Vehicle planning method, device and equipment based on manual guidance and storage medium

Abstract

The invention relates to the technical field of automatic driving, and discloses a vehicle planning method, device and equipment based on manual guidance and a storage medium, which are used for improving the operation efficiency of an automatic driving vehicle. The manual guidance-based vehicle planning method comprises the steps of obtaining passing point data corresponding to at least one passing point manually marked on a supervision platform; based on the data of the at least one passing point, combining map data, real-time road condition data and vehicle state data, and utilizing a spatio-temporal data fusion algorithm to generate a constraint node with a target priority; performing mixed path planning processing according to the constraint nodes to obtain an initial path; on the basis of the initial path, differential adjustment is carried out on the necessary arriving point and the winding avoiding point, local re-planning is carried out through a dynamic optimization algorithm, and an optimized path is obtained; and generating a control instruction which can be executed by the target autonomous vehicle based on the optimized path so as to control the target autonomous vehicle to operate.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving technology, and in particular to a vehicle planning method, device, equipment and storage medium based on manual guidance. Background Art

[0002] In today's era of rapidly developing automation technology, logistics and operations in closed environments such as mines and ports are rapidly moving towards intelligent and automated operations. Autonomous vehicles, with their advantages of high efficiency, stability, and long-term continuous operation, are gaining increasing popularity in these closed environments.

[0003] Currently, in closed environments like mines and ports, existing vehicle planning methods are mostly based on pre-set fixed rules and simple path planning algorithms. These methods typically pre-set basic information such as the vehicle's route and stops, and then generate the vehicle's route using static map data within the scene and the vehicle's initial state data. Existing methods lack the ability to effectively integrate human guidance. In actual operations, operators may need to make some manual interventions and adjustments, but existing planning methods struggle to quickly respond to and accurately execute these manual guidance, failing to meet the complex and ever-changing operational requirements of closed environments, resulting in low vehicle operation efficiency. Summary of the Invention

[0004] The present invention provides a vehicle planning method, device, equipment and storage medium based on manual guidance, so as to solve the problem of low vehicle operation efficiency due to lack of manual guidance in the prior art.

[0005] The first aspect of the present invention provides a vehicle planning method based on manual guidance, including: obtaining waypoint data corresponding to at least one waypoint manually marked on a supervision platform; based on at least one waypoint data, combined with map data, real-time road condition data and vehicle status data, using a spatiotemporal data fusion algorithm to generate constraint nodes with target priorities; performing hybrid path planning processing according to the constraint nodes to obtain an initial path; based on the initial path, performing differentiated adjustments to must-reach points and avoidance points, and performing local replanning through a dynamic optimization algorithm to obtain an optimized path; based on the optimized path, generating control instructions executable by the target autonomous driving vehicle to control the operation of the target autonomous driving vehicle.

[0006] In a feasible embodiment, each waypoint data includes the geographic coordinates, initial priority and traffic constraint attributes of the waypoint, and the generating of a constraint node with priority based on at least one waypoint data in combination with map data, real-time traffic data and vehicle status data includes: reading map data related to the area where the at least one waypoint is located from a preset map database, and obtaining real-time traffic data of the area surrounding the at least one waypoint and vehicle status data of the target autonomous driving vehicle; preprocessing the map data using a spatiotemporal data fusion algorithm to construct a spatiotemporal map model; based on at least one waypoint data, the real-time traffic data, the vehicle status data and the spatiotemporal map model, using a spatiotemporal graph neural network ST-GNN to perform data fusion analysis, and extract the influence parameters of road characteristics, road condition influencing factors and the current status of the vehicle on the accessibility of each waypoint respectively; assigning priority weights to the nodes corresponding to each waypoint according to the initial priority and corresponding influencing parameters in the waypoint data, and generating a constraint node with priority.

[0007] In a feasible embodiment, the map data is preprocessed using a spatiotemporal data fusion algorithm to construct a spatiotemporal map model, including: extracting node information and edge information of roads from the map data, wherein the node information includes the location of road intersections, and the edge information includes the connection relationship and length of roads; based on a preset time period division rule, a day is divided into multiple time periods, and historical traffic data of roads in each time period is obtained; based on the node information, edge information of the road and the historical communication data of each time period, a spatiotemporal map model is constructed using a spatiotemporal data fusion algorithm, wherein the spatiotemporal map model includes a road topology structure, and the road topology structure includes multi-time period dynamic characteristics of the road.

