Method and system for dynamic election and evaluation of access points for unattended streaming vehicle ld mapping

Abstract

The application relates to the technical field of automatic driving, in particular to a dynamic election and evaluation method and system of an unmanned logistics vehicle LD map access point. Through an innovative multi-source fusion strategy and an intelligent switching mechanism, a dynamic multi-access point election mechanism is adopted at the edge of a high-precision map, an optimal access point is intelligently selected, the accessibility, smoothness and road occupancy state of a line are considered in an access process, and dynamic updating is carried out during driving. The application greatly improves the autonomous operation capability and robustness of the unmanned logistics vehicle in a complex and dynamic traffic environment, reduces the safety degradation triggered by improper switching, and improves the overall operation efficiency and commercial value of the unmanned logistics vehicle.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of autonomous driving, in particular to a dynamic election and evaluation method and system for a logistics vehicle LD map access point. BACKGROUND

[0002] In existing autonomous logistics vehicle driving systems, when the vehicle needs to switch between different precision and type maps (such as SLAM high-precision maps and lightweight point cloud / feature maps LD maps), a method of presetting fixed switching points is usually adopted. This method can work in ideal conditions, but it has serious shortcomings in complex real-world environments. First, fixed switching points lack flexibility and cannot adapt to dynamically changing environments. Second, if the preset switching points become inaccessible or occupied due to road construction, temporary obstacles (such as illegally parked vehicles, piled goods), etc., the vehicle will be unable to complete the switching, resulting in task interruption. This rigid switching point selection strategy makes the autonomous driving system very vulnerable in the face of uncertainty, making it difficult to ensure continuity and reliability of travel.

[0003] Another core defect of traditional switching methods is that they often lack effective evaluation of the surrounding dynamic environment in the switching decision-making process. The system usually only focuses on whether the vehicle has reached the preset geographic coordinates, ignoring the real-time conditions around the coordinate point. For example, the system may not consider whether the switching point is occupied by other vehicles or pedestrians, nor evaluate whether the path from the current location to the switching point is safe and smooth. This "blind elephant" type of switching makes it easy for the vehicle to collide with dynamic obstacles during switching, or to seriously affect driving safety due to the need for emergency braking, sharp turning, and other dangerous operations. Therefore, a mechanism that can comprehensively and real-time evaluate the state of the switching point is crucial for improving the safety of autonomous driving systems.

[0004] Due to the above two shortcomings, the most direct consequence of traditional switching methods is that the vehicle "stuck" or appears in a critical situation during switching. When the preset switching point is occupied by an obstacle, the vehicle is in a dilemma: if it forces through, a collision may occur; if it waits, it will indefinitely delay the task. In this case, the system usually only triggers a safety degradation and requests manual intervention, which not only seriously affects the operational efficiency of the autonomous logistics vehicle, but also goes against the original intention of autonomous driving technology to improve efficiency and reduce labor costs. In addition, even if the switching point itself is not occupied, but if there are obstacles on the path to the point, or the switching requires emergency braking, sharp turning, and other drastic operations, it will still pose a great threat to driving safety. SUMMARY

[0005] Therefore, the present application aims to provide a dynamic election and evaluation method and system for unmanned flow vehicle LD map access points to solve the problems in the background art.

[0006] To achieve the above-mentioned purpose, the present application adopts the following technical solutions:

[0007] The dynamic election and evaluation method for unmanned flow vehicle LD map access points of the present application comprises the following steps:

[0008] When the unmanned flow vehicle travels to a switching area, the current positioning of the unmanned flow vehicle and a roaming link pool are obtained, wherein the roaming link pool comprises a plurality of pre-constructed access points, and the switching area is the area between the high-precision map and the lightweight automatic driving map;

[0009] A plurality of travel routes from the current positioning to the plurality of access points in the roaming link pool are generated, and a plurality of candidate routes and a plurality of candidate access points are selected from the plurality of travel routes;

[0010] The inherent features of the candidate routes are extracted, and the reachability score and the smoothness score of the plurality of candidate routes are calculated based on the inherent features, and the initial occupancy state score of the plurality of candidate routes is determined, wherein the inherent features include road length, congestion coefficient, number of road bends, and road feasible region width;

[0011] The comprehensive score of the plurality of candidate routes is calculated based on the reachability score, the smoothness score, and the initial occupancy state score, and the target route and the target access point are selected based on the comprehensive score;

[0012] When the unmanned flow vehicle travels to the target access point along the target route, the occupancy state score and the comprehensive score are updated based on the image data and the radar data collected by the unmanned flow vehicle in real time, and the target route and the target access point are updated based on the updated comprehensive score until any one target access point is reached.

