Automatic driving vehicle path planning method and system for vulnerable road users

Abstract

The invention relates to a vulnerable road user-oriented automatic driving vehicle path planning method and system. The method comprises the steps of obtaining a trajectory prediction result of a vulnerable road user in a traffic scene; based on the trajectory prediction result, adopting a non-uniform sampling strategy to sample vehicle motion parameters, and generating an alternative trajectory set; the safety cost, the efficiency cost and the comfort cost are calculated for each alternative track, and the safety cost is constructed according to the overlapping degree of the track and the predicted occupied area of the vulnerable road user; distributing a corresponding weight for each cost based on sensitivity analysis; on the basis of the weights and the corresponding costs, constructing a weighted cost function, and sorting the alternative tracks; and selecting the trajectory with the minimum weighted cost function as a final path planning result and outputting the final path planning result. Compared with the prior art, the data-driven path optimization method can improve the safety and decision adaptivity of the automatic driving system in the mixed driving scene of the motor vehicle and the vulnerable road user.

Description

Technical Field

[0001] This invention relates to the field of autonomous vehicle path planning, and in particular to an autonomous vehicle path planning method and system for vulnerable road users. Background Technology

[0002] As autonomous driving technology gradually expands into complex scenarios such as urban roads and intersections, the system's path planning capability in dynamic environments has become a key element in ensuring driving safety. Especially in mixed traffic flow scenarios that include vulnerable road users such as non-motorized vehicles and pedestrians, traditional path planning algorithms face enormous challenges due to the high uncertainty of their behavior and the instability of their movement patterns.

[0003] Existing autonomous driving path planning methods are mostly based on deterministic rules or global map guidance, lacking the ability to dynamically respond to the short-term local behaviors of vulnerable users. Some studies have attempted to integrate trajectory prediction and risk assessment methods, but some problems still exist. The behavior prediction models of vulnerable road users have large errors and poor generalization ability, and the path planner cannot effectively distinguish reliable trajectories. Current planning methods mainly focus on safety or efficiency indicators, neglecting the importance of comfort goals in passenger experience and stability control. Chinese patent CN115447607A discloses a method for planning vehicle driving trajectory. By determining the type of vulnerable road users and predicting their motion trajectory, it generates potential activity areas and avoidance transition areas, selects the planning trajectory with the minimum cost, optimizes the vehicle driving trajectory to reduce longitudinal speed and increase lateral distance, and introduces additional avoidance costs to improve safety and match human driving habits. Although this method considers vulnerable users, it relies on ST or SL diagrams to represent potential activity areas, which is a static or quasi-static modeling, making it difficult to capture the dynamic uncertainties of vulnerable road users. Furthermore, this method only introduces additional avoidance costs on top of the basic costs to match the driving habits of human users, but does not consider driving comfort and efficiency separately, making it difficult to balance multiple dimensions in the final trajectory planning result. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for autonomous vehicle path planning for vulnerable road users.

[0005] The objective of this invention can be achieved through the following technical solutions: A path planning method for autonomous vehicles targeting vulnerable road users includes the following steps: S1, Obtain the trajectory prediction results of vulnerable road users in the traffic scenario, and the trajectory prediction results are represented in the form of a dynamic occupancy grid map; S2, based on the trajectory prediction results, a non-uniform sampling strategy is used to sample the vehicle motion parameters to generate a set of candidate trajectories; S3, calculate the safety cost, efficiency cost, and comfort cost for each candidate trajectory, wherein the safety cost is constructed based on the degree of overlap between the trajectory and the predicted occupied area of ​​vulnerable road users, the efficiency cost is constructed based on the deviation between the average speed of the trajectory and the target expected speed, and the comfort cost is constructed based on the rate of change of the trajectory jerk. S4, assign corresponding weights to each cost based on sensitivity analysis; S5. Based on the weights and corresponding costs, construct a weighted cost function to sort the candidate trajectories; S6: Select the trajectory with the minimum weighted cost function as the final path planning result and output it.