[0008] In a feasible implementation, the data fusion analysis is performed using a spatiotemporal graph neural network ST-GNN based on at least one waypoint data, the real-time road condition data, the vehicle status data and the spatiotemporal map model to extract the road characteristics, road condition influencing factors and the impact parameters of the vehicle's current status on the accessibility of each waypoint, including: integrating the at least one waypoint data, the real-time road condition data, the vehicle status data and the relevant data in the spatiotemporal map model to construct an input data set; inputting the input data set into the spatiotemporal graph neural network ST-GNN, processing the input data set layer by layer through its multi-layer network structure to obtain an output structure; and extracting the road characteristics, road condition influencing factors and the impact parameters of the vehicle's current status on the accessibility of each waypoint based on the output structure.

[0009] In a feasible implementation, priority weights are assigned to nodes corresponding to each waypoint based on the priority and influencing parameters of each waypoint, and constraint nodes with target priorities are generated to determine initial weight values ​​corresponding to different initial priorities. The influencing parameters of each waypoint are analyzed, and for unfavorable factors affecting the accessibility of the waypoints, the initial weight values ​​of the corresponding waypoints are adjusted downward based on the degree of their impact on the accessibility. For favorable factors, the initial weight values ​​of the corresponding waypoints are adjusted upward based on the degree of their improvement on the accessibility. The adjusted weights are assigned to nodes corresponding to each waypoint, and constraint nodes with target priorities are generated, where a higher priority weight indicates a higher importance of the node in path planning.

[0010] In a feasible embodiment, the hybrid path planning processing is performed according to the constraint nodes to obtain an initial path, including: utilizing a graph search-based path planning algorithm, taking the current position of the target autonomous driving vehicle as the starting point, and constructing a search graph in combination with the constraint nodes; based on the search graph, performing a path search according to the target priority and traffic constraint attributes of each constraint node; when a target point that meets the conditions of all constraint nodes is found, determining the initial path based on the nodes passed during the search process.

[0011] In a feasible implementation, the must-reach points and avoidance points are differentially adjusted based on the initial path, and are partially replanned through a dynamic optimization algorithm to obtain an optimized path, including: identifying the must-reach points and avoidance points in the initial path, the must-reach points are manually marked mandatory waypoints, and the avoidance points are manually marked prohibited areas; based on the target priority of the must-reach points and the current status of the vehicle, the arrival time window of the must-reach points is dynamically adjusted, and the avoidance range of the avoidance points is expanded or reduced according to the real-time road conditions; using a dynamic optimization algorithm, the initial path is partially replanned with the adjusted time window of the must-reach points and the avoidance range of the avoidance points as constraints to obtain an optimized path.

[0012] In a feasible implementation, the arrival time window of the must-reach point is dynamically adjusted based on the target priority of the must-reach point and the current status of the vehicle, and the detour range of the detour point is expanded or reduced according to the real-time road conditions, including: for the first must-reach point with high priority, if the distance from the current position of the vehicle to the first must-reach point is greater than the preset distance and the expected arrival time is greater than the preset time range, the arrival time window is extended backward; if the current distance from the vehicle to the first must-reach point is not greater than the preset distance and the expected arrival time is less than the preset time range, the arrival time window is compressed forward; for the second must-reach point with low priority, the arrival time window is dynamically relaxed or tightened according to the current driving progress of the vehicle and the remaining path length; for each detour point, when the real-time road conditions show that the traffic around the avoidance point is congested, its detour range is expanded; when the real-time road conditions show that the traffic around the avoidance point is smooth, its detour range is reduced.

[0013] In a feasible implementation, the dynamic optimization algorithm is used to locally replan the initial path with the adjusted time window of the must-reach point and the avoidance range of the avoidance point as constraints to obtain an optimized path, including: using the adjusted time window of the must-reach point and the avoidance range of the avoidance point as constraints of the model, taking the shortest driving time as the main optimization goal, and taking the shortest path length as the secondary optimization goal to construct a dynamic optimization model; using the dynamic optimization algorithm to solve the dynamic optimization model, and finding a path plan that optimizes the optimization goal while satisfying the constraints; based on the solved path plan, locally adjust the initial path to generate an optimized path, and the optimized path meets the arrival requirements of the must-reach point and the avoidance requirements of the avoidance point.