[0013] In an embodiment of the present application, updating the occupancy state score and the comprehensive score based on the image data and the radar data collected by the unmanned flow vehicle in real time comprises:

[0014] Obtaining positioning information, inertial data, radar point cloud data, and visual data collected by the unmanned flow vehicle in real time;

[0015] Determining the initial pose of the unmanned flow vehicle based on the positioning information and the inertial data, and performing Kalman filtering on the initial pose to obtain an optimized pose;

[0016] Converting the coordinate system of the radar point cloud data into the coordinate system of the high-precision map, and preprocessing the radar point cloud data to obtain preprocessed point cloud data;

[0017] obtain a matching confidence of each point in the pre-processed point cloud data based on the optimized pose and matching the pre-processed point cloud data with reference point cloud data obtained from the high-precision map;

[0018] take points with a confidence greater than a preset confidence threshold as a scene point cloud, and take points with a confidence less than or equal to the preset confidence threshold as an obstacle point cloud, wherein the scene point cloud comprises a road domain point cloud;

[0019] identify the visual data based on a pre-constructed visual recognition model to obtain road marking features;

[0020] fuse the road marking features with the road domain point cloud to obtain a road domain model;

[0021] perform road occupancy state judgment based on the road domain model and the obstacle point cloud, and update an occupancy state score and a comprehensive score according to a judgment result.

[0022] In an embodiment of the present application, performing road occupancy state judgment based on the road domain model and the obstacle point cloud, and updating an occupancy state score and a comprehensive score according to a judgment result, comprises:

[0023] perform clustering on the obstacle point cloud to obtain a plurality of point cloud clusters, wherein each point cloud cluster represents an obstacle;

[0024] extract geometric features of the plurality of point cloud clusters, wherein the geometric features comprise a length , a width , and a height of a minimum rectangular frame;

[0025] match corresponding point cloud clusters at different time points based on the geometric features to obtain position information of a same point cloud cluster at different time points;

[0026] determine a motion speed of the obstacle based on the position information of the same point cloud cluster at different time points, and take an obstacle with a motion speed less than a preset speed threshold as a static obstacle;

[0027] calculate a proportion of a minimum circumscribed rectangular frame of the static obstacle to a size of the road domain, and set the occupancy state score to zero when the proportion of any one of the static obstacles exceeds a preset proportion threshold, or set the occupancy state score to 1 otherwise; update the comprehensive score based on the updated occupancy state score.

[0028]

[0029] ​​​In an embodiment of the present application, the method for constructing the roaming link pool comprises:

[0030] mapping the switching area into a two-dimensional map, and extracting a road network and road network nodes from the switching area;

[0031] dividing the road network based on the road network nodes to obtain a plurality of branch roads, and uniformly sampling an axis of each branch road to obtain a plurality of sampling points ;

[0032] extracting signal base station positions from the two-dimensional map, and calculating a first distance between a plurality of sampling points and the signal base station positions and a second distance between the plurality of sampling points and the nearest road network nodes ;

[0033] calculating an elimination score of the sampling points of each branch road based on the first distance and the second distance , wherein the elimination score of each branch road is calculated according to the following mathematical expression:

[0034]

[0035] In the formula, w1 is a first weight, and w2 is a second weight.

[0036] taking the sampling point with the lowest elimination score of each branch road as an access point, and constructing the roaming link pool based on a plurality of access points.

[0037] In an embodiment of the present application, the method for calculating the reachability score of a plurality of candidate routes based on the inherent characteristics comprises:

[0038] obtaining a congestion coefficient and a route distance of each candidate route;

[0039] calculating a reachability score of each candidate route based on the congestion coefficient and the route distance of each candidate route , wherein the reachability score of each candidate route is calculated according to the following mathematical expression:

[0040]

[0041] In the formula, c represents the congestion coefficient, d represents the route distance, and f represents a normalization processing function.

[0042] ​​​​​​In an embodiment of the present application, the comprehensive scores of the plurality of candidate routes are calculated based on the reachability scores, the smoothness scores and the initial occupancy state scores, and the target route and the target access point are selected based on the comprehensive scores, comprising:

[0043] reachability scores of each candidate route smoothness scores and occupancy state scores comprehensive scores of each candidate route are calculated , wherein the comprehensive scores The mathematical expression is:

[0044]

[0045] In the formula, is the third weight, is the fourth weight, indicates the serial number of the candidate route; when the candidate route is in the occupancy state, , when the candidate route is not in the occupancy state, ; the initial value of the occupancy state score of the plurality of candidate routes is 1;

[0046] The candidate route with the highest comprehensive score is taken as the target candidate route, and the access point corresponding to the target candidate route is taken as the target access point.

[0047] The present application also provides a dynamic election and evaluation system of unmanned flow car LD access points, comprising:

[0048] An acquisition module is configured to acquire a current positioning of an unmanned flow car and a roaming link pool when the unmanned flow car drives to a switching area, wherein the roaming link pool includes a plurality of pre-constructed access points, and the switching area is an area between a high-precision map and a lightweight automatic driving map.

[0049] A screening module is configured to generate a plurality of driving routes from the current positioning to a plurality of access points in the roaming link pool, and screen a plurality of candidate routes and a plurality of candidate access points from the plurality of driving routes.

