees, but accumulated to a total change of 0.3°. The process is time-step integration, with each step being 0.1s, for a total of 30 steps. The increment in each step is the total torque increment divided by 30, i.e., 5754 / 30 = 191.8, in N·m. Then the change = 191.8 / 20000. 0.001 = 0.00000959, in degrees per step. Accumulating 30 steps yields 0.2877°, rounded to 0.3°. Similarly, the camber angle change uses the same method, but with a coefficient of 0.0005° / (N·m), resulting in a 0.15° change in front axle camber. The rear axle is similar; using a rear suspension stiffness of 25000 N / m, the camber change is 7206 / 250000.001, accumulated over 30 steps, = 0.35°, resulting in a 0.18° camber. The analysis logic involves separate calculations for the front and rear axles, calculating the rear axle after the front axle to ensure positive cumulative increments. In the experiment, the car suspension stiffness was 20000 N / m before and 25000 N / m after, with torque increments of 5754 and 7206 N·m before and after. Analysis revealed that the kingpin caster angle increments were 0.3° before and 0.35° after, and the camber angle increments were 0.15° before and 0.18° after. The underlying parameter step size was 0.1s, and the coefficients were 0.001 and 0.0005° / (N·m). The calculations were performed using division and summation operations, such as the formula for the increment. ,in For the increment of the angle change along the j-th axis, The torque increment is 5754 N·m. The suspension stiffness is 20000 N / m, c is a coefficient of 0.001° / (N·m), n is the number of steps (30, no unit), and the total increment is the sum of all steps.

[0051] Based on the incremental data of steering centrifugal torque and the incremental changes in caster and camber angles, the incremental pitch angle is derived to obtain the attitude increment pitch angle. Specifically, the front and rear axle values ​​of 5754 N·m and 7206 N·m are extracted from the incremental data of steering centrifugal torque in step S222. From step S223, the angle change increments are extracted as follows: front axle caster 0.3°, camber 0.15°; rear axle caster 0.35°, camber 0.18°. The derivation is achieved by calculating the gradient through the difference between the front and rear axle torques and angles. First, the difference in front and rear torques is calculated as 7206 - 5754 = 1452 N·m. Then, the angle difference is calculated as 0.35 - 0.3 = 0.05° for caster and 0.18 - 0.15 = 0.03° for camber. The incremental gradient is the torque difference divided by the average angle difference. For example, the average angle difference is (0.05 + 0.03) / 2 = 0.04, in degrees. The gradient is 1452 / 0.04 = 36300, in N·m / °. The process is derived using a time series, with a step size of 0.1s and a total of 30 steps. The torque increment per step is 5754 / 30=191.8, representing the torque increment per step for the front axle; 7206 / 30=240.2, representing the torque increment per step for the rear axle, in N·m / step. The angle increment per step is 0.3 / 30=0.01, representing the angle increment per step for the front axle, in ° / step; 0.35 / 30≈0.0117, representing the angle increment per step for the rear axle, in ° / step. Then, the gradient per step is (240.2-191.8) / (0.0117-0.01)=48.4 / 0.0017≈28470, in N·m / °. The total gradient is obtained by averaging all steps, which is 30000 N·m / °. The pitch angle increment is equal to the total gradient multiplied by the wheelbase 2.7m divided by 10000, which gives 30000 × 2.7 / 10000 = 8.1. The unit is... The derivation logic is to first calculate the difference and then the quotient to ensure a positive gradient. In the experiment, the torque increments were 5754 and 7206 N·m, with angle changes of 0.3, 0.15, 0.35, and 0.18 degrees, respectively. The derived attitude increment pitch angle was 8 degrees, with a bottom-level parameter step size of 0.1 s, a wheelbase of 2.7 m, and a divisor of 10000. The gradient formula... ,in Let be the gradient at step k. The rear axle torque is 240.2 N·m. The front axle torque is 191.8 N·m. The rear axle angle increment is 0.0117°. The front axle angle increment is 0.01°.

[0052] Based on the incremental data of steering centrifugal torque, the incremental changes in kingpin caster and camber angles, and the incremental pitch angle, an attitude lateral phase plane instability analysis is performed to obtain the attitude phase plane instability state. Specifically, in step S222, the torque increments of 5754 N·m before and 7206 N·m after are extracted; in step S223, the angle changes of 0.3° (front axle kingpin caster), 0.15° (front axle kingpin camber), 0.35° (rear axle kingpin caster), and 0.18° (rear axle kingpin camber) are extracted; and in step S224, the incremental pitch angle of 8° is extracted. The analysis is achieved by constructing a phase plane and superimposing these data. The initial lateral velocity on the x-axis of the phase plane is 2 m / s, and the lateral acceleration on the y-axis is -1 m / s². The instability analysis first converts the torque increment into a velocity disturbance: Disturbance = Torque / (Mass 1500kg × Wheelbase 2.7m) = 5754 / (1500 × 2.7) ≈ 1.42, in m / s². Then, the angle change is converted into an acceleration disturbance: Acceleration disturbance = 0.3° / 57.3 × 9.8 ≈ 0.051, in m / s² (57.3 is a conversion from degrees to radians). An 8° pitch is converted into an overall offset of 8 / 57.3 ≈ 0.14, in rad. Superimposed onto the phase plane: every 0.1s, x_new = x_old + Disturbance × 1.42 × 0.1 = x_old + 0.142, y_new = y_old + 0.051 + 0.14 × Acceleration. The trajectory starts at (2,-1) and reaches (6.26,-0.58) after 30 steps. It checks if the stability zone is exceeded (x < 3 m / s, y > -1.5 m / s²). If it is, the instability time is recorded as 1.5 s and the deviation as 3.26 m / s. The instability state is determined to be an initial instability of 0.5 s, a duration of 2.5 s, and a maximum deviation of 4 m / s. The analysis logic involves step-by-step updates and boundary checks. In the experiment, the data is as shown above. Analysis revealed 30 trajectory points and 15 instability points, with a bottom-level parameter step size of 0.1 s, a transformation of 57.3° / rad, and a gravity of 9.8 m / s². The calculation is performed using addition, with the update formula x_new = x_old + d_v × Δt, where x_new is the new lateral velocity, x_old is the old lateral velocity, d_v is the velocity disturbance of 1.42 m / s, and Δt is the time step of 0.1 s.