[0014] The second aspect of the present invention provides a vehicle planning device based on manual guidance, including: an acquisition module for acquiring waypoint data corresponding to at least one waypoint manually marked on a supervision platform; a generation module for generating constraint nodes with target priorities based on at least one waypoint data, combined with map data, real-time road condition data and vehicle status data, using a spatiotemporal data fusion algorithm; a processing module for performing hybrid path planning processing according to the constraint nodes to obtain an initial path; an optimization module for performing differentiated adjustments to must-reach points and avoidance points based on the initial path, and performing local replanning through a dynamic optimization algorithm to obtain an optimized path; a control module for generating control instructions executable by a target autonomous driving vehicle based on the optimized path to control the target autonomous driving vehicle to operate.

[0015] In a feasible embodiment, the generation module includes: an acquisition unit, which is used to read map data related to the area where the at least one waypoint is located from a preset map database, and obtain real-time road condition data of the area around the at least one waypoint and vehicle status data of the target autonomous driving vehicle; a construction unit, which is used to pre-process the map data using a spatiotemporal data fusion algorithm to construct a spatiotemporal map model; an extraction unit, which is used to extract data based on at least one waypoint data, the real-time road condition data, the vehicle status data and the spatiotemporal map model, and use a spatiotemporal graph neural network ST-GNN to perform data fusion analysis to extract the influence parameters of road characteristics, road condition influencing factors and the current status of the vehicle on the accessibility of each waypoint; a generation unit, which is used to assign priority weights to the nodes corresponding to each waypoint according to the initial priority and corresponding influencing parameters in the data of each waypoint, and generate constraint nodes with priorities.

[0016] In a feasible embodiment, the construction unit is specifically used to: extract node information and edge information of roads from the map data, wherein the node information includes the location of road intersections, and the edge information includes the connection relationship and length of roads; based on a preset time period division rule, divide a day into multiple time periods, and obtain historical traffic data of roads in each time period; based on the node information, edge information and historical communication data of each time period of the road, use a spatiotemporal data fusion algorithm to construct a spatiotemporal map model, wherein the spatiotemporal map model includes a road topology structure, and the road topology structure includes multi-time period dynamic characteristics of the road.

[0017] In a feasible embodiment, the extraction unit is specifically used to: integrate the at least one waypoint data, the real-time road condition data, the vehicle status data and the relevant data in the spatiotemporal map model to construct an input data set; input the input data set into the spatiotemporal graph neural network ST-GNN, and process the input data set layer by layer through its multi-layer network structure to obtain an output structure; based on the output structure, respectively extract the road characteristics, road condition influencing factors and the impact parameters of the vehicle's current status on the accessibility of each waypoint. In a feasible embodiment, the generation unit is specifically used to: determine the initial weight values ​​corresponding to different initial priorities; analyze the influencing parameters of each waypoint, and for unfavorable factors affecting the accessibility of the waypoint, reduce the initial weight value of the corresponding waypoint according to the degree of its influence on the accessibility; for favorable factors, increase the initial weight value of the corresponding waypoint according to the degree of its improvement on the accessibility; assign the adjusted weight to the node corresponding to each waypoint, and generate a constraint node with a target priority, where the higher the priority weight, the higher the importance of the node in path planning. In a feasible embodiment, the processing module is specifically used to: utilize a graph search-based path planning algorithm, take the current position of the target autonomous driving vehicle as the starting point, and construct a search graph in combination with the constraint nodes; based on the search graph, perform a path search according to the target priority and traffic constraint attributes of each constraint node; when a target point that meets the conditions of all constraint nodes is searched, determine an initial path based on the nodes passed during the search process.

[0018] In a feasible embodiment, the optimization module includes: an identification unit for identifying the must-reach points and avoidance points in the initial path, the must-reach points are manually marked mandatory waypoints, and the avoidance points are manually marked prohibited areas; an adjustment unit for dynamically adjusting the arrival time window of the must-reach points based on the target priority of the must-reach points and the current status of the vehicle, and expanding or reducing the avoidance range of the avoidance points according to real-time road conditions; a replanning unit for using a dynamic optimization algorithm, with the adjusted time window of the must-reach points and the avoidance range of the avoidance points as constraints, to locally replan the initial path to obtain an optimized path.

