An Unmanned Vehicle Path Planning Method Based on an Improved Driving Safety Field

Abstract

The present invention relates to a path planning method for autonomous vehicles based on an improved driving safety field. Under the full consideration of complex traffic situations, this method proposes and verifies a path planning strategy and a control algorithm that meet the safety indicators of autonomous vehicles. It includes four major steps: (1) Detect the road traffic environment and calculate the road field strength distribution according to the driving safety field model; (2) Divide the sampling space, use the driving safety field as the weight for importance sampling, sample the road along the y-axis, and construct a matrix of discrete coordinate points of the road; (3) Calculate the risk evaluation index value - field force of each discrete coordinate point, and construct a driving risk matrix for autonomous vehicles; (4) Select the subscript of the sample point with the minimum field force from each layer of the sample space to obtain a sequence of path points with the minimum driving risk, and generate a path curve for the autonomous vehicle based on this. Based on this method, more accurate and reliable path planning for autonomous vehicles can be achieved.

Description

Technical Field

[0001] The present invention is directed to an autonomous vehicle, and proposes a path planning method for autonomous vehicles based on an improved driving safety field. The purpose is to propose and verify a path planning strategy and control algorithm that meet the safety indicators of autonomous vehicles under the full consideration of complex traffic situations, so as to achieve more accurate and reliable path planning for autonomous vehicles and ensure the safe driving of autonomous vehicles. The present invention belongs to the technical field of reliability engineering. Background Art

[0002] With the development of technology and the continuous progress of autonomy and intelligence, the development status of autonomous vehicles has gradually become an important indicator to measure the scientific and technological development level of a country. Compared with manned vehicles, autonomous vehicles have great advantages in reducing casualties caused by traffic accidents, reducing energy consumption and exhaust emissions, and improving traffic efficiency. Among them, path planning plays the role of a bridge for information perception and intelligent control in autonomous driving and is the basis of autonomous driving. Only when an autonomous vehicle ensures safety and reliability can it truly provide a good and convenient travel experience for the public and be accepted by the public. During the process of an autonomous vehicle driving on the road, accurate and reliable path planning can effectively avoid road congestion and reduce the traffic accident rate.

[0003] The traditional discrete sampling method is to uniformly sample along the center line of the lane on a structured road. However, the conditions of different parts of the road are different in actual driving, and the corresponding potential risks are also different. Importance sampling can sample more densely in areas with complex road conditions and more sparsely in areas with simple road conditions according to the weight, so as to improve the efficiency of route planning, and can also improve the accuracy and rationality of route planning.

[0004] In view of this, the present invention uses the field strength value of an improved driving safety field for autonomous vehicles (see 202210025985.8, a method for constructing a driving safety field model for autonomous vehicles) as the weight, and discretely samples the road surface and the environmental information of the road surface where the autonomous vehicle travels according to the principle of importance sampling to form a matrix of discrete point coordinates of the road, and calculates and outputs a set of coordinates of the minimum risk path, thereby completing the path planning of the autonomous vehicle. Summary of the Invention

[0005] The purpose of the present invention is to improve the random uniform safety discrete point sampling algorithm based on an improved driving safety field model, and use the principle of importance sampling to enable the algorithm to plan a more reasonable and reliable motion trajectory for autonomous vehicles under complex traffic situations.

[0006] Among them, the reliability of the path planning of the driverless vehicle refers to the ability of the driverless vehicle to perceive the surrounding environment, model according to on-vehicle equipment, make autonomous decisions through its own set program, plan the driving path, and track the expected path to reach the set destination, so as to complete the predetermined task.

[0007] The present invention provides a path planning method for a driverless vehicle based on an improved driving safety field, and the method flow is shown in the appendix Figure 1 . Among them, the present invention assumes that the driverless vehicle travels on a one-way double-lane road, and the driverless vehicle always travels along the road direction during the collision avoidance process. The present invention mainly includes the following steps:

[0008] Step 1: Detect the road traffic environment and calculate the road field strength distribution according to the driving safety field model.