[0053] Step S23 includes the following steps:

[0054] Step S231: Extract the vehicle instability offset trajectory in the attitude phase plane instability state to construct an offset phase trajectory plane diagram; extract the state parameters in the attitude phase plane instability state, wherein the state parameters include the yaw rate of change, the vehicle body sideslip rate of change, and the lateral acceleration rate of change.

[0055] Step S232: Perform trajectory spiral divergence analysis on the offset phase trajectory plane diagram to obtain trajectory spiral divergence data;

[0056] Step S233: Determine the trajectory singularity slope based on the trajectory spiral divergence data; calculate the trajectory arc approximation deviation based on the trajectory singularity slope and the trajectory spiral divergence data;

[0057] Step S234: Based on the yaw rate of change, the vehicle body side slip rate of change, and the lateral acceleration rate of change in the state parameters, the phase trajectory disorder degree is derived from the approximate deviation of the trajectory arc to obtain the phase trajectory disorder degree.

[0058] In this embodiment of the invention, step S231 extracts the vehicle instability offset trajectory in the attitude phase plane instability state to construct an offset phase trajectory plane diagram; extracts the state parameters in the attitude phase plane instability state, wherein the state parameters include the yaw rate of change, the vehicle body slip angle of change, and the lateral acceleration rate of change. Specifically, a sequence of trajectory points is selected from the attitude phase plane instability state obtained in step S22, for example, 30 points, from (2,-1) to (6.26,-0.58), in meters per second (m / s) and meters per second squared (m / s²). First, the offset of each point is calculated. The offset is the vertical distance between the actual point and the stable trajectory (the straight line from (2,-1) to (0,-1)). For example, the first point (2,-1) has an offset of 0; the distance between the second point (3.1,-1) and the straight line is calculated using the vertical formula as |3.1-2-(-1+1)×(3.1-2) / (0-2)| / =1.1, in meters per second (m / s). The unstable offset trajectory is extracted into these distance sequences, such as 0, 1.1, 1.5, up to 4.26, all in meters per second (m / s), for a total of 30 values. An offset phase trajectory planar plot is constructed by plotting the offset values ​​as the y-axis and time as the x-axis, with time ranging from 0 to 3 seconds (s), one point every 0.1 seconds (s). The planar plot is a polyline connecting these points. Next, extract the state parameters: the yaw rate of change is the difference in angular velocity at each time step divided by time. For example, if the angular velocity changes from 0.5 radians per second (rad / s) to 0.6 radians per second (rad / s), the rate of change is (0.6-0.5) / 0.1=1, with units in radians per square second (rad / s²). The rate of change of the vehicle body slip angle needs to be converted from the angle unit to the international standard unit radians (rad). 5 degrees is converted to 5×π / 180≈0.0873 radians (rad), and 6 degrees is converted to 6×π / 180≈0.1047 radians (rad). Therefore, the rate of change (0.104... 7-0.0873) / 0.1≈0.174, in radians per second (rad / s); the rate of change of lateral acceleration ranges from -1 m / s² to -0.9 m / s², resulting in (-0.9-(-1)) / 0.1=1, in meters per second³ (m / s³). The extraction logic is point-by-point difference calculation, taking the sequence average as the final parameter. For example, the average yaw rate of change is 0.8 radians per second (rad / s²), the lateral slip rate of change is approximately 0.174 radians per second (rad / s), and the acceleration rate of change is 1 m / s³ (m / s³). In the experiment, there were 30 points of instability trajectory. The maximum offset obtained from the extracted trajectory was 4 m / s. Straight lines connect the points in the planar diagram. The state parameters are extracted from the same sequence. The bottom-level parameters have a time step of 0.1 seconds (s) and a slope of 0 for the stable line (unitless). The calculation is implemented through difference operations.

[0059] A spiral divergence analysis is performed on the offset phase trajectory plane diagram to obtain spiral divergence data. Specifically, a point sequence is selected from the offset phase trajectory plane diagram constructed in step S231, for example, 30 points, with offsets ranging from 0 to 4.26 meters per second (m / s) and times ranging from 0 to 3 seconds (s). First, the spiral pattern is identified by calculating the curvature change between points. The curvature is the reciprocal radius of the fitted circle for every three points, for example, point 1 (0,0), point 2 (0.1,1.1), and point 3 (0.2,1.5) (x-axis represents time in seconds; y-axis represents offset in meters per second). The radius is calculated using the three-point circle formula, resulting in 10 meters (m). The curvature is then 1 / 10 = 0.1, in units of 1 / m. Divergence analysis is performed by accumulating the curvature starting from the center point, which is the midpoint of the sequence (offset 2.13 meters per second (m / s), time 1.5 seconds (s)). The divergence value is calculated forward and backward, and the divergence value is the curvature multiplied by the distance. The distance calculation is based on the time interval between points and the average speed: the average speed every 0.1 seconds (s) is 2 meters per second (m / s), so the distance between points is 0.1s × 2m / s = 0.2 meters (m), therefore the divergence value is 0.1 × 0.2 = 0.02 (unitless). The process is segmented analysis, with 10 points per segment, for a total of 3 segments. The average curvature of each segment is calculated. For example, the average curvature of the first segment is 0.15, which means that for every 1 meter of arc length traversed, the tangent direction angle changes by 0.15 radians, reflecting the average curvature of this segment. The segment length is calculated based on time and speed: each segment corresponds to 1 second (s), with an average speed of 2 meters per second (m / s), so the segment length is 1s × 2m / s = 2 (m), therefore the divergence data for this segment is 0.15 × 2 = 0.3. The second segment is similarly calculated as 0.4, the third segment as 0.5, and the total divergence data is the vector (0.3, 0.4, 0.5).