[0019] In a feasible embodiment, the adjustment unit is specifically used to: for the first must-reach point with high priority, if the distance from the current position of the vehicle to the first must-reach point is greater than the preset distance and the expected arrival time is greater than the preset time range, the arrival time window will be extended backward; if the current distance from the vehicle to the first must-reach point is not greater than the preset distance and the expected arrival time is less than the preset time range, the arrival time window will be compressed forward; for the second must-reach point with low priority, the arrival time window will be dynamically relaxed or tightened according to the current driving progress of the vehicle and the remaining path length; for each avoidance point, when the real-time road conditions show that the traffic around the avoidance point is congested, the avoidance range will be expanded; when the real-time road conditions show that the traffic around the avoidance point is smooth, the avoidance range will be reduced.

[0020] In a feasible implementation, the re-planning unit is specifically used to: use the adjusted time window of the must-reach point and the avoidance range of the avoidance point as the constraints of the model, with the shortest driving time as the main optimization goal and the shortest path length as the secondary optimization goal, so as to construct a dynamic optimization model; use a dynamic optimization algorithm to solve the dynamic optimization model, and find a path plan that optimizes the optimization goal while satisfying the constraints; according to the solved path plan, locally adjust the initial path to generate an optimized path, and the optimized path meets the arrival requirements of the must-reach point and the avoidance requirements of the avoidance point.

[0021] A third aspect of the present invention provides an electronic device comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the electronic device executes the above-mentioned vehicle planning method based on manual guidance.

[0022] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when executed on a computer, enable the computer to execute the above-mentioned vehicle planning method based on manual guidance.

[0023] In the technical solution provided by the present invention, waypoint data corresponding to at least one waypoint manually marked on the supervision platform is obtained; based on the at least one waypoint data, a spatiotemporal data fusion algorithm is used to generate constraint nodes with target priorities in combination with map data, real-time road condition data, and vehicle status data; hybrid path planning is performed based on the constraint nodes to obtain an initial path; based on the initial path, differential adjustments are made to the required points and avoidance points, and local replanning is performed using a dynamic optimization algorithm to obtain an optimized path; and based on the optimized path, control instructions executable by the target autonomous vehicle are generated to control the operation of the target autonomous vehicle. In an embodiment of the present invention, by obtaining manually marked waypoint data and integrating it into path planning, it is possible to quickly respond to and accurately execute manual guidance. By combining multi-source data with a spatiotemporal data fusion algorithm to generate constraint nodes and perform hybrid path planning, and then through differential adjustment and local replanning using a dynamic optimization algorithm, it can flexibly adapt to complex and changing operational requirements, plan a more accurate and reasonable path, and ultimately generate executable control instructions, effectively improving the operational efficiency of the target autonomous vehicle. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 Schematic diagram of an embodiment of a vehicle planning method based on manual guidance in an embodiment of the present invention; Figure 2 Schematic diagram of another embodiment of a vehicle planning method based on manual guidance in an embodiment of the present invention; Figure 3 Schematic diagram of another embodiment of a vehicle planning method based on manual guidance in an embodiment of the present invention; Figure 4 Schematic diagram of an embodiment of a vehicle planning device based on manual guidance in an embodiment of the present invention; Figure 5 Schematic diagram of another embodiment of a vehicle planning device based on manual guidance in an embodiment of the present invention; Figure 6 FIG. 1 is a schematic diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0025] Embodiments of the present invention provide a vehicle planning method, apparatus, device, and storage medium based on manual guidance. By acquiring waypoint data manually marked on a supervision platform and effectively integrating it into route planning, vehicle planning can quickly respond to and accurately execute manual intervention adjustments based on actual operating conditions, thereby improving vehicle operating efficiency.

[0026] The terms "first," "second," "third," "fourth," and so on (if any) in the description and claims of the present invention and in the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that shown or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus that includes a series of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, product, or apparatus.

[0027] It is understandable that the execution subject of the present invention may be a vehicle planning device based on manual guidance, or a terminal or a server, which is not limited here. The embodiment of the present invention is described by taking the server as the execution subject as an example.

