WO2022193584A1

WO2022193584A1 - Multi-scenario-oriented autonomous driving planning method and system

Info

Publication number: WO2022193584A1
Application number: PCT/CN2021/118608
Authority: WO
Inventors: 孙宏滨; 陈炜煌
Original assignee: 西安交通大学
Priority date: 2021-03-15
Filing date: 2021-09-15
Publication date: 2022-09-22
Also published as: CN112964271B; CN112964271A

Abstract

A multi-scenario-oriented autonomous driving planning method, comprising: acquiring a global reference route (S1); generating a local reference route according to whether there is an optimal trajectory, and the position of a vehicle and the global reference route: if there is no optimal trajectory, performing path point sampling according to the position of the vehicle and the global reference route to obtain a first local reference route, and also searching for and generating a reference route of a corresponding traffic road segment by using a hybrid A* algorithm, and splicing and integrating same into the first local reference route; performing smoothing processing on the reference route under the condition of taking a vehicle kinematic constraint and an obstacle constraint into consideration; and if there is an optimal trajectory, performing path point sampling according to the position of the vehicle and the global reference route to obtain the first local reference route, and performing smoothing processing (S2); acquiring an optimal trajectory according to the position of the vehicle, obstacle information and the final local reference route (S3); and cyclically performing S2 and S3 (S4). A multi-scenario-oriented autonomous driving planning system, comprising: a global planning module, a reference route provision module and a local planning module. By means of the method and the system, an autonomous driving vehicle can more robustly deal with diversified traffic scenarios.

Description

A multi-scenario-oriented autonomous driving planning method and system

technical field

The invention belongs to the technical field of automatic driving vehicles, and in particular relates to a multi-scene-oriented automatic driving planning method and system.

Background technique

Path planning or motion trajectory planning is one of the basic technologies for autonomous vehicles. Considering the motion model of the vehicle and the surrounding obstacles, path planning will generate a series of states to make the vehicle transition from the initial state to the desired state. As the target location and traffic environment change, autonomous vehicles need to adopt different behaviors to accomplish time-varying driving tasks. When driving in urban areas, road obstacles are easy to detect and track, traffic scenarios are relatively regulated and simple, and it is sufficient to navigate autonomous vehicles by following or changing lanes and overtaking obstacles ahead. When the road is remodeled by the traffic infrastructure, or the vehicle is required to reach the designated position, it is difficult to complete the task simply by following or changing the lane and overtaking the obstacles in front of the vehicle. More complex and precise planning algorithms are needed to achieve this task.

In recent years, many path planning methods have been proposed in the robotics literature. However, these planning methods only show good results in specific traffic scenarios or some traffic scenarios, and cannot effectively deal with most traffic scenarios at the same time. For structured urban traffic scenarios, the deviation between the global reference route planned based on the high-precision map and the drivable route of the real traffic scene is small. It is better to extract the local reference route from the global reference route and use the sampling-based planning algorithm. to meet the needs of autonomous driving trajectory planning. However, in traffic scenarios such as parking lots, narrow road U-turns, and construction occupied roads, there is a large deviation between the global reference route planned based on the high-precision map and the real drivable route. The global reference route of precision map planning is used for trajectory planning, and the trajectory planned by the sampling-based planning algorithm will become infeasible.

SUMMARY OF THE INVENTION

The purpose of the present invention is to overcome the above-mentioned shortcomings of the prior art, and propose a multi-scenario-oriented automatic driving planning method and system, which utilizes different reference route generation methods to deal with different traffic scenarios, and generates a road section that is highly drivable at the height of the current road section. For the local reference route of the route, feasible motion trajectories can be planned under the whole road section through the sampling-based planning algorithm, so that the autonomous vehicle can more robustly deal with diverse traffic scenarios.

Purpose of the present invention adopts following technical scheme to realize:

A multi-scenario-oriented autonomous driving planning method, comprising the following steps:

S1, analyze the map topology information, and use the analysis result to obtain a global reference route;

S2, generating a local reference route according to whether there is an optimal trajectory, the position of the self-vehicle and the global reference route, and the process includes;

If there is no optimal trajectory, the first local reference route is obtained by sampling the waypoints according to the position of the vehicle and the global reference route. At the same time, the hybrid A* algorithm is used to search and generate the reference route corresponding to the traffic section; the reference route is spliced and integrated into the first local reference route. In the reference route, a complete local reference route is obtained; the reference route is smoothed under the conditions of vehicle kinematic constraints and obstacle constraints to obtain the final local reference route;

If there is an optimal trajectory, the path point sampling is performed according to the position of the vehicle and the global reference route to obtain the first local reference route, and the first local reference route is smoothed to obtain the final local reference route;

S3, obtain the optimal trajectory according to the position of the self-vehicle, obstacle information and the final local reference route;

S4, S2-S3 are performed cyclically.