[0006] S2 includes the following steps: S21, Determine the dimension of the sampling parameters, and select the vehicle's longitudinal speed. Lateral displacement and longitudinal acceleration As a sampling parameter; S22, Design a non-uniform sampling distribution: Use a normal distribution centered on the current vehicle speed. As a longitudinal velocity sample, where, Indicates the first The target longitudinal velocity at the sampling time point of each candidate trajectory The mean is variance is The normal distribution This indicates the actual longitudinal speed of the autonomous vehicle at the current moment. This represents the maximum permissible standard deviation of the longitudinal speed sampling distribution; a normal distribution with the lane centerline as the mean is used. As a lateral displacement sample, where Indicates the first The lateral offset of each alternative trajectory relative to the lane centerline in the lateral coordinate system. This indicates that the mean is 0 and the variance is 0. The normal distribution Represents lane width, truncation constraint after sampling Avoid exceeding lane boundaries; use uniform distribution As longitudinal acceleration sampling, where, Indicates the first The candidate trajectories have a constant longitudinal acceleration value throughout the entire prediction time domain [0,T]. This indicates that the upper limit is met. and lower limit Uniform distribution; S23, the parameter combination obtained from sampling It is transformed into a continuous trajectory through a kinematic model: Based on uniform acceleration model Calculate the longitudinal displacement, where, This is the initial longitudinal position. To plan the time domain, For autonomous vehicles t The expected longitudinal displacement at time t; Through fifth-order polynomial interpolation Generates lateral displacement, where, The initial lateral position, coefficient Satisfy boundary smoothness constraints. For autonomous vehicles t The expected lateral displacement at time; S24 transforms the horizontal and vertical displacements in the Frenet coordinate system into global Cartesian coordinates, obtaining a complete set of candidate trajectories containing the position, velocity, and acceleration of each time step.

[0007] The security cost is expressed as: ; in, For the sake of security, This represents the occupancy probability of the dynamically occupied grid diagram. For trajectory points Does it fall into the grid? Indicator functions, For time-varying weights, To plan the time domain, For grid height, This represents the grid width.

[0008] The efficiency cost is expressed as: ; in, For the sake of efficiency, For the target desired speed, To incorporate time-varying speed limits that combine road speed limits with current traffic conditions, for Vehicle speed at any given moment For planning the time domain.

[0009] The cost of comfort is expressed as: ; in, For the price of comfort, , These represent longitudinal and lateral accelerations, respectively. , These represent the rates of change of longitudinal and lateral acceleration, respectively. For planning the time domain.

[0010] An autonomous vehicle path planning system for vulnerable road users includes: Trajectory prediction result acquisition module: Acquires trajectory prediction results of vulnerable road users in traffic scenarios, and the trajectory prediction results are represented in the form of a dynamic occupancy grid map; Candidate trajectory set generation module: Based on the trajectory prediction results, a non-uniform sampling strategy is used to sample the vehicle motion parameters to generate a candidate trajectory set; Cost calculation and alternative trajectory ranking module: For each alternative trajectory, calculate safety cost, efficiency cost, and comfort cost. The safety cost is constructed based on the overlap between the trajectory and the predicted occupied area by vulnerable road users. The efficiency cost is constructed based on the deviation between the trajectory's average speed and the target expected speed. The comfort cost is constructed based on the rate of change of the trajectory's jerk. A corresponding weight is assigned to each cost based on sensitivity analysis. Based on the weights and corresponding costs, a weighted cost function is constructed to rank the alternative trajectories. Output module: Select the trajectory with the minimum weighted cost function as the final path planning result and output it.

[0011] The candidate trajectory set generation module performs the following steps: S21, Determine the dimension of the sampling parameters, and select the vehicle's longitudinal speed. Lateral displacement and longitudinal acceleration As a sampling parameter; S22, Design a non-uniform sampling distribution: Use a normal distribution centered on the current vehicle speed. As a longitudinal velocity sample, where, Indicates the first The target longitudinal velocity at the sampling time point of each candidate trajectory The mean is variance is The normal distribution This indicates the actual longitudinal speed of the autonomous vehicle at the current moment. This represents the maximum permissible standard deviation of the longitudinal speed sampling distribution; a normal distribution with the lane centerline as the mean is used. As a lateral displacement sample, where Indicates the first The lateral offset of each alternative trajectory relative to the lane centerline in the lateral coordinate system. This indicates that the mean is 0 and the variance is 0. The normal distribution Represents lane width, truncation constraint after sampling Avoid exceeding lane boundaries; use uniform distribution As longitudinal acceleration sampling, where, Indicates the first The candidate trajectories have a constant longitudinal acceleration value throughout the entire prediction time domain [0,T]. This indicates that the upper limit is met. and lower limit Uniform distribution; S23, the parameter combination obtained from sampling It is transformed into a continuous trajectory through a kinematic model: Based on uniform acceleration model Calculate the longitudinal displacement, where, This is the initial longitudinal position. To plan the time domain, For autonomous vehicles t The expected longitudinal displacement at time t; Through fifth-order polynomial interpolation Generates lateral displacement, where, The initial lateral position, coefficient Satisfy boundary smoothness constraints. For autonomous vehicles t The expected lateral displacement at time; S24 transforms the horizontal and vertical displacements in the Frenet coordinate system into global Cartesian coordinates, obtaining a complete set of candidate trajectories containing the position, velocity, and acceleration of each time step.