[0050] An initial score module is configured to extract inherent features of the candidate routes, calculate reachability scores and smoothness scores of the plurality of candidate routes based on the inherent features, and determine initial occupancy state scores of the plurality of candidate routes, wherein the inherent features include congestion coefficients, numbers of road bends and widths of road feasible regions.

[0051] a comprehensive score module configured to calculate comprehensive scores of the plurality of candidate routes based on the reachability scores, the smoothness scores, and the initial occupancy state scores, and select a target route and a target access point based on the comprehensive scores;

[0052] an updating module configured to update the occupancy state scores and the comprehensive scores based on image data and radar data collected by the unmanned flow vehicle in real time while the unmanned flow vehicle travels along the target route to the target access point, and update the target route and the target access point based on the updated comprehensive scores until any one of the target access points is reached.

[0053] The application has the following beneficial effects: The dynamic election and evaluation method and system of the unmanned flow vehicle LD map access point of the application adopt a dynamic multi-access point election mechanism at the edge of a high-precision map, intelligently select an optimal access point, consider reachability, smoothness, and road occupancy state in the access process, and dynamically update during travel. The application greatly improves the autonomous operation capability and robustness of the unmanned flow vehicle in a complex and dynamic traffic environment, and also reduces the triggering of safety degradation due to improper switching, thereby improving the overall operation efficiency and commercial value of the unmanned flow vehicle. BRIEF DESCRIPTION OF DRAWINGS

[0054] The application will be further described below in conjunction with the drawings and embodiments:

[0055] Figure 1 is a system framework diagram shown in an embodiment of the application;

[0056] Figure 2 is a flowchart of the dynamic election and evaluation method of the unmanned flow vehicle LD map access point in an embodiment of the application;

[0057] Figure 3 is a structure diagram of the dynamic election and evaluation system of the unmanned flow vehicle LD map access point in an embodiment of the application. DETAILED DESCRIPTION

[0058] The embodiments of the application are described below through specific and concrete examples, and those skilled in the art can easily understand other advantages and effects of the application from the disclosure of the specification. The application can also be implemented or applied through other different specific embodiments, and the details in the specification can be modified or changed based on different views and applications without departing from the spirit of the application. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0059] It should be noted that the diagrams provided in the following embodiments only illustrate the basic concepts of the present application in a schematic manner, and only show the layers related to the present application in the diagrams, not according to the number, shape and size of the layers in actual implementation. The actual implementation of each layer may be a random change, and the layer layout pattern may be more complex.

[0060] In the following description, a large number of details are discussed to provide a more thorough explanation of the embodiments of the present application, however, it is obvious to those skilled in the art that the embodiments of the present application can be implemented without these specific details.

[0061] Figure 1 is a system framework diagram shown in an embodiment of the present application, as Figure 1 shown, the system applied to the dynamic election and evaluation method of the access point of the unmanned logistics vehicle LD map in the present application includes a perception module, a positioning module, a decision planning module and a control execution module.

[0062] The perception module is responsible for collecting real-time environmental information around the vehicle through sensors such as laser radar, camera, millimeter wave radar, etc., including obstacles, lane lines, traffic signs, etc. The positioning module is the core of the present application, including a GPS unit and an IMU unit, which is responsible for fusing multi-source sensor data and various map information (SLAM map, LD map), and performing dynamic access point election and map switching. The decision planning module makes behavior decisions and path planning according to the accurate pose provided by the GPS and IMU in the positioning module and the environmental information provided by the perception module, and generates a smooth driving trajectory. The control execution module is responsible for converting the planned trajectory into specific vehicle control instructions (such as steering, acceleration, braking), driving the vehicle to travel smoothly. Each module interacts with data through a high-speed bus to form a closed-loop control system, ensuring that the vehicle can safely and efficiently complete the smooth switching of the positioning mode. Figure 1

[0063] ​In an autonomous driving system, when a vehicle travels to a preset switching boundary of different maps (such as a SLAM high-precision map and a lightweight point cloud / feature map), a traditional method is to use a single-point hard switching mode. This mode sets a unique and fixed switching point on the switching boundary. When the vehicle reaches the point, the system immediately switches from the current positioning mode to another positioning mode. However, this seemingly simple switching method has a significant engineering risk: if the switching point coordinate is occupied by a temporarily parked vehicle, a pedestrian, a stacked cargo, or other dynamic obstacles, the vehicle will not be able to reach the point smoothly, thus being "stuck" in the switching area. At this time, the vehicle either passively waits for the obstacles to be cleared or is forced to trigger a degraded mode or request manual takeover, which seriously affects the continuity and efficiency of travel, especially in scenarios such as logistics transportation that require extremely high time efficiency. Such interruptions are unacceptable.

[0064] Based on the above scenario, the present application provides the following technical solutions to solve the above technical problems.