Preferably, the process of S1 includes:

Analyze the map topology information to obtain a road network including road physical information and semantic information;

According to the road network information, self-vehicle positioning information, and task starting point and target point, a global route search is performed to generate a set of global discrete route points with a contextual relationship, and the global discrete route points are used to generate a global reference route.

Preferably, in S2, when the hybrid A* algorithm is used to search and generate the reference route corresponding to the traffic section, the grid resolution is higher near the vehicle position and lower at the distance from the vehicle position.

Preferably, in S2, when the hybrid A* algorithm is used to search and generate the reference route corresponding to the traffic section, the point of the relative global path of the ego vehicle is determined, and the relative global path of the ego vehicle from the far and near directions within the preset distance from the point is determined. The point pushback of , combined with the grid map, to find the local target point, the local target point satisfies: within the range of the grid map, and there are no obstacles within the grid where this point is located and the preset distance range around it;

If the local target point is not found when pushing back to the point relative to the global path of the self-vehicle, a point in front of the self-vehicle with no surrounding obstacles is taken as the local target point.

Preferably, in S2, the process of splicing and integrating the reference route into the first local reference route includes:

The starting point and the ending point in the local reference route are selected as the splicing points, and the corresponding closest point in the first local reference route is searched according to the starting point and the ending point, and the splicing is performed.

Preferably, in S3, the starting planning point on the final local reference route is calculated according to the self-vehicle positioning information, and then the Cartesian coordinates in the local reference route are converted into Frenet coordinates;

Calculate the starting point of the autonomous vehicle in the Frenet coordinate system and the state information of the starting point, and the state information of the starting point is represented as the state at time T0;

Before sampling, the obstacle information is processed first, and the corresponding SL map (ie longitudinal displacement-horizontal displacement map) and ST map (ie longitudinal displacement-time map) are established based on the reference route when processing obstacles;

After the obstacle information processing is completed, start to sample the final state at the next T1 time, which is expressed as the state at the T1 time;

After sampling, polynomial fitting is performed between the state at time T1 and the state at time T0 to generate two polynomial trajectories in the horizontal and vertical directions, and two-dimensional synthesis of the two polynomial trajectories in the horizontal and vertical directions is performed to obtain a trajectory set;

Find the optimal trajectory from the trajectory set.

Preferably, the process of finding the optimal trajectory from the trajectory set includes:

Calculate the cost C _total for each trajectory in the trajectory set:

Perform physical limit detection and collision detection on the trajectory with the lowest cost. If the two conditions of vehicle motion constraint and no collision are met at the same time, the trajectory in the Frenet coordinate system is converted into the trajectory in the Cartesian coordinate system and returned to the trajectory; If at least one of the vehicle motion constraints and the no-collision condition is not met, the trajectory is eliminated, and the trajectory with the lowest cost is selected to continue detection until a trajectory that satisfies both of the above conditions is found;

If no matching trajectory is found, then proceed to S4.

Preferably: the cost C _total of each trajectory in the trajectory set is calculated by the following formula:

C _total = ω _{lon_obj} *C _{lon_obj} +ω _{lon_jerk} *C _{lon_jerk} +ω _collision *C _collision +ω _{lat_offset} *C _{lat_offset}

Among them, ω _{lon_obj} , ω _{lon_jerk} , ω _collision and ω _{lat_offset} are the weight of missed target, longitudinal jerk weight, collision weight and lateral offset weight, respectively, C _{lon_obj} , C _{lon_jerk} , C _collision and C _{lat_offset} are missed target, respectively Target cost, longitudinal jerk cost, collision cost, lateral offset cost.

Preferably, the position of the self-vehicle is obtained by reading GPS positioning and laser SLAM positioning on the vehicle.