[0009] (1) Based on the standardized lane on which the driverless vehicle travels, with the extension direction of the lane center line as the x-axis and the perpendicular line from the road starting point to the lane center line as the y-axis, a plane rectangular coordinate system is established, as shown in the appendix Figure 2 . The traffic environment of the road on which the driverless vehicle travels is detected by using the driverless vehicle environment perception system, including the vehicles, pedestrians, traffic signs and obstacles around the driverless vehicle.

[0010] (2) Based on the detected information, a driving safety field E S model of the driverless vehicle is established, and its mathematical model is:

[0011]

[0012] Among them, E V represents the field strength of the dynamic potential energy field, E D represents the safe behavior field, E V_ij represents the field strength of the dynamic potential energy field formed by object i at (x j , y j ), and E D_ij represents the safe behavior field formed by driverless vehicle j at (x i , y i ).

[0013] (3) Combining the patent 202210025985.8 A method for constructing a driving safety field model for a driverless vehicle, the field strength distribution of the whole road is calculated according to formula (1).

[0014] Step 2: Divide the sampling space, use the driving safety field as the weight of importance sampling, sample the road along the y-axis, and construct a road discrete coordinate point matrix.

[0015] (1) Using the driving safety field of the driverless vehicle on the road as the weight of importance sampling, perform lateral (x-axis) sampling on the structured road along the lane center line to obtain the lateral coordinate values x1, x2,..., xn Among them, the higher the field strength, the greater the sampling weight, indicating that the sampling space N in this area is denser.

[0016] (2) For the lane width L (meters), the driving lane is longitudinally sampled (y-axis) with L = x as the sampling interval to obtain the longitudinal coordinate values y1, y2,..., y m , thereby obtaining uniformly discrete state points within the sampling space.

[0017] (3) According to the strength of the driving safety field of the driverless vehicle, sampling spaces with different densities are divided on the transverse road N = {N1, N2,..., N n} and this space can be expressed as a sample point matrix P of m × n m×n , and its expression is as follows:

[0018]

[0019] (x n , y m ) represents the position information of each sampling point, where n represents the number of sampling spaces and m represents the number of discrete points in each layer of sampling space.

[0020] Step 3: Calculate the risk assessment index value - field force of each discrete coordinate point, and construct a driving risk matrix for the driverless vehicle.

[0021] (1) After completing the discretized sampling of the driving road space, the decision-making problem of the driverless vehicle's driving behavior can be transformed into the problem of evaluating the driving safety risk at the sampling points in the discrete space. That is, based on the current traffic situation on the road, the driving risk of the discrete sampling points of M discrete roads in each layer of sampling space N1, N2,..., N n is evaluated, so as to decide the sampling point with the minimum risk and meeting the driving rules in each layer of sampling space.

[0022] (2) Referring to the principle of electromagnetic field, calculate the field force of the driving safety field of the driverless vehicle, and its mathematical relationship is as follows:

[0023]

[0024] Among them, Ω j represents the equivalent mass of the driverless vehicle j, Q j represents the road condition influence factor of the section where the driverless vehicle j is located, D j represents the artificial intelligence (AI) driving risk factor of the driverless vehicle j, F ij represents the field force received by the driverless vehicle j at (x j , y j ) in the driving safety field formed by the object i, F ijCharacterizes the magnitude and direction of the driving risk caused by object i to the driverless vehicle j. F ij The larger the value of E, the greater the risk. S Represents the field strength of the driving safety field, Represents the field strength of the driving safety field of the driverless vehicle at j.