[0065] Figure 2 is a flowchart of the dynamic election and evaluation method of the access point of the logistics vehicle LD map in an embodiment of the present application, as shown in Figure 2 The dynamic election and evaluation method of the access point of the logistics vehicle LD map in the embodiment includes the following steps:

[0066] S210, when the logistics vehicle travels to a switching area, obtaining the current positioning of the logistics vehicle and a roaming link pool, wherein the roaming link pool includes a plurality of pre-constructed access points, and the switching area is the area between a SLAM high-precision map and a lightweight autonomous driving map (LD map);

[0067] S220, generating a plurality of travel routes from the current positioning to a plurality of access points in the roaming link pool, and screening a plurality of candidate routes and a plurality of candidate access points from the plurality of travel routes;

[0068] In the present application, the construction process of the roaming link pool includes:

[0069] (1) mapping the switching area to a two-dimensional map, and extracting a road network and road network nodes from the switching area;

[0070] It is ensured that the switching area has sufficient width to avoid sudden signal interruption at the boundary, and a clear geographical range is provided for subsequent access point selection, avoiding invalid positions of the access point outside the boundary.

[0071] (2) dividing the road network based on the road network nodes to obtain a plurality of branch roads, and uniformly sampling the axis of each branch road to obtain a plurality of sampling points ;

[0072] The uniform sampling ensures the uniform distribution of the access points on the road network, avoiding the over-concentration or sparseness of the access points; the division of the multiple branch roads can cover different driving paths, improving the reliability of the roaming.

[0073] (3) Extracting the signal base station positions from the two-dimensional map, and calculating the first distances between the multiple sampling points and the signal base station positions , and the second distances between the multiple sampling points and the nearest road network nodes ;

[0074] (4) Calculating the elimination scores of the sampling points of each branch road based on the first distances and the second distances , wherein the mathematical expression of the elimination score is as follows:

[0075]

[0076] In the formula, w1 is the first weight, and w2 is the second weight.

[0077] (5) Taking the sampling point with the lowest elimination score of each branch road as the access point, and constructing the roaming link pool based on the multiple access points.

[0078] The above process quantitatively evaluates the quality of each sampling point, providing an objective basis for the selection of the access point; by adjusting the weights w1 and w2, different environmental requirements (such as increasing the weight of w1 in the area with weak signal) can be adapted. It is ensured that the finally selected access point is optimal in signal and path, and the success rate of roaming is improved.

[0079] After the roaming link pool is constructed, multiple candidate access points are extracted from the pre-constructed roaming link pool based on the first positioning, including:

[0080] S221, calling the API interface of the navigation tool to determine multiple navigation routes from the first positioning to the target location, wherein the target location is located in the switching area;

[0081] S222, screening N candidate routes with the shortest distance from the multiple navigation routes; and extracting candidate access points located in the candidate routes from the roaming link pool.

[0082] ​​By screening the shortest N candidate routes, it is ensured that the selected route is the shortest in driving distance, improving driving efficiency and reducing energy consumption. The access points located on the candidate routes are extracted from the roaming link pool, ensuring that the access points are highly matched with the actual driving path, avoiding invalid selection of access points outside the path.

[0083] In S230, inherent features of the candidate routes are extracted, reachability scores and smoothness scores of the candidate routes are calculated based on the inherent features, and initial occupancy state scores of the candidate routes are determined.

[0084] After establishing the multi-candidate access point pool, how to select the optimal access point from it is another key technology of the present application. The present application proposes a dynamic real-time election mechanism. When the vehicle approaches the switching boundary and the confidence of the current positioning mode (such as SLAM) is still maintained at a high level, the system will activate the perception and prediction of the front LD map area in advance. The system will evaluate the state of all candidate access points in real time, and the core evaluation dimensions include: reachability (whether there is an unobstructed and traffic rule-compliant drivable path between the access point and the current position of the vehicle), occupancy state (whether the access point is occupied by static or dynamic obstacles detected by the vehicle-mounted perception system), and smoothness (whether the steering and speed adjustment required for switching from the current pose to the access point is smooth). The system will dynamically elect the currently optimal access point according to the comprehensive scores of these dimensions. This dynamic election mechanism enables the system to intelligently adapt to the real-time changing traffic environment and make optimal decisions, thereby ensuring the safety, efficiency and smoothness of the switching process.

[0085] The reachability score and the smoothness score are generated based on inherent features of the routes, and the reachability score of the candidate routes is calculated based on the inherent features, including:

[0086] In S231, the congestion coefficient and the route distance of each candidate route are obtained.

[0087] In S232, the reachability score of each candidate route is calculated based on the congestion coefficient and the route distance of each candidate route. The mathematical expression of the reachability score is:

[0088]

[0089] In the formula, c represents the congestion coefficient, d represents the route distance, and norm represents a normalization processing function.

[0090] ​​​​calculating smoothness scores of the plurality of routes based on the intrinsic features, comprising:

[0091] S233, extracting the number of target curves with curvatures greater than a set curvature threshold and the average road width of each candidate route;

[0092] S234, calculating the smoothness score of each candidate route based on the number of target curves and the average road width .

[0093] Since it is unknown whether each candidate route is occupied by an obstacle, the initial value of the occupancy state score of each candidate route is 1 in the initial state.