[0012] The security cost is expressed as: ; in, For the sake of security, This represents the occupancy probability of the dynamically occupied grid diagram. For trajectory points Does it fall into the grid? Indicator functions, For time-varying weights, To plan the time domain, For grid height, This represents the grid width.

[0013] The efficiency cost is expressed as: ; in, For the sake of efficiency, For the target desired speed, To incorporate time-varying speed limits that combine road speed limits with current traffic conditions, for Vehicle speed at any given moment For planning the time domain.

[0014] The cost of comfort is expressed as: ; in, For the price of comfort, , These represent longitudinal and lateral accelerations, respectively. , These represent the rates of change of longitudinal and lateral acceleration, respectively. For planning the time domain.

[0015] Compared with the prior art, the present invention has the following beneficial effects: (1) The present invention uses a dynamic occupancy grid diagram to represent the trajectory prediction results, which can more accurately describe the movement status and potential location of vulnerable road users and better reflect real-time changes in space occupancy, thereby improving the accuracy and safety of path planning.

[0016] (2) The present invention uses a non-uniform sampling strategy to generate a set of candidate trajectories, which can generate representative candidate trajectories more efficiently and improve the robustness and diversity of path planning.

[0017] (3) This invention clearly constructs three types of costs: safety, efficiency and comfort, and assigns corresponding weights to each cost based on sensitivity analysis to construct a weighted cost function. This can more scientifically weigh the relationship between different objectives, obtain better path planning results, and improve the decision-making safety and environmental adaptability of autonomous vehicles in complex scenarios. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is the architecture of the autonomous driving module and simulation platform used in the embodiments of the present invention; Figure 3 This is a distribution diagram of the weight coefficients of the three driving evaluation indicators under extreme scenarios in an embodiment of the present invention; Figure 4 This is a heatmap showing the correlation coefficients between the weights of the three types of costs in this invention and the average values ​​of each group of driving evaluation indicators. Detailed Implementation

[0019] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0020] This embodiment provides a path planning method for autonomous vehicles targeting vulnerable road users, such as... Figure 1 As shown, it includes the following steps: S1, Obtain the trajectory prediction results of vulnerable road users in the traffic scenario, and the trajectory prediction results are represented in the form of Dynamic Occupancy Grid Map (DOGM).

[0021] The DOGM discretizes the prediction time domain into multiple time steps and divides the space into a fixed-resolution two-dimensional grid. Each grid cell stores the occupancy probability value at the corresponding time, which characterizes the likelihood of a vulnerable road user appearing at that location. This invention does not limit the specific generation method of the DOGM; it can be output by an external vulnerable road user trajectory prediction module. This module can generate the future spatiotemporal occupancy probability distribution based on road topology features, historical trajectory sequences, and multi-source environmental sensing information (such as environmental data collected by LiDAR). One optional implementation method references the dynamic occupancy grid prediction technology described in the technical document "Dynamic Occupancy Grid Prediction for Urban Autonomous Driving: A Deep Learning Approach with Fully Automatic Labeling" (2018 IEEE International Conference on Robotics and Automation). This involves using a 360° environmental perception system comprised of one BEO LUX long-range laser scanner and four Velodyne VLP-16 laser scanners to collect environmental data on vulnerable road users such as pedestrians and cyclists in mixed urban traffic scenarios. Then, multi-sensor data fusion is performed using Sequential Monte Carlo filtering. The raw sensor data is first converted into a classic occupancy grid map, and then particle filtering is used to estimate the occupancy state and motion characteristics of each grid cell, ultimately generating a DOGM (Dynamic Occupancy Grid Prediction). The occupancy probability value stored in the grid cell directly represents the probability of a vulnerable road user appearing at the corresponding spatiotemporal location. By using the DOGM prediction results, this invention can directly quantify the future spatial occupancy of vulnerable road users during the path planning stage. Compared to schemes relying on single-point trajectory prediction, it can more comprehensively reflect prediction uncertainty and multimodal motion trends, thereby improving the safety and adaptability of planning decisions.