[0028] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 In one embodiment of the present invention, a vehicle planning method based on manual guidance includes: 101. Obtain the waypoint data corresponding to at least one waypoint manually marked on the supervision platform; Each waypoint can be accompanied by additional attributes, such as dwell time, initial priority, etc. The supervision platform provides an interactive map interface, where operators can mark key points on the map by clicking or dragging the mouse and set relevant parameters. The system will store the geographic coordinates, types, additional attributes and other data of these waypoints in a structured manner, such as using JSON or Protobuf format. At the same time, the system will verify the validity of the input data, such as whether the coordinates are within a reasonable range, whether the radius of the avoidance point is too large, etc., to avoid path planning anomalies due to incorrect input.

[0029] For example, the staff operated on the mine map through the user interaction interface of the supervision platform. The staff marked three transit points, namely: Point A (temporary ore storage area), which was set to high priority and was a must-reach point; Point B (a dangerous section with a slope of >15°), which was marked as an avoidance point and set to low priority; Point C (maintenance station), which was set to medium priority and required a 10-minute stop for equipment inspection.

[0030] 102. Based on at least one waypoint data, combined with map data, real-time traffic data, and vehicle status data, a spatiotemporal data fusion algorithm is used to generate a constraint node with a target priority; The method comprises the following steps: obtaining original map data of a corresponding area from a preset map database based on the geographic coordinate information of at least one waypoint; parsing the original map data to extract road elements therein, and reorganizing the road elements according to a preset data structure; constructing roads based on the reorganized road element data to obtain road network topology information.

[0031] Real-time traffic condition data of an area surrounding at least one waypoint is obtained, and traffic condition information of each road section is extracted from the real-time traffic condition data; based on a preset road condition and traffic capacity impact relationship model, the traffic condition information of each road section is used to calculate a road condition impact factor for each road section, where the road condition impact factor reflects the degree to which the road condition reduces the road traffic capacity.

[0032] The vehicle status data of the target autonomous vehicle is obtained and combined with at least one waypoint data, road network topology information, and a road condition influencing factor, and then a constraint node with a target priority is generated according to a preset priority algorithm. Specifically, a speed sensor, a power sensor, a load sensor, and other devices can be used to collect vehicle status data such as the speed, remaining power, and load of the target autonomous vehicle in real time. The vehicle status data, at least one waypoint data, road network topology information, and a road condition influencing factor calculated based on the real-time road condition data are then input into a preset priority algorithm model. The model first uses the preset initial priority of the waypoints as a basis and comprehensively considers the vehicle status. For example, when the vehicle battery is low, priority is given to waypoints that are required and close to the vehicle. At the same time, the model, in combination with the road condition influencing factor, avoids road sections around low-priority waypoints with poor road conditions that are not required to be reached. The priority of each waypoint is re-evaluated through a weighted calculation method, and ultimately generates a constraint node with a target priority.

[0033] For example, based on the coordinates of the marked waypoints A as a temporary ore storage area, B as a dangerous road section, and C as a maintenance station, the road data of the corresponding area is extracted from the mining area map database, and these data are analyzed to construct a topological network containing key attributes such as road slope and width, and obtain real-time road condition information. It is found that due to the dense ore transport vehicles around point A, the congestion index reaches 0.8. According to the road condition assessment model, it is calculated that the traffic capacity of this section has decreased by 60%; combined with the current status of the mine car, its load is 35 tons and the power is 28%. The priority algorithm begins to dynamically adjust, raising the priority of the must-reach point A from the preset 8 to 9. Due to limited power, this key task needs to be completed first; avoid point B The priority level 2 is maintained unchanged, and the avoidance radius is determined to be 50 meters; the priority of maintenance station C is reduced from 5 to 4, and a new constraint condition is added that it must be reached when the power level is lower than 30%; finally, three constraint nodes are output, namely point A, whose coordinates are (x1, y1), the target priority is 9 and it is a must-reach point; point B, whose coordinates are (x2, y2), the target priority is 2, and the avoidance radius is 50 meters; point C, whose coordinates are (x3, y3), the target priority is 4, and it needs to stay for 10 minutes and meet the power trigger condition.