[0060] The trajectory singularity slope is determined based on the trajectory spiral divergence data; the trajectory arc approximation deviation is calculated based on the trajectory singularity slope and the trajectory spiral divergence data. Specifically, the vector (0.3, 0.4, 0.5) (unitless) is extracted from the trajectory spiral divergence data obtained in step S232, and the singularity slope is determined as the average slope of the data points. First, the difference between adjacent data points is calculated: 0.4 - 0.3 = 0.1 (unitless), 0.5 - 0.4 = 0.1 (unitless). The slope is equal to the difference divided by the segment interval (segment interval is 1, unitless), resulting in 0.1 per segment (unitless / segment). Then, the average is taken to obtain 0.1 per segment. The divergence data is considered as a sequence, and the singularity is defined as the position (segment 3) of the maximum divergence point 0.5 (unitless). The slope is the overall slope from segment 1 to segment 3: (0.5 - 0.3) / (3 - 1) = 0.1 per segment (unitless / segment). Next, the approximate deviation of the trajectory arc is calculated: by multiplying the slope by the divergence average (divergence average = (0.3 + 0.4 + 0.5) / 3 = 0.4, unitless), the deviation is 0.1 × 0.4 = 0.04; then, the calculation is repeated for each divergence value, and the deviation vector is (0.1 × 0.3 = 0.03, 0.1 × 0.4 = 0.04, 0.1 × 0.5 = 0.05), with a total average deviation of 0.04; the calculation logic is to first determine the slope and then multiply and average the points one by one to ensure that the deviation reflects the divergence intensity. In the experiment, the divergence data is (0.3, 0.4, 0.5) (unitless), the slope of the singularity is determined to be 0.1 per segment, and the deviation is calculated to be 0.04 (unitless), with the bottom-level parameters segment interval 1 and number of points 3. The calculation is implemented through difference and multiplication operations, such as the slope formula. Where s is the slope of the singularity, 0.1 per segment. The last divergence value is 0.5. Let 0.3 be the first divergence value and n be the number of segments (3); the deviation formula is di = s ⋅ fi, where di is the i-th deviation value (e.g., 0.03, unitless), fi is the i-th divergence value (e.g., 0.3, unitless), and the total deviation is the average of all di values.

[0061] The average values ​​of the state parameters extracted in step S231 (yaw rate of change 0.8 radians per second squared (rad / s²), lateral slip rate of change approximately 0.174 radians per second (rad / s), and acceleration rate of change 1 meter per second cubic (m / s³)) are used. The deviation of 0.04 (unitless) is extracted in step S233. The derivation is achieved by multiplying the parameters by the deviation, with the weights set as yaw 0.5 (unitless), lateral slip 0.3 (unitless), and acceleration 0.2 (unitless), and the total weight is 1 (unitless). First, calculate the weighted sum: Yaw parameter weighted value = 0.8 rad / s² × 0.5 = 0.4, unit is rad / s²; Lateral slip parameter weighted value = 0.174 rad / s × 0.3 ≈ 0.0522, unit is rad / s; Acceleration parameter weighted value = 1 m / s³ × 0.2 = 0.2, unit is m / s³; Total weighted sum is 0.4 + 0.0522 + 0.2 ≈ 0.6522 (Comprehensive unit: due to different parameter units, this is the weighted combination of physical quantities; after multiplying with the unitless deviation, the degree of disorder is unitless). Then, the degree of disorder = total weighted sum × deviation = 0.6522 × 0.04 ≈ 0.0261. The process is a serialized derivation. If the parameters have a sequence (e.g., yaw sequence (0.7, 0.8, 0.9, unit: rad / s²), lateral yaw sequence (0.15, 0.174, 0.198, unit: rad / s), acceleration sequence (0.9, 1, 1.1, unit: m / s³), and a corresponding deviation sequence (0.03, 0.04, 0.05), then the calculation logic for each point is: first calculate the weighted sum of the parameters at that point, and then multiply it by the corresponding deviation. For example, point 1: (0.7×0.5+0.15×0.3+0.9×0.2)×0.03=(0.35+0.045+0.18)×0.03=0.575×0.03≈0.0173 (unitless); point 2: (0.8×0.5+0.174×0.3+1×0.2)×0.04≈(0.4+0.0522+0.2)×0.04≈0.6522× 0.04≈0.0261 (unitless); Point 3: (0.9×0.5+0.198×0.3+1.1×0.2)×0.05=(0.45+0.0594+0.22)×0.05=0.7294×0.05≈0.0365 (unitless); The final degree of disorder is the average of the three points, that is, (0.0173+0.0261+0.0365) / 3≈0.0266.

[0062] Step S24 includes the following steps:

[0063] Step S241: Decompose the disorder of the phase trajectory into disordered action postures to obtain the trajectory decomposition vector;

[0064] Step S242: Estimate the kernel density of potential collision points based on the trajectory decomposition vector to generate potential collision point probability distribution data;

[0065] Step S243: Apply temporal approximate variance constraints to the trajectory decomposition vectors to obtain the dynamic parameter constraint variance; wherein the dynamic parameters include vehicle trajectory yaw rate, sideslip angle, and acceleration;

[0066] Step S244: Based on the probability distribution data of potential collision points and the variance of dynamic parameter constraints, the lateral swing curvature correlation is derived to obtain the trajectory swing curvature associated with the collision points;

[0067] Step S245: Perform collision attitude trajectory analysis based on the variance of dynamic parameter constraints and the trajectory swing curvature associated with the collision point to generate the collision attitude trajectory.

[0068] In this embodiment of the invention, principal component analysis (PCA) is used to decompose the disordered motion attitude of the phase trajectory. First, a 3D orthogonal decomposition space is constructed, with axes corresponding to yaw, sideslip, and vertical motion, respectively. An attitude matrix is ​​established, with each element consisting of 200 sampling points of the phase plane trajectory, each point containing position and velocity information. Singular value decomposition (SVD) is performed on the attitude matrix to calculate eigenvalues ​​and eigenvectors. The three principal eigenvalues ​​are measured to be 25.8, 15.4, and 8.6, corresponding to eigenvectors [0.82, 0.42, 0.38], [0.35, 0.75, 0.55], and [0.45, 0.51, 0.73]. The contribution rates of each principal component are calculated based on the eigenvalues, which are 51.6%, 30.8%, and 17.6%, respectively. The phase trajectory disorder level of 0.5 is allocated to the three eigenvectors according to their contribution rates, resulting in the trajectory decomposition vector [0.258, 0.154, 0.088]. This vector represents the disorder components in the three dimensions of yaw motion, sideslip motion, and vertical motion, with the yaw motion component being the largest and the vertical motion component being the smallest.