[0022] S2, based on the trajectory prediction results, uses a non-uniform sampling strategy to sample vehicle motion parameters and generate a set of candidate trajectories.

[0023] S2 includes the following steps: S21, Determine the dimension of the sampling parameters, and select the vehicle's longitudinal speed. Lateral displacement and longitudinal acceleration As sampling parameters, the three correspond to the longitudinal velocity characteristics, lateral offset characteristics, and longitudinal acceleration / deceleration characteristics of the trajectory, respectively, and together determine the local motion trajectory shape of the vehicle.

[0024] S22, Design a non-uniform sampling distribution.

[0025] Using a normal distribution centered on the current vehicle speed As a longitudinal speed sample, it ensures that the sampled values ​​are distributed around the current driving state, with a focus on covering the safe speed range. Among these, Indicates the first The target longitudinal velocity at the sampling time point of each candidate trajectory The mean is variance is The normal distribution This indicates the actual longitudinal speed of the autonomous vehicle at the current moment. This represents the maximum permissible standard deviation of the longitudinal velocity sampling distribution.

[0026] Using a normal distribution with the lane centerline as the mean As a lateral displacement sample, where Indicates the first The lateral offset of each alternative trajectory relative to the lane centerline in the lateral coordinate system. This indicates that the mean is 0 and the variance is 0. The normal distribution Represents lane width, truncation constraint after sampling Avoid going beyond the lane boundary.

[0027] Use uniform distribution As longitudinal acceleration sampling, it covers the feasible acceleration range of the vehicle, where... Indicates the first The candidate trajectories have a constant longitudinal acceleration value throughout the entire prediction time domain [0,T]. This indicates that the upper limit is met. and lower limit The uniform distribution.

[0028] S23, the parameter combination obtained from sampling It is transformed into a continuous trajectory through a kinematic model: Based on uniform acceleration model Calculate the longitudinal displacement, where, This is the initial longitudinal position. To plan the time domain, For autonomous vehiclest The expected longitudinal displacement at time t.

[0029] Through fifth-order polynomial interpolation Generates lateral displacement, where, The initial lateral position, coefficient Satisfy boundary smoothness constraints. For autonomous vehicles t The expected lateral displacement at time.

[0030] S24 transforms the horizontal and vertical displacements in the Frenet coordinate system into global Cartesian coordinates, obtaining a complete set of candidate trajectories containing the position, velocity, and acceleration of each time step.

[0031] The non-uniform sampling strategy focuses on sampling key motion parameters and core areas, which reduces invalid computation while ensuring trajectory coverage, and the generated alternative trajectories all meet vehicle dynamics constraints.

[0032] S3, calculate the safety cost, efficiency cost, and comfort cost for each candidate trajectory, wherein the safety cost is constructed based on the degree of overlap between the trajectory and the predicted occupied area of ​​vulnerable road users, the efficiency cost is constructed based on the deviation between the average speed of the trajectory and the target expected speed, and the comfort cost is constructed based on the rate of change of the trajectory jerk.

[0033] In this embodiment, the security cost is expressed as: ; in, For the sake of security, This represents the occupancy probability of the dynamically occupied grid diagram. For trajectory points Does it fall into the grid? Indicator functions, It is a time-varying weight (automatically increases when approaching a conflict zone). To plan the time domain, For grid height, This represents the grid width.

[0034] The efficiency cost is expressed as: ; in, For the sake of efficiency, For the target desired speed, To achieve efficiency evaluation of context-awareness by combining road speed limits with time-varying speed limits based on current traffic conditions, for The vehicle speed at any given moment.

[0035] The cost of comfort is expressed as: ; in, For the price of comfort, , These represent longitudinal and lateral accelerations, respectively. , These represent the rates of change of longitudinal and lateral acceleration, respectively. For planning the time domain.