[0034] 103. Perform hybrid path planning processing according to the constraint nodes to obtain an initial path; Based on the constraint nodes, a graph search algorithm, such as the A* algorithm or the Dijkstra algorithm, is used. The appropriate algorithm is selected based on the actual scenario requirements. If dynamic factors such as real-time traffic conditions need to be considered, a modified A* algorithm can be used. A search graph is constructed based on road network topology information, with constraint nodes serving as key nodes in the graph. The weights of each road segment in the search graph are dynamically adjusted based on real-time traffic data. For example, a higher road condition impact factor indicates a higher weight for the segment, indicating a higher travel cost. Vehicle status data, such as the impact of load and battery level on vehicle driving capacity, is also incorporated as an additional constraint. These factors are comprehensively considered during the search process. The algorithm iteratively searches for the optimal path from the starting point to the end point that passes through each constraint node and meets the vehicle driving capacity and real-time traffic requirements. This path is then used as the initial path. The algorithm traverses the paths based on target priority.

[0035] For example, based on the A, B, and C constraint nodes, an improved A* algorithm is used for path planning. Specifically, a search graph is constructed based on the road network topology information of the mining area, with points A, B, and C as key nodes in the graph. Combined with real-time traffic data, the weights of each road section in the search graph are dynamically adjusted. Due to the congestion of the section near point A and the large road condition impact factor, the weight of this section increases, indicating a high travel cost. At the same time, the vehicle status data of the mine car, such as load and battery power, is taken into account. The load affects the vehicle's acceleration and braking performance, while the battery power limits the mileage. These factors are incorporated into the search process as additional constraints. The algorithm traverses based on the target priority and, starting from the current position of the mine car, searches for an optimal path that passes through point A, avoids point B, passes through point C, and meets the vehicle's driving capacity and real-time road conditions. This path is used as the initial path.

[0036] 104. Based on the initial path, differentiated adjustments are made to the must-reach points and avoidance points, and local replanning is performed using a dynamic optimization algorithm to obtain the optimized path. For must-reach points, the initial path is checked to see if it ensures that the vehicle can arrive accurately. If it is found that a must-reach point is not reasonably included in the path, for example, the must-reach point is bypassed during the path planning process to avoid a severely congested road section, the path will be readjusted. The adjustment method includes re-searching for feasible paths within a certain range around the must-reach point, taking into account the real-time road conditions and road topology to ensure that the vehicle can reach the must-reach point in the shortest time and at the lowest cost. For avoidance points, the system will continuously monitor the distance between the vehicle and the avoidance point during driving. If the vehicle tends to approach the avoidance point due to changes in road conditions or other reasons, the dynamic optimization algorithm will immediately initiate local replanning. The algorithm will use the current position of the vehicle as the new starting point, and combine the remaining constraint nodes, real-time road conditions and vehicle status to replan a local path while avoiding the avoidance point. During the local replanning process, the continuity and smoothness of the path will be fully considered to avoid vehicle instability due to sudden changes in the path. At the same time, the dynamic optimization algorithm will evaluate the advantages and disadvantages of different path plans in real time, select the optimal local path plan through continuous iterative optimization, and gradually correct and improve the initial path, ultimately obtaining an optimized path that meets the requirements of reaching the must-reach point and effectively avoids detour points.

[0037] For example, when it is found that due to changes in real-time road conditions, the congestion of the road section originally planned to pass through point A has intensified, and continuing to drive along the original route may result in failure to reach point A on time, adjustments must be made to the must-reach point A; a feasible path is re-searched within a certain range around point A, and the updated real-time road conditions and road topology are comprehensively considered to ensure that the mine car can reach point A in the shortest time and at the lowest cost; at the same time, it is monitored that the direction of the mine car is approaching the avoidance point B because the road ahead is partially closed due to construction, and the originally planned path is affected; at this time, the dynamic optimization algorithm immediately starts local replanning, taking the current position of the mine car as the new starting point, and combining the remaining constraint nodes, such as points A and C, real-time road conditions such as avoiding congestion and construction sections, and vehicle status, such as 28% battery power and full load, to replan a local path on the premise of avoiding point B; during the planning process, the continuity and smoothness of the path are taken into consideration to avoid unstable driving conditions such as sudden braking or sharp turns caused by sudden changes in the path of the mine car. Through continuous iterative optimization, the optimal local path solution is selected, and the initial path is corrected and improved. Ultimately, an optimized path is obtained that ensures accurate arrival at the required point A, passing through point C for maintenance at the appropriate time, and effectively avoiding the detour point B.

[0038] 105. Generate control instructions executable by the target autonomous driving vehicle based on the optimized path to control the target autonomous driving vehicle to operate.