The present invention provides a multi-scenario-oriented automatic driving planning system for implementing the multi-scenario-oriented automatic driving planning method of the present invention. The system includes:

Global planning module: used to parse the map topology information, obtain the global reference route, and output the global reference route;

Reference route providing module: used to generate a local reference route according to whether there is an optimal trajectory, the position of the vehicle and the global reference route. The process includes;

And output the local reference route;

Local planning module: It is used to obtain the optimal trajectory according to the position of the self-vehicle, obstacle information and the final local reference route, and output the result of whether there is an optimal trajectory.

Compared with the prior art, the beneficial effects of the present invention are:

The multi-scene-oriented automatic driving planning method of the present invention adopts the hybrid A* algorithm for complex scenes such as parking lots, narrow road U-turns, and construction road occupations to generate drivable routes that are highly in line with real traffic, and splicing and integrating them into high-precision In the local reference route planned by the map, the generated reference route is highly similar to the actual drivable route in the traffic scene, and then the motion trajectory planning is carried out through the sampling-based method, which enhances the robustness of the planning algorithm and enables the autonomous vehicle to effectively Cope with complex and diverse traffic scenarios.

Description of drawings

Fig. 1 is a schematic diagram of the overall planning flow of the present invention;

2 is a schematic diagram of a semi-structured scene of the present invention;

Fig. 3 is the schematic diagram of behavior Frenet coordinate system in the method of the present invention;

4 is a schematic diagram of a lateral trajectory in the method of the present invention;

FIG. 5 is a schematic diagram of a longitudinal trajectory in the method of the present invention.

Detailed ways

Below in conjunction with accompanying drawing and embodiment, the present invention is described in further detail:

Referring to FIG. 1, FIG. 3-FIG. 5, the multi-scenario-oriented automatic driving planning method of the present invention includes the following steps:

S1: The global planning module outputs a global reference route containing high-precision map semantic information and topology information;

S1 includes the following sub-steps:

S11: Analyze the map topology information to obtain a road network including road physical information and semantic information;

S12: Perform a global route search according to the starting point and the end position to obtain an initial global reference route.

S2: The reference route providing module outputs the local reference route to the local planning module according to the global reference route and the position of the vehicle;

S2 includes the following sub-steps:

S21: Read GPS positioning and laser SLAM positioning to determine the position of the vehicle;

S22: Generate a local reference route and output it to the local planning module according to the state of the vehicle, the state of the local planning module in the previous frame, and the global reference route.

Step S22 includes: if the state of the local planning module in the previous frame is returned successfully (the first frame at the beginning of the planning is successful by default), sampling the path points according to the position of the vehicle and the global reference route to obtain a local reference route PATH_1, and smoothing the PATH_1 After processing, it is sent to the local planning module. If the module status of the local planning module in the previous frame fails to return, firstly obtain a local reference route PATH_1 by sampling the waypoints according to the position of the vehicle and the global reference route, and use the hybrid A* algorithm to search and generate the reference route PATH_2 corresponding to the traffic section; then The PATH_2 is spliced and integrated into PATH_1 to obtain a complete reference route; finally, the reference route is smoothed under the conditions of vehicle kinematic constraints and obstacle constraints, and the smoothed reference route is sent to the local planning module.

The splicing and integration process of PATH_1 and PATH_2 in the sub-step S22 specifically includes the following sub-steps:

S221: Select the start point and the end point in PATH_2 as the splicing point to ensure that the entire section of PATH_1 is retained;

S222: Find the closest point in PATH_1 according to the starting point and the ending point in S241;

S223: Select the front and rear splicing points on PATH_1 according to the two closest points and related parameters;

S224: Using cubic spline interpolation, splicing the front end of PATH_1, the entire segment of PATH_2, and the end of PATH_1 to form a complete reference route.

S3: The local planning module uses the planning method based on the Frenet coordinate system to generate a feasible and safe motion trajectory;

S3 specifically includes the following sub-steps:

S31: Calculate the starting planning point on the local reference route according to the position of the vehicle, and initialize the Frenet information;

S32: Establish corresponding SL map and ST map based on the local reference route according to the obstacle information;

S33: Generate a trajectory set based on the sampling method;

S34: Calculate the cost of each trajectory;

S35: Loop detection, and return to the motion track that meets the conditions; if there is no motion track that meets the conditions, skip to step S2, and re-plan the reference route.