[0094] S240, calculating the comprehensive score of the plurality of candidate routes based on the accessibility score, the smoothness score and the initial occupancy state score, and selecting a target route and a target access point based on the comprehensive score;

[0095] Based on the above accessibility score , smoothness score and occupancy state score , the mathematical expression of the comprehensive score is:

[0096]

[0097] In the formula, is the third weight, is the fourth weight, indicates the serial number of the candidate route; when the candidate route is in the occupancy state, , when the candidate route is not in the occupancy state, ; the initial value of the occupancy state score of the plurality of candidate routes is 1;

[0098] In the above formula, the accessibility score , the smoothness score are weighted and summed, and then multiplied by the occupancy state score . The internal logic is: when the road is not occupied by an obstacle, normal driving, at this time, the priority is determined by the accessibility score , the smoothness score . When the road is occupied by an obstacle, it means that the road is not drivable, and its priority is directly zero.

[0099] Finally, the candidate route with the highest comprehensive score is selected as the target candidate route, and the access point corresponding to the target candidate route is selected as the target access point.

[0100] ​S250, while the unmanned flow vehicle is driving along the target line to the target access point, updating the occupancy state score and the comprehensive score based on the image data and radar data collected by the unmanned flow vehicle in real time, and updating the target line and the target access point based on the updated comprehensive score until reaching any one of the target access points.

[0101] After obtaining the target candidate line and the target access point, the system then automatically controls the unmanned flow vehicle to drive according to the target candidate line and the target access point. In this process, the positioning module in the unmanned flow vehicle obtains positioning information in real time, the inertial measurement unit obtains inertial data in real time, the camera collects visual data in real time, and the radar collects radar point cloud data in real time. The occupancy state score is automatically updated based on the above data, and the process includes:

[0102] S251, obtaining the positioning information, inertial data, radar point cloud data and visual data collected by the unmanned flow vehicle in real time;

[0103] Specifically, the data is collected in real time through the vehicle CAN bus or the ROS (Robot Operating System) data bus, which specifically includes:

[0104] Positioning information: from RTK-GPS (accuracy ±1cm) or Beidou positioning system.

[0105] Inertial data: from IMU (inertial measurement unit, including accelerometer, gyroscope, magnetometer).

[0106] Radar point cloud data: from 4 vehicle-grade solid-state laser radars (360° scanning, 150m detection range).

[0107] Visual data: from 14 fisheye cameras (covering a range of 20 meters around the vehicle body, resolution 1920x1080).

[0108] S252, determining an initial pose of the unmanned flow vehicle based on the positioning information and the inertial data, and performing Kalman filtering on the initial pose to obtain an optimized pose;

[0109] The initial pose (x, y, z, yaw, pitch, roll) is obtained from GPS / IMU, wherein x, y, z are three-dimensional positioning, yaw is the yaw angle, pitch is the pitch angle, and roll is the roll angle.

[0110] Kalman filtering fuses GPS (high position accuracy but susceptible to shielding) and IMU (position prone to drift but fast response) data through prediction and update steps; effectively suppresses GPS signal fluctuation and IMU drift, and improves positioning accuracy.

[0111] S253, convert the coordinate system of the radar point cloud data into the coordinate system of the high-precision map, and pre-process the radar point cloud data to obtain pre-processed point cloud data;

[0112] The coordinate system conversion is performed through a rotation matrix and a translation vector pre-calibrated, and the pre-processing includes statistical filtering (removing outliers), voxel filtering (down-sampling to improve computational efficiency), and ground point cloud separation (based on a height threshold).

[0113] S254, based on the optimized pose, intercept reference point cloud data from the high-precision map, and match the pre-processed point cloud data with the reference point cloud data to obtain the matching confidence of each point in the pre-processed point cloud data;

[0114] According to the pose information (position + attitude) of the vehicle, the reference point cloud is intercepted from the high-precision map, which is essentially a spatial query process: taking the current position of the vehicle as the center, the point cloud data within a specified range in the high-precision map database is retrieved.

[0115] In this application, the GICP algorithm is used for point cloud matching, and the confidence of each point The mathematical expression is:

[0116]

[0117] In the formula, represents the current iteration error, represents the initial error.

[0118] GICP considers the local geometric characteristics of points through the covariance matrix, and the matching confidence reflects the matching quality. The higher the value, the more reliable the matching, and the more similar to the reference point cloud. Conversely, it is not similar and may be an obstacle point cloud.

[0119] S255, points with a confidence greater than a pre-set confidence threshold are taken as scene point clouds, and points with a confidence less than or equal to the pre-set confidence threshold are taken as obstacle point clouds, wherein the scene point clouds include road domain point clouds;

[0120] Based on the matching confidence threshold (default 0.8), high-confidence points indicate good matching with the high-precision map and belong to the road domain. Low-confidence points indicate poor matching with the map and may belong to obstacles.

[0121] S256, based on a pre-constructed visual recognition model, the visual data is recognized to obtain road marking features;

[0122] Specifically, semantic segmentation is performed using a deep learning model (such as U-Net), and the output includes the coordinates, type (solid / dashed), width, and other information of the road markings. This provides semantic information of the road and makes up for the shortcomings of point cloud perception.