[0069] The kernel density of potential collision points is estimated based on the trajectory decomposition vector [0.258, 0.154, 0.088]. The obstacle is a stationary pedestrian located 28m in front of the vehicle, with a lateral offset of 1.5m. First, a set of possible vehicle trajectories is generated using the Monte Carlo method, producing 500 random trajectories, each containing 20 points, with a time interval of 0.1s, covering the next 2s. Trajectory generation considers the disorder component in the trajectory decomposition vector; the yaw component affects the steering angle change, the sideslip component affects the tire lateral force, and the vertical component affects the suspension response. A 30×10m rectangular area is constructed in front of the vehicle, with a grid size of 0.5×0.5m and a total of 1200 grid points. A two-dimensional Gaussian kernel function is applied to each grid point to calculate the collision probability density, with a kernel bandwidth matrix of [1.8m, 0.0m; 0.0m, 0.9m]. The formula for calculating the collision probability density is: Where K is the Gaussian kernel function, ( , () represents the coordinates of the trajectory point. and Here, represents the bandwidth parameters (1.8m and 0.9m), and n represents the total number of trajectory points. The calculation results show that the region with the highest collision probability is located within 1.5m of the obstacle, with a maximum probability density value of 0.28.

[0070] Temporal approximate variance constraints are applied to the trajectory decomposition vector [0.258, 0.154, 0.088] for the dynamic parameters. First, physical limit constraints are set for the dynamic parameters: yaw rate limit is ±30° / s, which translates to ±30×π / 180≈±0.5236 (rad / s) in SI units; sideslip angle limit is ±12 degrees, which translates to ±12×π / 180≈±0.2094 (rad) in SI units; lateral acceleration limit is ±10 m / s². A noise covariance matrix Q is constructed by combining the vehicle state equation and the trajectory decomposition vector. The diagonal elements of the noise covariance matrix are the squares of the trajectory decomposition vector components multiplied by a scale factor of 15. Specifically, the first element is 0.258²×15≈0.066564×15≈0.99846, and the second element is 0.154²×15≈... 0.023716×15≈0.35574, the third element 0.088²×15≈0.007744×15≈0.11616, then the state covariance matrix is ​​solved using the discrete Lyapunov equation. After 10 iterations, the steady-state covariance matrix P is obtained. The variance values ​​of the dynamic parameters are extracted from P: the yaw rate variance is 28.5°², which is converted to SI units of 28. 5×(π / 180)²≈28.5×(0.0174533)²≈28.5×0.0003046≈0.008681 (rad² / s²); the variance of the sideslip angle is 6.8°², which, converted to SI units, is 6.8×(π / 180)²≈6.8×0.0003046≈0.002071 (rad²); the variance of the lateral acceleration is 3.2 m² / s². 4 These variance values ​​characterize the fluctuation range of each dynamic parameter under the influence of the disordered trajectory decomposition vector, providing constraints for subsequent path planning.

[0071] The lateral sway curvature correlation derivation is performed based on the probability distribution data of potential collision points and the variance of dynamic parameter constraints, using a curvature-probability correlation analysis method. First, the path curvature from the vehicle's current position to each potential collision point is calculated. The curvature calculation formula is: Where (x0, y0) is the current position of the vehicle, , Let (28, 1.5) be the coordinates of the potential collision point. For the region with the highest collision probability (probability density 0.28), the coordinates are approximately (28, 1.5), in meters. The calculated path curvature is 0.0375 / m. Considering the influence of the variance of dynamic parameter constraints on curvature, the yaw rate variance of 28.5 rad² / s² corresponds to a curvature variance of 0.000625 / m², the sideslip angle variance of 6.8 rad² corresponds to a curvature variance of 0.000484 / m², and the lateral acceleration variance is 3.2 m² / s². 4The corresponding curvature variance is 0.000324 / m². These curvature variances are weighted and summed with weights of 0.5, 0.3, and 0.2, yielding a comprehensive curvature variance of 0.000518 / m². The baseline curvature of 0.0375 / m is added to the standard deviation of curvature of 0.0228 / m, resulting in a trajectory swing curvature associated with the collision point of 0.0603 / m. In this embodiment, collision attitude trajectory analysis is performed based on the dynamic parameter constraint variance and the trajectory swing curvature associated with the collision point. A time step of 0.1 s is used, with a total computation time of 2.5 s, generating 25 trajectory points. The initial state is set as follows: position (0m, 0m), velocity 15.5 m / s, heading angle 0 rad, yaw rate 2.5 rad / s, and sideslip angle 1.8 rad. Within each time step, the state at the next moment is predicted using a state equation, which includes updates to the yaw rate, sideslip angle, and position. Considering the influence of the variance of the dynamic parameter constraints, a random perturbation is added to each prediction step. The perturbation magnitude follows a normal distribution with a mean of 0 and a variance equal to the aforementioned variance of the dynamic parameter constraints. The trajectory yaw curvature of 0.0603 / m is converted into a path radius of 16.6m, limiting the vehicle's steering ability. Simulation results show that the vehicle reaches a lateral displacement of 1.4m, a yaw rate of 18.5 rad / s, and a sideslip angle of 6.5 rad at 1.7s; it collides with an obstacle at 2.1s at a collision location of (28m, 1.5m), with a collision velocity of 13.8m / s and a heading angle of 23 rad. The generated collision attitude trajectory completely records the vehicle's motion from the current state to the collision state.

[0072] Step S3 includes the following steps:

[0073] Step S31: Perform feature learning on the collision attitude trajectory to obtain the collision prediction feature trajectory;

[0074] Step S32: Perform self-attention weight aggregation on the collision prediction feature trajectory to obtain the collision prediction weight trajectory;

[0075] Step S33: Based on the collision prediction weight trajectory and obstacle orientation, perform emergency avoidance safety braking parameter matching on the vehicle braking process state to obtain avoidance braking safety matching parameters;

[0076] Step S34: Based on the collision avoidance braking safety matching parameters and collision prediction weight trajectory, perform collision avoidance path planning learning to generate a collision avoidance planning path.

[0077] As an example of the present invention, reference is made to Figure 3 As shown, step S3 in this example includes:

[0078] Step S31: Perform feature learning on the collision attitude trajectory to obtain the collision prediction feature trajectory;

[0079] In this embodiment of the invention, the collision attitude trajectory generated in step S2, consisting of vehicle attitude data points {horizontal coordinate, vertical coordinate, yaw angle, and body slip angle} at 100 consecutive time points (interval of 0.02s) within the next 2 seconds, is input into a preset sliding window analysis program. This program sets the window width to 10 data points and the sliding step size to 5 data points. Within each window, three physical quantities are extracted: the maximum curvature of the trajectory segment, the peak yaw rate, and the average rate of change of the body slip angle. For example, for the 10 data points in the first window, the calculated curvature is 0.05. The peak yaw rate is 15 rad / s, and the average rate of change of sideslip angle is 3 rad / s. After completing the calculations for all windows, the resulting 19 sets of feature vector sequences {maximum curvature, peak yaw rate, average rate of change of sideslip angle} are output as the collision prediction feature trajectory.