Example

As shown in FIG. 1 , the multi-scenario-oriented automatic driving planning method in this embodiment includes the following steps:

S1: The global planning module outputs a global reference route containing high-precision map semantic information and topology information:

As shown in Figure 1, the global planning module first parses the map topology information, mainly including lane-related information and lane topology relationship, and parses to obtain a road network containing road physical information and semantic information. Perform global route search based on road network information, positioning information, and mission starting and target points. The search algorithm adopts the Dijkstra algorithm to preliminarily determine the road sections that need to be passed, and then determine the route within the road section according to the connection relationship, generate a set of global discrete route points with context, and generate a global reference route according to the set of global discrete route points with context. .

S2: The reference route providing module outputs the reference route to the local planning module according to the position of the vehicle and the global reference route:

The process of the reference route providing module mainly includes: reading GPS positioning and laser SLAM positioning, determining the position of the vehicle; generating a local reference route according to the vehicle state, the state of the local planning module in the previous frame and the global reference route and outputting it to the local planning module: if If the status of the local planning module in the previous frame is returned successfully (the first frame at the beginning of the planning is successful by default), then a local reference route PATH_1 is obtained by sampling the path points according to the position of the vehicle and the global reference route, and the PATH_1 is sent to the local after smoothing processing. planning module. If the state of the local planning module in the previous frame fails to return (generally occurs in semi-structured scenarios such as parking lots, construction occupied roads, etc., as shown in Figure 2), firstly, sampling the path points according to the position of the vehicle and the global reference route to obtain a local Refer to the route PATH_1, and use the hybrid A* algorithm to search and generate the reference route PATH_2 corresponding to the traffic section; then splicing and integrating PATH_2 into PATH_1 to obtain a complete reference route; finally, considering the vehicle kinematic constraints and obstacle constraints The reference route is smoothed, and the smoothed reference route is sent to the local planning module.

The input of the hybrid A* algorithm is the obstacle grid map detected by lidar, as well as the initial state of the vehicle s ₀ =<x,y,θ> ₀ and the target state s _g =<x,y,θ> _g , where <x, y, θ> are the position of the vehicle and the heading of the vehicle, respectively. The desired output _is a trajectory (sequence of vehicle states _s ₀ , _s ₁ , _. _i-1 ||≤δ _s ). The present invention improves and supplements the hybrid A* method. First, according to the characteristics of the perception range, the present invention changes the resolution of the grid, and changes the resolution of the grid to a The location is high and the distance is low, which can speed up the search speed while still retaining a high path accuracy around the vehicle body. The second is to analyze and describe how to determine the local target of a search when the global path is determined. The global reference path indicates However, due to the change of road conditions and the appearance of new obstacles at any time in reality, it is necessary to combine the global reference path and the obstacles around the vehicle to determine the local end point of a hybrid A* algorithm planning. The method is to first determine the point of the own vehicle relative to the global path, start pushing back from the distance 100 meters away from the point (that is, from the 200th point), and combine the grid map to try to find the first such point: that is, in the grid Within the range of the grid map, and there are no obstacles in the grid where this point is located and within 2 to 5 meters around it. Take this point as the local target point. If the local target point is not found after pushing back to the point where the vehicle is located, then Try to use a point directly in front of the vehicle with no surrounding obstacles as the local target point.

The process of splicing two reference paths into a reference route includes selecting the starting point and the ending point in the path PATH_2 planned by the hybrid A* as the splicing point, and searching the reference route PATH_1 extracted from the global reference path according to the starting point and the ending point. the closest point. For the selection of splicing points in PATH_1, a splicing interval of 1.5 to 2.5 meters is reserved to ensure that the path after interpolation and smoothing is better close to the real environment. The interpolation method used in this application is cubic spline interpolation, and a higher-order spline interpolation method can also be used if necessary.

S3: The local planning module uses the planning method based on the Frenet coordinate system to generate a feasible and safe motion trajectory