[0036] By considering both longitudinal and lateral jerk, and optionally suppressing high-frequency components through Fourier transform, passenger-perceived discomfort and vehicle control instability can be reduced. By introducing time-varying weights, context-aware speed limits, and bidirectional jerk assessment, the cost index not only quantifies the safety, efficiency, and comfort of the trajectory but also adapts to the dynamic changes of different traffic scenarios, thereby enhancing the creativity and applicability of planning decisions.

[0037] S4 assigns a corresponding weight to each cost based on sensitivity analysis.

[0038] S4 includes the following steps: S41 extracts trajectory data from existing typical scenario datasets of conflicts between vulnerable road users and motor vehicles, as well as trajectory data from test scenario sets, covering kinematic parameters such as speed, acceleration, and steering angle, and interactive behavior characteristics in conflict events.

[0039] Step S42: Based on the data extracted in step S41, simulation is performed using the MATLAB Autonomous Driving Toolbox to simulate the trajectory planning and control behavior of autonomous vehicles and generate multiple alternative trajectories.

[0040] Step S43: Use Morris sensitivity analysis to quantify the impact of changes in the three types of costs (safety, efficiency, and comfort) on the total cost, rank their importance, and assign fixed weights accordingly. , , .

[0041] S5. Based on the weights and corresponding costs, construct a weighted cost function C to sort the candidate trajectories.

[0042] .

[0043] S6: Select the trajectory with the minimum weighted cost function as the final path planning result and output it.

[0044] The above is an introduction to the method embodiments. The following system embodiments will further illustrate the solution of the present invention.

[0045] An autonomous vehicle path planning system for vulnerable road users includes: Trajectory prediction result acquisition module: Acquires trajectory prediction results of vulnerable road users in traffic scenarios, and the trajectory prediction results are represented in the form of a dynamic occupancy grid map; Candidate trajectory set generation module: Based on the trajectory prediction results, a non-uniform sampling strategy is used to sample the vehicle motion parameters to generate a candidate trajectory set; Cost calculation and alternative trajectory ranking module: For each alternative trajectory, calculate safety cost, efficiency cost, and comfort cost. The safety cost is constructed based on the overlap between the trajectory and the predicted occupied area by vulnerable road users. The efficiency cost is constructed based on the deviation between the trajectory's average speed and the target expected speed. The comfort cost is constructed based on the rate of change of the trajectory's jerk. A corresponding weight is assigned to each cost based on sensitivity analysis. Based on the weights and corresponding costs, a weighted cost function is constructed to rank the alternative trajectories. Output module: Select the trajectory with the minimum weighted cost function as the final path planning result and output it.

[0046] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the described module can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0047] This embodiment utilizes a local natural driving research dataset containing 280 conflict events (speed, acceleration, steering angle, TTC, etc., covering four types of scenarios: longitudinal, lateral, forward, and steering) to test the autonomous driving trajectory planning method.

[0048] Based on step S1, the non-motorized vehicle trajectory prediction results are obtained and the Dynamic Occupied Grid Map (DOGM) is used.

[0049] Based on step S2, in the Frenet coordinate system, with the road centerline as the reference path, a non-uniform sampling strategy (normal distribution center sampling, sampling dimensions including lateral displacement, longitudinal velocity, and longitudinal acceleration) is adopted to generate candidate trajectories. The hyperparameter settings are shown in Table 1.

[0050] Table 1 Table 1 This represents the total number of time steps within the planning time domain. L represents the length of the autonomous vehicle.

[0051] Simulation platform architecture such as Figure 2As shown, this demonstrates the complete process from acquiring environmental information from the perception module, through trajectory planning and local trajectory processing, to finally outputting control commands such as steering wheel, accelerator, and brake through the vehicle control module.

[0052] Based on step S3, calculate the cost index for each trajectory. Figure 3 The distribution of weight coefficients for minTTC, TDD, and jerk under extreme scenarios is shown to reflect the balance of the indicators.

[0053] Following step S41, a dataset of 280 non-collision events is extracted. Following step S42, based on the MATLAB Autonomous Driving Toolbox simulation, perception, trajectory planning, and vehicle control are simulated to generate multiple candidate trajectories. Following step S43, Morris sensitivity analysis is used to analyze the impact of parameters on minTTC, TDD, and jerk, optimizing the weights to safety 0.2, efficiency 0.4, and comfort 0.4. The correlation between weights and indicators is as follows: Figure 4 As shown, the heat map displays... , , With average minTTC, TDD, The correlation coefficient indicates that the efficiency and comfort weights are highly sensitive.