[0025] (3) Calculate P m×n The values of the risk evaluation indicators of the driverless vehicle at each sampling point in the matrix, and then obtain the driving safety risk matrix F of the driverless vehicle m×n In the matrix, each column represents the driving safety risks that the driverless vehicle has to bear at different longitudinal offsets at the same abscissa position, and each row represents the driving safety risks that the driverless vehicle has to bear at different lateral offsets at the same ordinate position. F m×n The expression of F is as follows:

[0026]

[0027] Step 4: Select the subscript of the sample point with the minimum field force from each layer of the sample space to obtain the sequence of path points with the minimum driving risk, and generate the path curve of the driverless vehicle accordingly.

[0028] Select the subscript of the sampling point with the minimum value of the driving safety evaluation index from each layer of the sample space, and correspond it to the value of P in the position coordinate matrix m×n Then the coordinate point where the driverless vehicle is affected by the minimum risk at the current moment can be obtained, and thus a sequence of path points representing the minimum driving risk can be obtained. Description of the Drawings

[0029] Figure 1 Flowchart of a path planning method for a driverless vehicle based on an improved driving safety field

[0030] Figure 2 Schematic diagram of the road rectangular coordinate system

[0031] Figure 3 Schematic diagram of a multi-obstacle scenario

[0032] Figure 4 Schematic diagram of the field strength of the driving safety field of a multi-obstacle driverless vehicle

[0033] Figure 5 Animation diagram of a multi-obstacle driverless vehicle avoiding collision

[0034] Figure 6 Simulation result diagram of the path of uniformly safe discrete points on a multi-obstacle road

[0035] Figure 7 Simulation result diagram of the path of importance sampling safe discrete points on a multi-obstacle road

[0036] Figure 8 TTC comparison diagram of two algorithms

[0037] Figure 9 Comparison Chart of Path Planning Efficiency in Complex Multi-Obstacle Scenarios Specific Implementation Manner

[0038] [Example] Establish a rectangular coordinate system with the extension direction of the lane center line as the x-axis and the perpendicular line from the starting point of the road to the lane center line as the y-axis. In addition, a cylindrical obstacle is set at (125m, 1m) on the road, a triangular obstacle is set at (100m, 2m), and a yellow barrel-shaped obstacle is set at (180m, -2m). The unit of m in the example is meters, as shown in the appendix Figure 3 as follows

[0039] Build a simulation model of the driverless vehicle in the CarSim software. Select the E-Class vehicle in the CarSim database as the prototype, and set the vehicle model parameters of the driverless vehicle as shown in Table 1

[0040] Table 1 Simulation Parameters Related to the Whole Vehicle of the Vehicle

[0041]

[0042] Set the state parameters of the driverless vehicle during the movement process as shown in Table 2

[0043] Table 2 Setting of State Parameters of the Driverless Vehicle during the Movement Process

[0044]

[0045] Step 1: Generate the path trajectory of the driverless vehicle in the multi-obstacle scenario

[0046] Input the scene information and the state of the driverless vehicle into MATLAB, and the field strength model of the driving safety field of the driverless vehicle formed by the current multi-obstacle road scene can be obtained as shown in the appendix Figure 4 as follows

[0047] According to the driving safety field of the road formed by multiple obstacles, the minimum risk sequence points are obtained by using the uniform sampling safety discrete point path planning method and the importance sampling-based safety discrete point sampling method for the driverless vehicle respectively

[0048] Given the road length interval [a, b] for the driverless vehicle to travel, and divide the road interval according to the minimum path sequence points Δ: a = x0 < x1 < … < x n-1 < x n = b, then the cubic spline function of this road trajectory is

[0049]

[0050] where The cubic spline function S3(x) with respect to the partition Δ interval, x0, x1,..., x n are spline nodes, x1, x2,..., x n-1 are interior nodes, x0, x n are boundary points, and α, β are undetermined coefficients.