After the reference route is determined, the first thing to do is to calculate the starting planning point on the reference route (that is, the starting point of each cycle trajectory planning) according to the GPS positioning of the vehicle, and then convert the Cartesian coordinates (X-Y) in the reference route. is the Frenet coordinate (S-L, longitudinal displacement - lateral displacement). The Frenet coordinate system is shown in Figure 3, where S represents the longitudinal displacement of the vehicle in the Frenet coordinate system, that is, the distance along the reference route; L represents the lateral displacement of the vehicle in the Frenet coordinate system, that is, the distance from the reference route to the left and right positions . After the coordinates are converted, the starting point of the autonomous vehicle in the Frenet coordinate system and the state information of the starting point are calculated, which are expressed as the state at time T0. Before sampling, the obstacles need to be processed first. The method we use is to establish the corresponding SL map and ST map based on the reference route. After the obstacle information processing is completed, start to sample the final state at the next time T1, which is represented as the state at time T1. After sampling, the state at time T1 and the state at time T0 are polynomially fitted to generate two polynomial trajectories in the horizontal and vertical directions, as shown in Figure 4 and Figure 5. Finally, the two polynomial trajectories in the horizontal and vertical directions are synthesized in two dimensions. According to the above, the process of generating the trajectory includes: given a series of times, calculating the longitudinal and lateral offsets of the vehicle at the series of times, and generating a series of two-dimensional plane trajectory points according to the reference route. Through a series of time points, a series of trajectory points are obtained by calculation, and then a trajectory set is generated. The trajectory set is obtained, and the cost of each trajectory is calculated. This cost mainly considers the feasibility, comfort and safety of the trajectory:

C _total = ω _{lon_obj} *C _{lon_obj} +ω _{lon_jerk} *C _{lon_jerk} +ω _collision *C _collision +ω _{lat_offset} *C _{lat_offset} , where ω _{lon_obj} , ω _{lon_jerk} , ω _collision and ω _{lat_offset} are the weight of the missed target and the longitudinal jerk, respectively C _{lon_obj} , C _{lon_jerk} , C _collision and C _{lat_offset} are the missed target cost, longitudinal jerk cost, collision cost, and lateral offset cost, respectively.

The last step is to find the optimal trajectory, which is a process of cyclic detection of the trajectory. In this process, physical limit detection and collision detection are first performed on the least expensive trajectories. If both conditions are satisfied at the same time, the trajectory in the Frenet coordinate system is converted into a trajectory in the Cartesian coordinate system and returned; if either condition is not satisfied, the trajectory is eliminated, and then the lowest cost trajectory is selected Continue to detect until a trajectory that satisfies both conditions is found. If a matching motion trajectory cannot be found, set the local planning module status to fail, and then jump to the reference route providing module to regenerate the operation reference route.

To sum up, the present invention is a multi-scenario-oriented automatic driving planning method for navigating an automatic driving vehicle. The semi-structured scenes such as parking lots, narrow road U-turns, and construction road occupations are effectively integrated with the reference routes of urban roads, viaducts and other structured scenes, ensuring that the reference route is similar to the actual drivable route. An algorithmic framework for the planning of motion trajectories greatly enhances the robustness of the planning method, enabling autonomous vehicles to effectively deal with complex and diverse traffic scenarios.

Claims

A multi-scene-oriented automatic driving planning method, characterized in that it includes the following steps:

S1, analyze the map topology information, and use the analysis result to obtain a global reference route;

S2, generating a local reference route according to whether there is an optimal trajectory, the position of the self-vehicle and the global reference route, and the process includes;

If there is no optimal trajectory, the first local reference route is obtained by sampling the waypoints according to the position of the vehicle and the global reference route. At the same time, the hybrid A* algorithm is used to search and generate the reference route corresponding to the traffic section; the reference route is spliced and integrated into the first local reference route. In the reference route, a complete local reference route is obtained; the reference route is smoothed under the conditions of vehicle kinematic constraints and obstacle constraints to obtain the final local reference route;

If there is an optimal trajectory, the path point sampling is performed according to the position of the vehicle and the global reference route to obtain the first local reference route, and the first local reference route is smoothed to obtain the final local reference route;

S3, obtain the optimal trajectory according to the position of the self-vehicle, obstacle information and the final local reference route;

S4, S2-S3 are performed cyclically.
A multi-scenario-oriented automatic driving planning method according to claim 1, wherein the process of S1 comprises:

Analyze the map topology information to obtain a road network including road physical information and semantic information;

According to the road network information, self-vehicle positioning information, and task starting point and target point, a global route search is performed to generate a set of global discrete route points with a contextual relationship, and the global discrete route points are used to generate a global reference route.
A multi-scene-oriented automatic driving planning method according to claim 1, wherein in S2, when using the hybrid A* algorithm to search and generate a reference route corresponding to the traffic section, the grid resolution is the distance from the vehicle position A form that is higher in the vicinity and lower in the distance from the vehicle's position.
The multi-scenario-oriented automatic driving planning method according to claim 1, wherein in S2, when using the hybrid A* algorithm to search and generate a reference route corresponding to the traffic section, the point of the relative global path of the self-vehicle is determined, and in S2 Within the range of the preset distance from this point, push back from the far and near points of the ego vehicle relative to the global path, and combine the grid map to find a local target point, the local target point satisfies: within the range of the grid map, and this There are no obstacles within the grid where the point is located and the preset distance around it;