[0039] The optimized path is discretized, dividing the continuous path curve into multiple path points at regular time intervals or distance intervals. Each path point contains key information such as the vehicle's desired geographic coordinates, speed, acceleration, and steering angle at that location. Based on the vehicle's dynamic and kinematic models and incorporating current vehicle state data, such as speed, steering angle, and battery charge, the control parameters corresponding to each path point are further optimized and adjusted. For example, the speed parameter at each path point is appropriately limited based on the vehicle's maximum acceleration and braking capabilities to ensure that excessive speed changes during driving do not affect driving safety and comfort. The optimized path point information is then encapsulated into control commands according to specific communication protocols and interface specifications. These control commands are transmitted in real time to the vehicle's various control modules, such as the engine control module, brake control module, and steering control module, via an onboard communication system, such as the CAN bus or Ethernet. Upon receiving the control commands, each control module controls the vehicle's driving state based on the parameter information contained in the commands, ensuring that the vehicle accurately follows the optimized path.

[0040] In an embodiment of the present invention, by acquiring manually annotated waypoint data and integrating it into path planning, it can quickly respond to and accurately execute manual guidance. By combining multi-source data with a spatiotemporal data fusion algorithm to generate constraint nodes and perform hybrid path planning, and then undergoing differentiated adjustment and local replanning using a dynamic optimization algorithm, it can flexibly adapt to complex and changing operational requirements, plan a more accurate and reasonable path, and ultimately generate executable control instructions, effectively improving the operating efficiency of the target autonomous driving vehicle.

[0041] See also Figure 2 Another embodiment of the vehicle planning method based on manual guidance in the embodiment of the present invention includes: 201. Obtaining waypoint data corresponding to at least one waypoint manually marked on the supervision platform; 202. Reading map data related to an area where at least one waypoint is located from a preset map database, and obtaining real-time traffic data of an area surrounding the at least one waypoint and vehicle status data of the target autonomous driving vehicle; Map data provides basic information about regional roads, real-time traffic data reflects current road conditions, and vehicle status data includes the vehicle's own operating conditions.

[0042] 203. Preprocess the map data using spatiotemporal data fusion algorithm to construct a spatiotemporal map model; Node information and edge information of roads are extracted from map data, where node information includes the location of road intersections, and edge information includes the connection relationship and length of roads. Based on the preset time period division rules, a day is divided into multiple time periods, and the historical traffic data of roads in each time period is obtained. Based on the node information, edge information of the road and the historical communication data of each time period, a spatiotemporal map model is constructed using a spatiotemporal data fusion algorithm. The spatiotemporal map model includes the road topology structure, and the road topology structure includes the dynamic characteristics of the road in multiple time periods. Vector analysis tools are used to identify and extract the road layer in the map, converting the roads into vector lines composed of a series of discrete points. For the road intersection locations in the node information, the spatial analysis function of GIS is used to detect the intersection points of the vector lines. Combined with the coordinate system of the map, the longitude and latitude coordinates of each intersection in the real geographic space are determined, and the intersections are classified and recorded according to their shape characteristics. When extracting edge information, the distance measurement tool of GIS is used to calculate the actual length between adjacent nodes along the center line of the road for the acquired vector road lines to ensure the accuracy of the length data. Graph theory algorithms are used to treat each node as a vertex in the graph and the road lines as edges. By analyzing the connection between the vertices, a road connection relationship matrix is ​​constructed to present the connection status of each road with other roads, thereby completely extracting the node information and edge information of the road.

[0043] Based on pre-set time period division rules, a day can be divided into multiple time periods. Specifically, based on the cyclical patterns of traffic flow and actual demand, a day can be initially divided into 24 time periods, with each hour as the basic unit. Furthermore, for special periods with drastic traffic flow fluctuations, such as morning and evening rush hours, this can be further subdivided into 15-minute or 30-minute time periods. To obtain historical road traffic data for each time period, a legal, secure, and stable interface can be established with the traffic management department's data system to retrieve key data such as vehicle flow statistics, average driving speeds, and the number and duration of congestion on each road during the specified time period from its massive database. Furthermore, intelligent traffic sensors deployed at key road nodes, such as important intersections and bridges, can be used to collect real-time information such as license plate information and transit times of passing vehicles. By filtering, organizing, and analyzing this raw data by time period, historical communication data such as the actual number of vehicles passing through the road and traffic density during each time period can be obtained.