[0054] Based on step S5, construct the weighted total cost function, which is a weighted summation.

[0055] According to step S6, the trajectory with the minimum total cost is selected as the final path planning result and output.

[0056] Overall analysis shows that efficiency and comfort weights (0.4 each) have a significant impact on trajectory planning, safety weight (0.2) ensures conflict avoidance, minTTC indicates sufficient safety distance, TDD reflects efficient path planning, and jerk distribution indicates that comfort optimization is effective.

[0057] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A path planning method for autonomous vehicles targeting vulnerable road users, characterized in that, Includes the following steps: S1, Obtain the trajectory prediction results of vulnerable road users in the traffic scenario, and the trajectory prediction results are represented in the form of a dynamic occupancy grid map; S2, based on the trajectory prediction results, a non-uniform sampling strategy is used to sample the vehicle motion parameters to generate a set of candidate trajectories; S3, calculate the safety cost, efficiency cost, and comfort cost for each candidate trajectory, wherein the safety cost is constructed based on the degree of overlap between the trajectory and the predicted occupied area of ​​vulnerable road users, the efficiency cost is constructed based on the deviation between the average speed of the trajectory and the target expected speed, and the comfort cost is constructed based on the rate of change of the trajectory jerk. S4, assign corresponding weights to each cost based on sensitivity analysis; S5. Based on the weights and corresponding costs, construct a weighted cost function to sort the candidate trajectories; S6: Select the trajectory with the minimum weighted cost function as the final path planning result and output it.

2. The autonomous vehicle path planning method for vulnerable road users according to claim 1, characterized in that, S2 includes the following steps: S21, Determine the dimension of the sampling parameters, and select the vehicle's longitudinal speed. Lateral displacement and longitudinal acceleration As a sampling parameter; S22, Design a non-uniform sampling distribution: Use a normal distribution centered on the current vehicle speed. As a longitudinal velocity sample, where, Indicates the first The target longitudinal velocity at the sampling time point of each candidate trajectory The mean is variance is The normal distribution This indicates the actual longitudinal speed of the autonomous vehicle at the current moment. This represents the maximum permissible standard deviation of the longitudinal speed sampling distribution; a normal distribution with the lane centerline as the mean is used. As a lateral displacement sample, where Indicates the first The lateral offset of each alternative trajectory relative to the lane centerline in the lateral coordinate system. This indicates that the mean is 0 and the variance is 0. The normal distribution Represents lane width, truncation constraint after sampling Avoid exceeding lane boundaries; use uniform distribution As longitudinal acceleration sampling, where, Indicates the first The candidate trajectories have a constant longitudinal acceleration value throughout the entire prediction time domain [0,T]. This indicates that the upper limit is met. and lower limit Uniform distribution; S23, the parameter combination obtained from sampling Transformed into a continuous trajectory using a kinematic model: Based on uniform acceleration model Calculate the longitudinal displacement, where, This is the initial longitudinal position. To plan the time domain, For autonomous vehicles t The expected longitudinal displacement at time t; Through fifth-order polynomial interpolation Generates lateral displacement, where, The initial lateral position, coefficient Satisfy boundary smoothness constraints. For autonomous vehicles t The expected lateral displacement at time; S24 transforms the horizontal and vertical displacements in the Frenet coordinate system into global Cartesian coordinates, obtaining a complete set of candidate trajectories containing the position, velocity, and acceleration of each time step.

3. The autonomous vehicle path planning method for vulnerable road users according to claim 1, characterized in that, The security cost is expressed as: ； in, For the sake of security, This represents the occupancy probability of the dynamically occupied grid. For trajectory points Does it fall into the grid? Indicator functions, For time-varying weights, To plan the time domain, For grid height, This represents the grid width.

4. The autonomous vehicle path planning method for vulnerable road users according to claim 1, characterized in that, The efficiency cost is expressed as: ； in, For the sake of efficiency, For the target desired speed, To incorporate time-varying speed limits that combine road speed limits with current traffic conditions, for Vehicle speed at any given moment For planning the time domain.