If the local target point is not found when pushing back to the point relative to the global path of the self-vehicle, a point in front of the self-vehicle with no surrounding obstacles is taken as the local target point.
The multi-scene-oriented automatic driving planning method according to claim 1, wherein in S2, the process of splicing and integrating the reference route into the first local reference route comprises:

The starting point and the ending point in the local reference route are selected as the splicing points, and the corresponding closest point in the first local reference route is searched according to the starting point and the ending point, and the splicing is performed.
The multi-scenario-oriented automatic driving planning method according to claim 1, wherein in S3, the starting planning point on the final local reference route is calculated according to the self-vehicle positioning information, and then the Convert Cartesian coordinates to Frenet coordinates;

Calculate the starting point of the autonomous vehicle in the Frenet coordinate system and the state information of the starting point, and the state information of the starting point is represented as the state at time T0;

Before sampling, the obstacle information is processed first, and the corresponding longitudinal displacement-lateral displacement map and longitudinal displacement-time map are established based on the reference route when the obstacle is processed;

After the obstacle information processing is completed, start to sample the final state at the next T1 time, which is expressed as the state at the T1 time;

After sampling, polynomial fitting is performed between the state at time T1 and the state at time T0 to generate two polynomial trajectories in the horizontal and vertical directions, and two-dimensional synthesis of the two polynomial trajectories in the horizontal and vertical directions is performed to obtain a trajectory set;

Find the optimal trajectory from the trajectory set.
A multi-scene-oriented automatic driving planning method according to claim 6, wherein the process of finding the optimal trajectory from the trajectory set comprises:

Calculate the cost C total for each trajectory in the trajectory set:

Perform physical limit detection and collision detection on the trajectory with the lowest cost. If both the vehicle motion constraint conditions and the collision-free condition are met, the trajectory in the Frenet coordinate system is converted into the trajectory in the Cartesian coordinate system and returned to the trajectory; if not satisfied If at least one of the vehicle motion constraints and the no-collision condition is satisfied, the trajectory is eliminated, and the trajectory with the lowest cost is selected to continue detection until a trajectory that satisfies both of the above conditions is found;

If no matching trajectory is found, then proceed to S4.
A multi-scene-oriented automatic driving planning method according to claim 7, characterized in that:

The cost C total of each trajectory in the trajectory set is calculated by the following formula:

C total = ω lon_obj *C lon_obj +ω lon_jerk *C lon_jerk +ω collision *C collision +ω lat_offset *C lat_offset

Among them, ω lon_obj , ω lon_jerk , ω collision and ω lat_offset are the weight of missed target, longitudinal jerk weight, collision weight and lateral offset weight, respectively, C lon_obj , C lon_jerk , C collision and C lat_offset are missed target, respectively Target cost, longitudinal jerk cost, collision cost, lateral offset cost.
The multi-scenario-oriented automatic driving planning method according to claim 1, wherein the position of the self-vehicle is obtained by reading GPS positioning and laser SLAM positioning on the vehicle.
A multi-scenario-oriented automatic driving planning system, characterized in that, for realizing the multi-scenario-oriented automatic driving planning method according to any one of claims 1-9, the system includes:

Global planning module: used to parse the map topology information, obtain the global reference route, and output the global reference route;

Reference route providing module: used to generate a local reference route according to whether there is an optimal trajectory, the position of the vehicle and the global reference route. The process includes;

If there is no optimal trajectory, the first local reference route is obtained by sampling the waypoints according to the position of the vehicle and the global reference route. At the same time, the hybrid A* algorithm is used to search and generate the reference route corresponding to the traffic section; the reference route is spliced and integrated into the first local reference route. In the reference route, a complete local reference route is obtained; the reference route is smoothed under the conditions of vehicle kinematic constraints and obstacle constraints to obtain the final local reference route;

If there is an optimal trajectory, the first local reference route is obtained by waypoint sampling according to the position of the vehicle and the global reference route, and the first local reference route is smoothed to obtain the final local reference route;

And output the local reference route;

Local planning module: It is used to obtain the optimal trajectory according to the position of the self-vehicle, obstacle information and the final local reference route, and output the result of whether there is an optimal trajectory.