[0123] S257, the road marking features are fused with the road domain point cloud to obtain a road domain model;

[0124] Specifically, the road markings are converted to the point cloud coordinate system, and then the road markings are fused with the road domain point cloud to obtain a road domain model containing image segmentation semantic information.

[0125] S258, based on the road domain model and the obstacle point cloud, a road occupancy state is determined, and an occupancy state score and a comprehensive score are updated according to the determination result, specifically including:

[0126] S2581, the obstacle point cloud is clustered to obtain a plurality of point cloud clusters, wherein each point cloud cluster represents an obstacle;

[0127] The density-based clustering algorithm clusters spatially adjacent points into a class, effectively distinguishes independent obstacles, and avoids merging multiple obstacles.

[0128] S2582, geometric features of the plurality of point cloud clusters are extracted, wherein the geometric features include length , width , and height of a minimum rectangular frame;

[0129] The minimum rectangular frame (AABB) of each point cloud cluster is calculated, and the corresponding length , width , and height are extracted. The size of the obstacle is quantified, which is used for subsequent occupancy state calculation and provides basic data for obstacle classification and identification.

[0130] S2583, corresponding point cloud clusters at different time points are matched based on the geometric features to obtain position information of the same point cloud cluster at different time points;

[0131] The same obstacle in consecutive frames is matched by spatial position similarity, effectively realizing obstacle tracking.

[0132] S2584, the motion speed of the obstacle is determined based on the position information of the same point cloud cluster at different time points, and the obstacle with a motion speed less than a preset speed threshold is regarded as a static obstacle;

[0133] The mathematical expression of the motion speed of the object is:

[0134]

[0135] wherein, represents the position of the centroid of the rectangular frame at time point , represents the position of the centroid of the rectangular frame at time point , represents the time interval (usually 0.1S).

[0136] Obstacles with a speed less than a preset threshold (such as 0.5m / s) are considered static, and the obstacles mainly targeted by the present application are static obstacles. If they are dynamic, the vehicle can be driven in a following manner.

[0137] S2585, the proportion of the size of the minimum circumscribed rectangular frame of the static obstacle to the road domain is calculated, and when the proportion of any one static obstacle exceeds a preset proportion threshold, the occupancy state score is set to zero, otherwise the occupancy state score is set to 1.

[0138] Specifically, the width of the rectangular frame can be projected to the road section, and if the proportion exceeds 0.8, it indicates that the obstacle seriously occupies the road. The current road is in an impassable state. In the impassable state, , otherwise, .

[0139] S2586, the comprehensive score is updated based on the updated occupancy state score.

[0140] Finally, the comprehensive score is updated based on the updated occupancy state score , and the road with the highest comprehensive score at the current position is selected. Thus, the traffic strategy can be dynamically adjusted in real time to avoid the deadlock of the unmanned flow vehicle.

[0141] The present application effectively solves the difficult problem of the switch point being occupied by obstacles in the traditional switching mode through the innovative "LD map access point roaming link" mechanism. In the traditional single-point hard switching mode, if the preset switching point is occupied by temporarily parked vehicles, pedestrians or other obstacles, the vehicle will be forced to "get stuck", seriously affecting the driving efficiency. However, the present application assigns great flexibility to the system by presetting multiple candidate access points and dynamically selecting the optimal one. Even if a certain access point is occupied, the system can immediately select another available access point, thereby avoiding the dilemma of the vehicle being trapped in waiting or requiring manual intervention. This innovation greatly improves the autonomous operation ability and robustness of the unmanned flow vehicle in complex and dynamic traffic environments.

[0142] As shown in Figure 3 , the present application also provides a dynamic election and evaluation system for the LD map access points of the unmanned flow vehicle, comprising:​​​

[0143] an acquisition module, configured to acquire a current positioning of the unmanned flow vehicle and a roaming link pool when the unmanned flow vehicle drives to a switching area, wherein the roaming link pool comprises a plurality of pre-constructed access points, and the switching area is an area between a high-precision map and a lightweight automatic driving map;

[0144] a screening module, configured to generate a plurality of driving routes from the current positioning to the plurality of access points in the roaming link pool, and screen a plurality of candidate routes and a plurality of candidate access points from the plurality of driving routes;

[0145] an initial scoring module, configured to extract inherent features of the candidate routes, calculate reachability scores and smoothness scores of the plurality of candidate routes based on the inherent features, and determine initial occupancy state scores of the plurality of candidate routes, wherein the inherent features comprise a congestion coefficient, a number of road bends, and a road feasible region width;

[0146] a comprehensive scoring module, configured to calculate comprehensive scores of the plurality of candidate routes based on the reachability scores, the smoothness scores, and the initial occupancy state scores, and select a target route and a target access point based on the comprehensive scores;

[0147] an updating module, configured to update the occupancy state scores and the comprehensive scores based on image data and radar data collected by the unmanned flow vehicle in real time when the unmanned flow vehicle drives along the target route to the target access point, and update the target route and the target access point based on the updated comprehensive scores until any one target access point is reached.