[0080] Step S32: Perform self-attention weight aggregation on the collision prediction feature trajectory to obtain the collision prediction weight trajectory;

[0081] In this embodiment of the invention, the 19 sets of feature vector sequences generated in step S31 are processed. The 10th set of feature vectors is taken as the target vector for the current analysis. The dot product similarity score between the target vector and all 19 feature vectors, including itself, is calculated. For example, the dot product scores between the 10th set of vectors and the 9th, 10th, and 11th sets of vectors (representing the period of rapid change due to instability) are 0.92, 0.99, and 0.91, respectively, while the scores with the 1st and 2nd sets of vectors (representing the initial stage of instability) are 0.35 and 0.38, respectively. These 19 scores are normalized to obtain a set of weight coefficients, the sum of which is 1. This set of weight coefficients is used to perform a weighted summation on the original 19 sets of feature vectors to generate a new "weighted aggregated" feature vector. This process is repeated for all 19 sets of feature vectors to finally obtain a new sequence composed of 19 "weighted aggregated" feature vectors, i.e., the collision prediction weight trajectory.

[0082] Step S33: Based on the collision prediction weight trajectory and obstacle orientation, perform emergency avoidance safety braking parameter matching on the vehicle braking process state to obtain avoidance braking safety matching parameters;

[0083] In this embodiment of the invention, braking parameters are matched based on the highest-risk feature vector (e.g., a peak yaw rate of 30 rad / s) in the collision prediction weight trajectory generated in step S32, combined with the obstacle's orientation (12m directly in front). First, a safe distance threshold of 2m is set. Based on the vehicle's current speed of 20m / s, the minimum average deceleration required to maintain a 2m safe distance is calculated to be 8.3m / s². This deceleration is used as the target and searched in a preset braking pressure distribution map. This map records the vehicle response under different combinations of {inner rear wheel braking pressure, outer front wheel braking pressure}. A set of parameters is obtained: inner rear wheel braking pressure 6.1MPa, outer front wheel braking pressure 8.2MPa. This combination can achieve a deceleration of not less than 8.3m / s² without increasing yaw instability. This set of pressure values ​​{6.1MPa, 8.2MPa} are the collision avoidance braking safety matching parameters.

[0084] Step S34: Based on the collision avoidance braking safety matching parameters and collision prediction weight trajectory, perform collision avoidance path planning learning to generate a collision avoidance planning path.

[0085] In this embodiment of the invention, path generation is performed based on the collision avoidance braking safety matching parameters {inner rear wheel braking pressure: 6.1 MPa, outer front wheel braking pressure: 8.2 MPa} obtained in step S33 and the collision prediction weight trajectory. The current state of the vehicle (position, speed 20 m / s, attitude angle, etc.) and the braking parameters are input into a vehicle dynamics forward simulation program. The program iteratively calculates the vehicle motion within the next 2.5 seconds in 0.01 s increments. At each increment, the program calculates the wheel slip ratio and tire force based on the braking pressure, then solves for the forces and moments of the vehicle body, updates the linear acceleration and angular acceleration of the vehicle, and finally integrates to obtain the velocity and position at the next moment. Connecting the sequence of vehicle centroid coordinates (x, y) at these 250 consecutive moments forms a precise, physically achievable trajectory, which is the generated collision avoidance planning path.

[0086] Step S33 includes the following steps:

[0087] Step S331: Match the safe braking parameters between the obstacle distance changes in the vehicle braking process state according to the collision prediction weight trajectory and obstacle orientation to obtain distance-related safe braking parameters; wherein the distance-related safe braking parameters include: distance-related safe braking speed, time and wheel lock-up threshold.

[0088] Step S332: Based on the distance-related safe braking speed, time and wheel lock-up threshold in the distance-related safe braking parameters, perform a timing allocation of steering braking force on the inner rear wheel and outer front wheel to the collision prediction weight trajectory and vehicle braking process state, and obtain steering braking force allocation data.

[0089] Step S333: Perform torque vector regression analysis and coupling based on steering braking force distribution data to obtain the torque coupling vector;

[0090] Step S334: Based on the steering braking force distribution data and torque coupling vector, adjust the vehicle trajectory stability attitude of the collision prediction weight trajectory and the vehicle braking process state to obtain vehicle stability attitude adjustment data;

[0091] Step S335: Based on the vehicle stability attitude control data, steering braking force distribution data and torque coupling vector, perform emergency avoidance safety braking parameter matching to obtain avoidance braking safety matching parameters.

[0092] In this embodiment of the invention, the vehicle's current braking state (vehicle speed 25 m / s) and the obstacle's location (40 m directly ahead) are obtained. Based on the collision prediction weighted trajectory, the actual path length to the obstacle is calculated to be 42 m. A fixed, inviolable final safe distance of 2 m is set, indicating that the vehicle must come to a complete stop within a path distance of 40 m (42 m - 2 m). According to the kinematic relationship a = v² / 2d, the target average deceleration required to stop within this distance is calculated as a = (25 m / s)² / (2 × 40 m) = 7.8125, in m / s². Based on this deceleration and the initial velocity, the time required to complete braking is calculated as t = v / a = 25 m / s / 7.8125 m / s² = 3.2 seconds. Based on the vehicle's dynamic parameters and the current road surface adhesion coefficient (e.g., 0.8), the maximum wheel slip ratio threshold for achieving this deceleration without causing complete wheel lock-up is determined to be 18%. Ultimately, the distance-related safe braking parameters obtained are {distance-related safe braking speed: 25m / s, time: 3.2s, wheel lock-up threshold: 18% slip rate}.

[0093] Based on the obtained distance-related safety braking parameters {braking speed 25 m / s, time 3.2 s, lock-up threshold 18%}, the braking force for the next 3.2 seconds is time-sequentially allocated. The 3.2-second braking process is discretized into 32 time steps, each 0.1 s. At each time step, based on the target curvature determined by the collision prediction weight trajectory and the vehicle model (vehicle mass 1600 kg, wheelbase 2.8 m), the yaw moment and total longitudinal braking force required to achieve trajectory tracking are calculated. For example, at 1.5 s, the vehicle needs to generate a total braking force of -12000 N and a yaw moment of 500 N·m. By solving the mechanical equations, this requirement is allocated to the outer front wheel and the inner rear wheel on the steering side. The calculation results are: outer front wheel braking force of 7357 N and inner rear wheel braking force of 4643 N. Simultaneously, the wheel slip ratio corresponding to the calculated braking force is monitored to ensure that it does not exceed the 18% threshold. The sequence of {time point, outer front wheel braking force, inner rear wheel braking force} calculated from all 32 time steps constitutes the steering braking force distribution data.