[0051] The cubic spline function S3(x) of the unmanned vehicle trajectory road needs to satisfy the following conditions:

[0052] (1) On each subinterval [x i , x i-1 (i = 0, 1,..., n - 1), it is a cubic polynomial, that is

[0053]

[0054] (2) S3(x) has second-order continuous derivatives on [a, b], that is

[0055]

[0056] (3) The interpolation nodes satisfy:

[0057]

[0058] (4) Satisfy any one of the following sets of boundary conditions:

[0059]

[0060] According to the constraint relationship of equations (6)-(9), the undetermined coefficients in the unmanned vehicle trajectory curve can be solved to obtain a complete, continuous and smooth spline function. Based on this method, the path trajectory of the unmanned vehicle in this multi-obstacle scenario is generated. In the CarSim simulation software, the animation screenshot of the unmanned vehicle avoiding collisions and bypassing in this multi-obstacle scenario is as attached Figure 5 shown.

[0061] Step 2: Verify the effectiveness and rationality of the path interpolation fitting planning algorithm based on the cubic spline curve.

[0062] Step 1: Safety comparison

[0063] On the premise of ensuring the sampling accuracy, select 29 layers of sampling space to simulate and verify the safety of the uniform safe discrete point sampling path planning algorithm and the safe discrete point sampling path planning algorithm based on importance sampling.

[0064] ① Uniform safe discrete point sampling path curve

[0065] There are three different types of obstacles on the road. The positions of the obstacles are at 100m, 125m, and 180m on the x-axis coordinate of the road. In this area, the range of the x-axis coordinate [60, 200] is selected to simulate and analyze the path trajectory of the driverless vehicle. With a road sampling granularity of 5m, the road section of [60, 200] is evenly divided into 29 sampling spaces. The obtained trajectory is input into CarSim, and the simulation model is run to obtain as shown in the appendix Figure 6 as shown.

[0066] ② Sampling path curve of safety discrete points with importance sampling

[0067] For the range of the x-axis coordinate [60, 200], importance sampling is performed according to the size of the driving safety field of the driverless vehicle generated in the appendix Figure 4 Near the obstacles, the field strength value of the road is large and the sampling space is denser. Similarly, 29 sampling spaces are obtained, and the path trajectory of the driverless vehicle is generated based on the cubic spline interpolation function. After CarSim simulation, the true driving trajectory of the driverless vehicle is obtained as shown in the appendix Figure 7 as shown.

[0068] ③ TTC curve

[0069] The time to collision remaining TTC characterizes the time length before the host vehicle collides with the following vehicle on the premise that the current driving environment and driving task remain unchanged, that is, the time of collision when the relative speed between the ego vehicle and the leading vehicle remains unchanged. The formula is as follows:

[0070]

[0071] where, v s (t) represents the speed of the driverless vehicle itself at time t, v o (t) represents the speed of the road target at time t, v r (t) represents the relative speed between the driverless vehicle and the road target at time t, and D(t) represents the relative distance between the driverless vehicle and the road target at time t.

[0072] For the multi-obstacle road scenario, there are three road targets on the road, namely a cylindrical obstacle, a triangular obstacle, and a yellow barrel obstacle. To compare the safety of the uniform safety discrete point sampling strategy and the safety discrete point sampling strategy based on importance sampling, the remaining collision time TTC(s) of the driverless vehicle from the three road targets during driving is calculated respectively, and the results can be obtained as shown in the appendix Figure 8The curves are as follows. It can be seen from the figure that whether it is a cylindrical obstacle, a triangular obstacle or a yellow barrel obstacle, the remaining collision time between the path planned by the importance sampling safety discrete point strategy and the obstacle is greater than that between the path planned by the uniform safety discrete point sampling strategy and the obstacle. This indicates that the path accuracy obtained by the unmanned vehicle safety discrete point driving path planning algorithm based on importance sampling is higher, and the potential collision risk during the driving of the unmanned vehicle is lower, that is, the safety of the unmanned vehicle during the execution of tasks is higher.