5. The autonomous vehicle path planning method for vulnerable road users according to claim 1, characterized in that, The cost of comfort is expressed as: ； in, For the price of comfort, , These represent longitudinal and lateral accelerations, respectively. , These represent the rates of change of longitudinal and lateral acceleration, respectively. For planning the time domain.

6. A path planning system for autonomous vehicles targeting vulnerable road users, characterized in that, include: Trajectory prediction result acquisition module: Acquires trajectory prediction results of vulnerable road users in traffic scenarios, and the trajectory prediction results are represented in the form of a dynamic occupancy grid map; Candidate trajectory set generation module: Based on the trajectory prediction results, a non-uniform sampling strategy is used to sample the vehicle motion parameters to generate a candidate trajectory set; Cost calculation and alternative trajectory ranking module: For each alternative trajectory, calculate safety cost, efficiency cost, and comfort cost. The safety cost is constructed based on the overlap between the trajectory and the predicted occupied area by vulnerable road users. The efficiency cost is constructed based on the deviation between the trajectory's average speed and the target expected speed. The comfort cost is constructed based on the rate of change of the trajectory's jerk. A corresponding weight is assigned to each cost based on sensitivity analysis. Based on the weights and corresponding costs, a weighted cost function is constructed to rank the alternative trajectories. Output module: Select the trajectory with the minimum weighted cost function as the final path planning result and output it.

7. The autonomous driving vehicle path planning system for vulnerable road users according to claim 6, characterized in that, The candidate trajectory set generation module performs the following steps: S21, Determine the dimension of the sampling parameters, and select the vehicle's longitudinal speed. Lateral displacement and longitudinal acceleration As a sampling parameter; S22, Design a non-uniform sampling distribution: Use a normal distribution centered on the current vehicle speed. As a longitudinal velocity sample, where, Indicates the first The target longitudinal velocity at the sampling time point of each candidate trajectory The mean is variance is The normal distribution This indicates the actual longitudinal speed of the autonomous vehicle at the current moment. This represents the maximum permissible standard deviation of the longitudinal speed sampling distribution; a normal distribution with the lane centerline as the mean is used. As a lateral displacement sample, where Indicates the first The lateral offset of each alternative trajectory relative to the lane centerline in the lateral coordinate system. This indicates that the mean is 0 and the variance is 0. The normal distribution Represents lane width, truncation constraint after sampling Avoid exceeding lane boundaries; use uniform distribution As longitudinal acceleration sampling, where, Indicates the first The candidate trajectories have a constant longitudinal acceleration value throughout the entire prediction time domain [0,T]. This indicates that the upper limit is met. and lower limit Uniform distribution; S23, the parameter combination obtained from sampling Transformed into a continuous trajectory using a kinematic model: Based on uniform acceleration model Calculate the longitudinal displacement, where, This is the initial longitudinal position. To plan the time domain, For autonomous vehicles t The expected longitudinal displacement at time t; Through fifth-order polynomial interpolation Generates lateral displacement, where, The initial lateral position, coefficient Satisfy boundary smoothness constraints. For autonomous vehicles t The expected lateral displacement at time; S24 transforms the horizontal and vertical displacements in the Frenet coordinate system into global Cartesian coordinates, obtaining a complete set of candidate trajectories containing the position, velocity, and acceleration of each time step.

8. The autonomous vehicle path planning system for vulnerable road users according to claim 6, characterized in that, The security cost is expressed as: ； in, For the sake of security, This represents the occupancy probability of the dynamically occupied grid. For trajectory points Does it fall into the grid? Indicator functions, For time-varying weights, To plan the time domain, For grid height, This represents the grid width.

9. A path planning system for autonomous vehicles targeting vulnerable road users according to claim 6, characterized in that, The efficiency cost is expressed as: ； in, For the sake of efficiency, For the target desired speed, To incorporate time-varying speed limits that combine road speed limits with current traffic conditions, for Vehicle speed at any given moment For planning the time domain.

10. A path planning system for autonomous vehicles oriented towards vulnerable road users according to claim 6, characterized in that, The cost of comfort is expressed as: ； in, For the price of comfort, , These represent longitudinal and lateral accelerations, respectively. , These represent the rates of change of longitudinal and lateral acceleration, respectively. For planning the time domain.

Images