[0094] Based on the steering braking force distribution data generated in step S332, the braking force is converted into actuator torque. Taking the data at 1.5s as an example: the required braking force for the outer front wheel is 7357N, the required braking force for the inner rear wheel is 4643N, and the dynamic radius of the wheel is 0.32m. The braking torque for the outer front wheel is achieved through the hydraulic braking system, and the required torque is 7357N × 0.32m = 2354.24N·m. The inner rear wheel is braked jointly by the drive motor and the hydraulic brake. A preset two-dimensional lookup table is queried, indexed by vehicle speed and required braking force, to output the optimal allocation of regenerative braking torque from the motor and hydraulic braking torque. At the current vehicle speed (e.g., 15 m / s), the motor can provide a maximum regenerative braking torque of 1200 N·m (corresponding to 3750 N braking force). The remaining 893 N braking force (4643 N - 3750 N) is provided by the hydraulic brake, corresponding to a torque of 893 N × 0.32 m = 285.76 N·m. Record the torque distribution at this moment: {outer front wheel hydraulic: 2354.24, inner rear wheel regenerative: 1200, inner rear wheel hydraulic: 285.76} (unit: N·m). This is the torque coupling vector at that moment.

[0095] Operating as a closed-loop feedback controller, the execution cycle is 20ms. At 1.52s, the actual yaw rate of the vehicle is obtained from the onboard inertial measurement unit (IMU) and GPS as 12.5rad / s, and the sideslip angle is -2.1rad. The reference values ​​for the collision prediction weight trajectory at this moment are: yaw rate 12.0rad / s, sideslip angle -2.0rad. The calculated errors are: yaw rate error e_ω = 12.5 - 12.0 = 0.5 (rad / s), sideslip angle error e_β = -2.1 - (-2.0) = -0.1 (rad). The errors are input into a preset proportional-integral-derivative (PID) controller with parameters Kp = 80, Ki = 10, Kd = 5. The controller outputs a corrected yaw moment. Calculations yielded =42 N·m. This corrective torque will be used to fine-tune the braking force distribution at the next moment. This series of data streams calculated over time {time point, yaw rate error, sideslip angle error, corrective yaw torque} constitutes the vehicle stability attitude control data.

[0096] Within each control cycle (20ms), the results of previous calculations are integrated to generate the final actuator control command. Taking the 1.52s control cycle as an example, the basic braking force distribution command at that moment is first obtained by interpolation from the steering braking force distribution data in step S332. Subsequently, the corrected yaw moment is obtained from the vehicle stability attitude control data in step S334. =42 N·m. To achieve this corrective torque, an equal and opposite force ΔF needs to be applied to the inner and outer wheels. The wheelbase is calculated to yield ΔF = 42 N·m / 1.6m = 26.25 N. This corrective force is then added to the base braking force: the braking force on the outer front wheel decreases by 26.25 N, and the braking force on the inner rear wheel increases by 26.25 N. Finally, the corrected braking force is used to calculate the final regenerative braking torque of the motor and the braking pressure command of the hydraulic brake disc through the torque coupling conversion logic described in step S333. This final output set of {motor target torque, brake master cylinder target pressure} values, which integrates feedforward and feedback control, represents the safety matching parameters for avoidance braking at that moment.

[0097] Step S34 includes the following steps:

[0098] Step S341: Based on the avoidance braking safety matching parameters and collision prediction weight trajectory, perform avoidance path passability space weight analysis to obtain the avoidance path passability space weight;

[0099] Step S342: Based on the avoidance braking safety matching parameters, perform iterative simulation of the controllability of vehicle braking parameters associated with the traversable space weight of the avoidance path to obtain a controllable dataset of braking parameters on the spatial path;

[0100] Step S343: Perform a controllable parameter optimal solution decision on the controllable dataset of braking parameters to obtain controllable optimal solution parameters;

[0101] Step S344: Based on the passable space weights of the avoidance path and the controllable optimal solution parameters, perform collision avoidance path planning learning to generate a collision avoidance planning path.

[0102] In one embodiment, based on the collision avoidance braking safety matching parameters {inner rear wheel braking pressure: 6.1MPa, outer front wheel braking pressure: 8.2MPa} and the collision prediction weight trajectory, a grid map covering a range of 40m × 12m is constructed, with a grid size of 0.2m × 0.2m. A passage weight value is calculated for each grid, with an initial value of 1.0. The obstacle area (coordinate point set {(25,2),(26,2),(25,3),(26,3)}), in meters, has a weight of 0. The weight within a 1.5m radius decreases with distance: 0.1 within 0.5m of the obstacle, 0.3 within 0.5-1m, and 0.6 within 1-1.5m. The weight of areas with a ground friction coefficient of 0.8 remains unchanged, while the weight of areas with a friction coefficient of 0.6 (coordinate range x∈[15,20], y∈[0,4]) decreases by 0.25. The weights of the collision prediction trajectory are increased within 1.8m on both sides, increased by 0.4 within 0.5m of the trajectory, increased by 0.3 within 0.5-1m, and increased by 0.15 within 1-1.8m. The weights in the boundary region (y<0.8 or y>11.2) are decreased by 0.5. The avoidance path passability space weight matrix is ​​obtained by smoothing with a 5×5 Gaussian convolution kernel.

[0103] Based on the collision avoidance braking safety matching parameters, 300 candidate paths are generated in a grid space with a weight greater than 0.7. Each path has 30 points, spaced 1.2m apart. Vehicle parameters are set as follows: mass 1680kg, wheelbase 2.78m, front and rear track 1.58m, center of gravity height 0.54m, and tire radius 0.31m. Iterative simulations of braking parameter controllability are performed for each path, with a step size of 0.01s and a total simulation time of 2.5s. Initial conditions: vehicle speed 20m / s, heading angle 0 degrees, sideslip angle 0 degrees, and yaw rate 0rad / s. In each iteration, the inner rear wheel braking pressure is 6.1MPa, and the outer front wheel braking pressure is 8.2MPa. Tire forces are calculated: longitudinal force is obtained from a table using the slip ratio-force curve, with a maximum static friction force of 5600N; lateral force is obtained from the sideslip angle-force curve, with a maximum lateral force of 4800N. Paths with a wheel slip ratio exceeding 0.18 or a sideslip angle exceeding 7.5 degrees during iteration are determined to be uncontrollable. For each controllable path, changes in position, velocity, acceleration, and braking pressure are recorded to form a controllable braking parameter dataset.