[0073] Step 2: Comparison of path planning efficiency

[0074] To ensure that a safe path is planned in as little sampling space as possible, the number of layers of the sampling space in which the trajectories generated by the algorithm in the same scenario can complete the task and meet the safety requirements can be compared. Among them, the fewer the number of layers, the higher the efficiency, and the more effective the algorithm is. The efficiency of path planning is measured by the number of layers of the sampling space that meets the safety requirements. The formula is as follows:

[0075] min[n j ,n z (11)

[0076] Among them, n j refers to the minimum number of layers for the sampling space N in the uniform sampling safety discrete point algorithm to complete the task and meet the safety and reasonable requirements, and n z refers to the minimum number of layers for the sampling space N in the safety discrete point sampling algorithm based on importance sampling to complete the task and meet the safety and reasonable requirements.

[0077] For a complex multi-obstacle scenario, by using the method of uniform safety discrete point sampling and sampling uniformly at positions 14m apart on the x-axis of the road, 11 layers of sampling space in the road interval [60, 200] can be obtained. Similarly, 11 minimum risk sequence points can be obtained, and then through the cubic spline fitting method, the sampling path as shown in the upper right figure of the appendix can be obtained. It can be seen from the figure that in the road interval ranges of [60, 80] and [180, 200], the trajectory accuracy of the curve is not high, and there is an unreasonable upward convex situation in the actual driving scenario, which does not match the actual driving habits of the unmanned vehicle, indicating that when the sampling space is 11 layers, the method of using uniform safety sampling points for sampling can no longer achieve the required ideal trajectory curve. When the number of sampling spaces is reduced to 10 layers, the trajectory accuracy is even lower, and the generated trajectory is as shown in the upper left figure of the appendix, and there is also an unreasonable upward convex of the curve. Figure 9 The minimum risk sequence points and the corresponding generated unmanned vehicle path trajectories of 10 layers of sampling space and 11 layers of sampling space are obtained respectively by using the safety discrete point driving path planning method based on importance sampling, as shown in the appendix Figure 8 The upper left figure shows that there is also an unreasonable upward convex of the curve.

[0078] The minimum risk sequence points and the corresponding generated unmanned vehicle path trajectories of 10 layers of sampling space and 11 layers of sampling space are obtained respectively by using the safety discrete point driving path planning method based on importance sampling, as shown in the appendix Figure 9As shown in the lower left and lower right figures. As can be seen, the path and curvature obtained by using the unmanned vehicle safety discrete point path planning algorithm based on importance sampling are continuous and smooth, and there is no undesirable upward convex curve. The simulation results are consistent with the feasible path of the actual unmanned vehicle during driving.

[0079] In summary, when using the uniform safety discrete point sampling path planning algorithm, it is no longer possible to generate a reasonable driving trajectory for the unmanned vehicle to perform collision avoidance tasks when the sampling space is 11 layers. However, when using the unmanned vehicle safety discrete point driving path planning method based on importance sampling, a reasonable and smooth-curvature driving trajectory for the unmanned vehicle can still be generated when the sampling space is reduced to 10 layers. For complex multi-obstacle scenarios, the path planning efficiency of the unmanned vehicle safety discrete point driving path planning method based on importance sampling is higher than that of the uniform safety discrete point sampling path planning algorithm, and it can generate reasonable trajectories in a smaller sampling space.