[0148] The dynamic election and evaluation method and system of the LD map access point of the unmanned flow vehicle provided in the application greatly improve the autonomous operation ability and robustness of the unmanned flow vehicle in a complex and dynamic traffic environment, and also reduce the safety degradation triggered due to improper switching, thereby improving the overall operation efficiency and commercial value of the unmanned flow vehicle.

[0149] The embodiment also provides an electronic terminal, comprising a processor and a memory.

[0150] The memory is used for storing a computer program, and the processor is used for executing the computer program stored in the memory, so that the terminal executes any one method in the embodiment.

[0151] The computer readable storage medium in the embodiment can be understood by those skilled in the art that all or part of the steps of the above-mentioned method embodiments can be completed by a computer program related hardware. The foregoing computer program can be stored in a computer readable storage medium. The program executes the steps of the above-mentioned method embodiments when executed; and the foregoing storage medium includes ROM, RAM, magnetic disc or optical disc and various storage medium that can store program codes.

[0152] The electronic terminal provided in the embodiment includes a processor, a memory, a transceiver and a communication interface. The memory and the communication interface are connected with the processor and the transceiver and complete communication between each other. The memory is used for storing a computer program. The communication interface is used for communication. The processor and the transceiver are used for running the computer program, so that the electronic terminal executes each step of the method.

[0153] In the embodiment, the memory can include a random access memory (RAM) and can also include a non-volatile memory, for example, at least one disk memory.

[0154] The processor described above can be a general processor, including a central processing unit (CPU), a network processor (NP) and the like; can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component.

[0155] In the above-mentioned embodiments, although the application has been described in combination with specific embodiments of the application, many alternatives, modifications and variations of the embodiments will be apparent to those skilled in the art according to the foregoing description. The embodiments of the application are intended to cover all such alternatives, modifications and variations which fall within the broad scope of the appended claims.

[0156] The above embodiments are only illustrative of the principles of the present application and its effects, and are not intended to limit the present application. Any modification or change made by any person skilled in the art without departing from the spirit and scope of the present application shall be covered by the claims of the present application.

Claims

1. A method for dynamic election and evaluation of access points in the LD map of unmanned logistics vehicles, characterized in that, Including the following steps: When the unmanned logistics vehicle travels to the switching area, the current location of the unmanned logistics vehicle and the roaming link pool are obtained. The roaming link pool includes multiple pre-built access points. The switching area is the area between the high-precision map and the lightweight autonomous driving map. Generate multiple travel routes from the current location to multiple access points in the roaming link pool, and filter out multiple candidate routes and multiple candidate access points from the multiple travel routes; Extract the inherent features of candidate routes, and calculate the accessibility score and smoothness score of multiple candidate routes based on the inherent features, and determine the initial occupancy status score of multiple candidate routes. The inherent features include road length, congestion coefficient, number of road curves and width of road feasible area. The comprehensive score of multiple candidate lines is calculated based on the reachability score, the smoothness score, and the initial occupancy score, and the target line and target access point are selected based on the comprehensive score. When the unmanned logistics vehicle travels along the target route to the target access point, the occupancy status score and comprehensive score are updated based on the image data and radar data collected in real time by the unmanned logistics vehicle. The target route and target access point are also updated based on the updated comprehensive score until any target access point is reached. The updating process of the occupancy status score and comprehensive score includes: by fusing multi-source sensor data of the unmanned logistics vehicle, performing pose optimization, point cloud registration and confidence screening, and combining road marking features recognized by vision to construct a road domain model and obstacle point cloud, and judging the road occupancy status and updating the occupancy status score and comprehensive score based on the road domain model and the obstacle point cloud.

2. The method for dynamic election and evaluation of LD map access points for unmanned logistics vehicles according to claim 1, characterized in that, The occupancy status score and comprehensive score are updated based on the real-time image and radar data collected by the unmanned logistics vehicle, including: Acquire real-time positioning information, inertial data, radar point cloud data, and visual data collected by unmanned logistics vehicles; The initial pose of the unmanned logistics vehicle is determined based on the positioning information and the inertial data, and Kalman filtering is applied to the initial pose to obtain the optimized pose. The coordinate system of the radar point cloud data is converted into the coordinate system of a high-precision map, and the radar point cloud data is preprocessed to obtain preprocessed point cloud data. Based on the optimized pose, reference point cloud data is extracted from the high-precision map, and the preprocessed point cloud data is matched with the reference point cloud data to obtain the matching confidence of each point in the preprocessed point cloud data. Points with a confidence level greater than a preset confidence threshold are used as scene point clouds, and points with a confidence level less than or equal to the preset confidence threshold are used as obstacle point clouds. The scene point clouds include road domain point clouds. The visual data is identified based on a pre-built visual recognition model to obtain road marking features; The road marking features are fused with the road domain point cloud to obtain a road domain model; The road occupancy status is determined based on the road domain model and the obstacle point cloud, and the occupancy status score and comprehensive score are updated according to the determination results.