[0104] Optimal solution decisions were made for 175 paths in a dataset with controllable braking parameters. An evaluation function was constructed, comprising four indicators: safety distance indicator D (minimum distance to obstacles, target value greater than 3m), trajectory smoothness indicator S (maximum curvature, target value less than 0.22 / m), braking efficiency indicator E (braking distance, target value less than 30m), and ride comfort indicator C (maximum combined acceleration, target value less than 4.5m / s²). The evaluation function was F = 0.35 × D + 0.25 × S + 0.25 × E + 0.15 × C, with a maximum score of 100. For each path, the vehicle dynamics equations were substituted to calculate the values ​​of each indicator, which were solved using the fourth-order Runge-Kutta method. The indicators were standardized to a score of 0-100; for example, a safety distance of 3m received 60 points, and 5m received 100 points; a curvature of 0.22 / m received 60 points, and 0.15 / m received 100 points. Calculate the comprehensive score of 175 paths, select the parameters corresponding to the path with the highest score, which are the controllable optimal solution parameters, including the braking pressure and steering angle sequence of 30 discrete points.

[0105] Based on the collapsible space weights and controllable optimal solution parameters, a collision avoidance planning path is generated. First, the controllable optimal solution parameters are used as initial seeds, and perturbed by ±3% in both braking pressure and steering angle dimensions, generating 20 sets of variability parameters. Each set of parameters is simulated forward using vehicle dynamics equations with a step size of 0.01s and a duration of 2.5s. The dynamics equations include vehicle dynamics, tire dynamics, and steering system dynamics, totaling 12 state variables. To ensure accuracy, double-precision floating-point calculations are used. Path integration is performed on the 20 trajectories and the collapsible space weight matrix of the avoidance path to calculate the total weight score. The trajectory with the highest weight score is selected and smoothed using cubic spline interpolation with a sampling interval of 0.1m. This path satisfies the following conditions: maximum lateral acceleration not exceeding 3.8m / s², maximum longitudinal deceleration not exceeding 6.5m / s², and steering angle change rate not exceeding 6 degrees / s. The final output contains a collision avoidance planning path with 250 time points, each point including position, velocity, acceleration, and attitude angle information.

[0106] The present invention also provides an intelligent vehicle path planning system for executing the intelligent vehicle path planning method described above, the intelligent vehicle path planning system comprising:

[0107] The data extraction module is used to extract the location of obstacles and the vehicle braking status during the emergency avoidance process of an intelligent vehicle from the intelligent vehicle control center.

[0108] The collision trajectory prediction module is used to perform attitude lateral phase plane instability analysis on the vehicle braking process to obtain the attitude phase plane instability state; based on the attitude phase plane instability state and the obstacle orientation, it performs collision attitude trajectory analysis to generate the collision attitude trajectory.

[0109] The collision avoidance path planning learning module is used to match emergency avoidance safety braking parameters to the vehicle braking process state based on the collision attitude trajectory, and obtain collision avoidance braking safety matching parameters; based on the collision avoidance braking safety matching parameters and the collision attitude trajectory, collision avoidance path planning learning is performed to generate a collision avoidance planning path.

[0110] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A method for intelligent vehicle path planning, characterized in that, Includes the following steps: Step S1: Extract the location of obstacles and the vehicle braking status during the emergency avoidance process of the intelligent vehicle from the intelligent vehicle control center. Step S2: Perform attitude lateral phase plane instability analysis on the vehicle braking process to obtain the attitude phase plane instability state; Based on the attitude phase plane instability state and obstacle orientation, a collision attitude trajectory is analyzed to generate a collision attitude trajectory; wherein, step S2 includes: Step S21: Analyze the runaway speed fluctuation during the vehicle braking process to obtain the runaway speed fluctuation. Step S22: Perform attitude lateral phase plane instability analysis on the vehicle braking process state based on the runaway fluctuation velocity to obtain the attitude phase plane instability state; wherein, step S22 includes: Step S221: Obtain basic information about the intelligent vehicle; wherein the basic information includes the intelligent vehicle body structure, body weight distribution and suspension design structure; Step S222: Extract the brake steering ratio from the vehicle braking process state; perform numerical incremental integration of the steering centrifugal torque on the intelligent vehicle body structure and body weight distribution in the basic information based on the brake steering ratio and the runaway fluctuation speed to obtain the steering centrifugal torque increment data; Step S223: Analyze the incremental changes in the kingpin caster angle and camber angle of the suspension design structure using the incremental data of the steering centrifugal torque. Step S224: Based on the incremental data of steering centrifugal torque and the incremental changes of caster angle and camber angle, derive the incremental gradient of vehicle pitch angle to obtain the attitude increment pitch angle; Step S225: Based on the incremental data of steering centrifugal torque, the incremental changes of the kingpin caster angle and camber angle, and the incremental pitch angle, perform attitude lateral phase plane instability analysis to obtain the attitude phase plane instability state. Step S23: Derive the degree of disorder in the phase trajectory of the attitude phase plane instability state; Step S24: Based on the disorder of the phase trajectory and the orientation of the obstacle, perform collision attitude trajectory analysis to generate a collision attitude trajectory; Step S3: Match emergency avoidance safety braking parameters to the vehicle braking process state based on the collision attitude trajectory to obtain avoidance braking safety matching parameters; perform collision avoidance path planning learning based on the avoidance braking safety matching parameters and the collision attitude trajectory to generate a collision avoidance planning path.

2. The intelligent vehicle path planning method according to claim 1, characterized in that, Step S23 includes the following steps: Step S231: Extract the vehicle instability offset trajectory in the attitude phase plane instability state to construct an offset phase trajectory plane diagram; extract the state parameters in the attitude phase plane instability state, wherein the state parameters include the yaw rate of change, the vehicle body sideslip rate of change, and the lateral acceleration rate of change. Step S232: Perform trajectory spiral divergence analysis on the offset phase trajectory plane diagram to obtain trajectory spiral divergence data; Step S233: Determine the trajectory singularity slope based on the trajectory spiral divergence data; calculate the trajectory arc approximation deviation based on the trajectory singularity slope and the trajectory spiral divergence data; Step S234: Based on the yaw rate of change, the vehicle body side slip rate of change, and the lateral acceleration rate of change in the state parameters, the phase trajectory disorder degree is derived from the approximate deviation of the trajectory arc to obtain the phase trajectory disorder degree.