Claims

1. An unmanned vehicle path planning method based on an improved driving safety field, characterized in that It includes the following steps: Step 1: Detect the road traffic environment and calculate the road field strength distribution according to the driving safety field model; (1) Based on the standardized lane for the unmanned vehicle to drive, establish a plane rectangular coordinate system with the extension direction of the lane center line as the x-axis and the perpendicular line from the road starting point to the lane center line as the y-axis; Use the unmanned vehicle environment perception system to detect the traffic environment of the road where the unmanned vehicle is driving, including the vehicles, pedestrians, traffic signs and obstacles around the unmanned vehicle; (2) Establish the driving safety field E of the driverless vehicle based on the detected information S The mathematical model of the model is as follows: Among them, E V represents the field strength of the dynamic potential energy field, E D represents the safe behavior field, E V_ij represents the field strength of the dynamic potential energy field formed by object i at (x j , y j ), and E D_ij represents the safe behavior field formed by the driverless vehicle j at (x i , y i ). (3) Calculate the field strength distribution of the whole road according to the above formula; Step 2: Divide the sampling space, use the driving safety field as the weight for importance sampling, sample the road along the y-axis, and construct a road discrete coordinate point matrix; (1) Taking the driving safety field of the driverless vehicle on the road as the weight for importance sampling, sampling is carried out in the x-axis direction along the center line of the lane on the structured road to obtain the lateral coordinate values x1, x2,..., x n ; among them, the higher the field strength, the greater the sampling weight, indicating that the sampling space N in this area is denser; (2) For the lane width L, the driving lane is sampled in the y-axis direction with L = x as the sampling interval, and the longitudinal coordinate values y1, y2,..., y m are obtained, so as to obtain uniformly discrete state points in the sampling space; (3) According to the strength of the driving safety field of the driverless vehicle, sampling spaces N = {N1, N2,..., N n} with different densities are divided on the horizontal road, and this space can be expressed as a sample point matrix P m×n of m×n, and its expression is as follows: (x n , y m ) represents the position information of each sampling point, where n represents the number of sampling spaces and m represents the number of discrete points in each layer of sampling space; Step 3: Calculate the risk evaluation index value of each discrete coordinate point - field force, and construct an unmanned vehicle driving risk matrix; (1) After the discretized sampling of the driving road space is completed, the decision-making problem of the unmanned vehicle's driving behavior can be transformed into the problem of evaluating the driving safety risk at the sampling points in the discrete space; that is, based on the current traffic situation of the road, the driving risk of the discrete sampling points of M discrete roads in each layer of sampling space N1, N2,..., N n is evaluated to decide the sampling point with the minimum risk and meeting the driving rules in each layer of sampling space; (2) Refer to the electromagnetic field principle to calculate the field force of the unmanned vehicle driving safety field, and its mathematical relationship is as follows: Among them, Ω j represents the equivalent mass of the driverless vehicle j, Q j represents the road condition influence factor of the section where the driverless vehicle j is located, D j represents the artificial intelligence (AI) driving risk factor of the driverless vehicle j, F ij represents the field force received by the driverless vehicle j at (x j , y j ) in the driving safety field formed by the object i, F ij characterizes the magnitude and direction of the driving risk caused by the object i to the driverless vehicle j; F ij The larger the value of F S , the greater the risk; E represents the field strength of the driving safety field of the driverless vehicle at j; (3) Calculate P m×n values of the risk assessment indicators of the driverless vehicle for each sampling point in the matrix, and then obtain the driving safety risk matrix F of the driverless vehicle m×n , each column in the matrix represents the driving safety risks that the driverless vehicle has to bear at different longitudinal offsets under the same abscissa position, and each row represents the driving safety risks that the driverless vehicle has to bear at different lateral offsets under the same ordinate position; The expression of F m×n is as follows: Step 4: Select the subscript of the sample point with the minimum field force from each layer of sample space to obtain the path point sequence with the minimum driving risk, and generate the path curve of the unmanned vehicle accordingly; Select the subscript of the sampling point with the smallest value of the driving safety evaluation index from each layer of the sample space, and correspond it to the value of P in the position coordinate matrix, then the coordinate point where the minimum risk affects the driverless vehicle at the current moment can be obtained, and thus a sequence of path points representing the minimum driving risk can be obtained. m×n The value corresponds, then the coordinate point where the minimum risk affects the driverless vehicle at the current moment can be obtained, and thus a sequence of path points representing the minimum driving risk can be obtained.

Images