3. The method for dynamic election and evaluation of LD map access points for unmanned logistics vehicles according to claim 2, characterized in that, Based on the road domain model and the obstacle point cloud, the road occupancy status is determined, and the occupancy status score and comprehensive score are updated according to the determination result, including: The obstacle point cloud is clustered to obtain multiple point cloud clusters, where each point cloud cluster represents an obstacle; Extract geometric features from multiple point cloud clusters, wherein the geometric features include the length of the smallest bounding box. ,Width and high ; Based on geometric features, point cloud clusters at different time points are matched to obtain the location information of the same point cloud cluster at different time points; The movement speed of obstacles is determined based on the location information of the same point cloud cluster at different time points, and obstacles with movement speeds less than a preset speed threshold are classified as static obstacles. Calculate the ratio of the minimum bounding rectangle of a static obstacle to the size of the road domain. If the ratio of any static obstacle exceeds a preset threshold, the obstacle will be assigned an occupation status score. Set to zero; otherwise, it will occupy the state score. Set to 1; Based on the updated occupancy status score Update the overall rating.

4. The method for dynamic election and evaluation of LD map access points for unmanned logistics vehicles according to claim 1, characterized in that, The method for constructing the roaming link pool includes: The switching area is mapped onto a two-dimensional map, and the road network and road network nodes are extracted from the switching area; The road network is divided based on the road network nodes to obtain multiple branch roads, and the axis of each branch road is uniformly sampled to obtain multiple sampling points. ; The locations of signal base stations are extracted from the two-dimensional map, and multiple sampling points are calculated. First distance from the location of the signal base station And calculate multiple sampling points Second distance to the nearest road network node ; Based on the first distance and the second distance Calculate the elimination score for each branch road sampling point. The elimination score The mathematical expression is: In the formula, As the first weight, As the second weight; The sampling point with the lowest elimination score for each branch road is used as the access point, and a roaming link pool is built based on multiple access points.

5. The method for dynamic election and evaluation of LD map access points for unmanned logistics vehicles according to claim 1, characterized in that, Based on the inherent characteristics, the reachability scores of multiple candidate routes are calculated, including: Obtain the congestion coefficient and distance for each candidate route; An accessibility score for each candidate route is calculated based on its congestion coefficient and distance. The reachability score The mathematical expression is: In the formula, Indicates the congestion coefficient. Indicates the distance of the route. This represents the normalization function.

6. The method for dynamic election and evaluation of LD map access points for unmanned logistics vehicles according to claim 1, characterized in that, The process involves calculating a comprehensive score for multiple candidate routes based on the reachability score, the smoothness score, and the initial occupancy score, and selecting a target route and target access point based on the comprehensive score, including: Based on the accessibility score of each candidate route Smoothness score and occupation status score Calculate the overall score for each candidate route. The comprehensive score The mathematical expression is: In the formula, As the third weight, As the fourth weight, Indicates the sequence number of the candidate line; when the candidate line is in an occupied state, When the candidate line is not in an occupied state, Occupancy status scoring of multiple candidate lines The initial value is set to 1; The candidate line with the highest comprehensive score is selected as the target candidate line, and the access point corresponding to the target candidate line is selected as the target access point.

7. A dynamic election and evaluation system for LD map access points of unmanned logistics vehicles, characterized in that, include: The acquisition module is used to acquire the current location and roaming link pool of the unmanned logistics vehicle when it travels to the switching area. The roaming link pool includes multiple pre-built access points, and the switching area is the area between the high-precision map and the lightweight autonomous driving map. The filtering module is used to generate multiple driving routes from the current location to multiple access points in the roaming link pool, and to filter multiple candidate routes and multiple candidate access points from the multiple driving routes; The initial scoring module is used to extract the inherent features of candidate routes, calculate the accessibility and smoothness scores of multiple candidate routes based on the inherent features, and determine the initial occupancy status scores of multiple candidate routes. The inherent features include congestion coefficient, number of road curves, and width of road feasible area. The comprehensive scoring module is used to calculate a comprehensive score for multiple candidate lines based on the reachability score, the smoothness score, and the initial occupancy status score, and to select a target line and a target access point based on the comprehensive score. The update module is used to update the occupancy status score and comprehensive score based on the image data and radar data collected in real time by the unmanned logistics vehicle when it travels along the target route to the target access point, and to update the target route and target access point based on the updated comprehensive score, until any target access point is reached. The update process of the occupancy status score and comprehensive score includes: by fusing multi-source sensor data of the unmanned logistics vehicle, performing pose optimization, point cloud registration and confidence filtering, and combining road marking features recognized by vision to construct a road domain model and obstacle point cloud, and judging the road occupancy status and updating the occupancy status score and comprehensive score based on the road domain model and the obstacle point cloud.

8. An electronic device, characterized in that, include: Processor and memory; The memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to cause the electronic device to perform the method as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 6.

Images