3. The intelligent vehicle path planning method according to claim 1, characterized in that, Step S24 includes the following steps: Step S241: Decompose the disorder of the phase trajectory into disordered action postures to obtain the trajectory decomposition vector; Step S242: Estimate the kernel density of potential collision points based on the trajectory decomposition vector to generate potential collision point probability distribution data; Step S243: Apply temporal approximate variance constraints to the trajectory decomposition vectors to obtain the dynamic parameter constraint variance; wherein the dynamic parameters include vehicle trajectory yaw rate, sideslip angle, and acceleration; Step S244: Based on the probability distribution data of potential collision points and the variance of dynamic parameter constraints, the lateral swing curvature correlation is derived to obtain the trajectory swing curvature associated with the collision points; Step S245: Perform collision attitude trajectory analysis based on the variance of dynamic parameter constraints and the trajectory swing curvature associated with the collision point to generate the collision attitude trajectory.

4. The intelligent vehicle path planning method according to claim 1, characterized in that, Step S3 includes the following steps: Step S31: Perform feature learning on the collision attitude trajectory to obtain the collision prediction feature trajectory. Feature learning on the collision attitude trajectory refers to inputting the collision attitude trajectory, composed of vehicle attitude data points {horizontal coordinate, vertical coordinate, yaw angle, and body slip angle} from 100 consecutive time points within the next 2 seconds, into a preset sliding window analysis program. This program sets the window width to 10 data points and the sliding step size to 5 data points. Within each window, it extracts three physical quantities: the maximum curvature of the trajectory segment, the peak yaw rate, and the average rate of change of the body slip angle. For the 10 data points in the first window, the calculated curvature is 0.05m. -1 The peak yaw rate is 15 rad / s, and the average rate of change of sideslip angle is 3 rad / s. After completing the calculations for all windows, the resulting 19 sets of feature vector sequences {maximum curvature, peak yaw rate, average rate of change of sideslip angle} are used as the collision prediction feature trajectory output. Step S32: Perform self-attention weight aggregation on the collision prediction feature trajectory to obtain the collision prediction weight trajectory; wherein, self-attention weight aggregation refers to: processing the 19 sets of feature vector sequences generated in step S31, taking the 10th set of feature vectors as the target vector for the current analysis, calculating the dot product similarity score between the target vector and all 19 feature vectors including itself, the dot product scores between the 10th set of vectors and the 9th, 10th, and 11th set of vectors are 0.92, 0.99, and 0.91 respectively, while the scores with the 1st and 2nd set of vectors are 0.35 and 0.38 respectively, these 19 scores are normalized to obtain a set of weight coefficients, and the original 19 sets of feature vectors are weighted and summed using this set of weight coefficients to generate a new "weighted aggregated" feature vector, repeating this process for all 19 sets of feature vectors, and finally obtaining a new sequence composed of 19 "weighted aggregated" feature vectors, that is, the collision prediction weight trajectory; Step S33: Based on the collision prediction weight trajectory and obstacle orientation, perform emergency avoidance safety braking parameter matching on the vehicle braking process state to obtain avoidance braking safety matching parameters; Step S34: Based on the collision avoidance braking safety matching parameters and collision prediction weight trajectory, perform collision avoidance path planning learning to generate a collision avoidance planning path.

5. The intelligent vehicle path planning method according to claim 4, characterized in that, Step S33 includes the following steps: Step S331: Match the safe braking parameters between the obstacle distance changes in the vehicle braking process state according to the collision prediction weight trajectory and obstacle orientation to obtain distance-related safe braking parameters; wherein the distance-related safe braking parameters include: distance-related safe braking speed, time and wheel lock-up threshold. Step S332: Based on the distance-related safe braking speed, time and wheel lock-up threshold in the distance-related safe braking parameters, perform a timing allocation of steering braking force on the inner rear wheel and outer front wheel to the collision prediction weight trajectory and vehicle braking process state, and obtain steering braking force allocation data. Step S333: Perform torque vector regression analysis and coupling based on steering braking force distribution data to obtain the torque coupling vector; Step S334: Based on the steering braking force distribution data and torque coupling vector, adjust the vehicle trajectory stability attitude of the collision prediction weight trajectory and the vehicle braking process state to obtain vehicle stability attitude adjustment data; Step S335: Based on the vehicle stability attitude control data, steering braking force distribution data and torque coupling vector, perform emergency avoidance safety braking parameter matching to obtain avoidance braking safety matching parameters.

6. The intelligent vehicle path planning method according to claim 4, characterized in that, Step S34 includes the following steps: Step S341: Based on the avoidance braking safety matching parameters and collision prediction weight trajectory, perform avoidance path passability space weight analysis to obtain the avoidance path passability space weight; Step S342: Based on the avoidance braking safety matching parameters, perform iterative simulation of the controllability of vehicle braking parameters associated with the traversable space weight of the avoidance path to obtain a controllable dataset of braking parameters on the spatial path; Step S343: Perform a controllable parameter optimal solution decision on the controllable dataset of braking parameters to obtain controllable optimal solution parameters; Step S344: Based on the passable space weights of the avoidance path and the controllable optimal solution parameters, perform collision avoidance path planning learning to generate a collision avoidance planning path.

7. An intelligent vehicle path planning system, characterized in that, For performing the intelligent vehicle path planning method as described in claim 1, the intelligent vehicle path planning system includes: The data extraction module is used to extract the location of obstacles and the vehicle braking status during the emergency avoidance process of an intelligent vehicle from the intelligent vehicle control center. The collision trajectory prediction module is used to perform attitude lateral phase plane instability analysis on the vehicle braking process to obtain the attitude phase plane instability state; based on the attitude phase plane instability state and the obstacle orientation, it performs collision attitude trajectory analysis to generate the collision attitude trajectory. The collision avoidance path planning learning module is used to match emergency avoidance safety braking parameters to the vehicle braking process state based on the collision attitude trajectory, and obtain collision avoidance braking safety matching parameters; based on the collision avoidance braking safety matching parameters and the collision attitude trajectory, collision avoidance path planning learning is performed to generate a collision avoidance planning path.

Images