[0044] The initial static topological structure of the road is constructed with the node information of the road as the vertex and the edge information as the edge, which is expressed in the form of a graph, where the nodes record the coordinates of key locations such as road intersections, and the edges record the road connection relationship and fixed length; for the historical traffic data of each time period, cluster analysis in data mining is used to classify the data of different time periods according to characteristics such as traffic flow and average speed, and the typical traffic mode of the road in each time period is extracted, such as the congestion mode during peak hours and the smooth mode during off-peak hours; the spatiotemporal interpolation algorithm is used to interpolate the traffic patterns between adjacent time periods according to the time intervals and traffic mode changes of different time periods. Data interpolation is performed to fill in data gaps and make the traffic data continuous and smooth in the time dimension. Afterwards, the processed traffic data for each time period is integrated with the static topological structure. By assigning dynamic attributes of different time periods to each edge, such as travel time and congestion probability in different time periods, the static road topological structure is expanded into a spatiotemporal topological structure that includes dynamic characteristics of multiple time periods. Finally, three-dimensional visualization technology is used to display the road topology in a two-dimensional plane, and visual elements such as color and texture are used to present the dynamic changes of the road in different time periods in the time dimension, thus constructing an intuitive and information-rich spatiotemporal map model.

[0045] Existing technologies often rely solely on static map data for route planning, failing to reflect the dynamic characteristics of roads over time, resulting in a disconnect between planned routes and actual traffic conditions. This solution, however, accurately extracts road node and edge information from map data to construct a basic road topology framework. This solution then rationally divides time periods and obtains historical traffic data for each period to fully understand the traffic patterns of roads at different times. Using a spatiotemporal data fusion algorithm, it deeply integrates static road structure with dynamic traffic data to construct a spatiotemporal map model that incorporates the dynamic characteristics of roads across multiple time periods. This allows route planning to fully consider the actual traffic capacity of roads at different times, significantly improving the accuracy, rationality, and real-time performance of route planning.

[0046] 204. Based on at least one waypoint data, real-time road condition data, vehicle status data, and a spatiotemporal map model, a spatiotemporal graph neural network (ST-GNN) is used to perform data fusion analysis to extract parameters affecting the accessibility of each waypoint, including road characteristics, road condition influencing factors, and vehicle current status; The input dataset is constructed by integrating at least one waypoint data, real-time road condition data, vehicle status data and relevant data in the spatiotemporal map model. The input dataset is input into the spatiotemporal graph neural network ST-GNN, and the input dataset is processed layer by layer through its multi-layer network structure to obtain an output structure. Based on the output structure, the road characteristics, road condition influencing factors and the impact parameters of the vehicle's current status on the accessibility of each waypoint are extracted respectively.

[0047] Preprocess the data of at least one waypoint obtained and convert it into a unified coordinate format; for real-time traffic data, parse out key indicators such as the congestion level, average vehicle speed, and accident occurrence of each road section; extract vehicle status data such as the vehicle's current position, speed, remaining power / fuel, and driving direction; extract data such as the road topology structure and road traffic patterns in different time periods related to the area where at least one waypoint is located from the spatiotemporal map model, and splice and associate these data from different sources according to a specific data structure to construct a structured input data set to ensure a clear correspondence between each data element.

[0048] The constructed input dataset is fed into a pre-trained spatiotemporal graph neural network (ST-GNN). This network has a multi-layered architecture consisting of an input layer, multiple hidden layers, and an output layer. After receiving the input dataset, the input layer allocates data such as waypoint coordinates, real-time road condition indicators, vehicle status information, road topology, and traffic patterns to individual neurons based on their functional characteristics. The hidden layers then take over the data assigned by the input layer and perform deep processing using nonlinear transformations. Using spatiotemporal convolution, the input data combines spatial relationships between different road sections and the relative positions of waypoints and roads, as well as temporal variations in real-time road conditions and traffic patterns over time. This captures correlations in both spatial and temporal dimensions, uncovering underlying spatiotemporal patterns and regularities. Furthermore, through an attention mechanism, different weights are assigned to each data feature based on its importance to the accessibility of each waypoint, highlighting key features and allowing the network to focus on factors that have the greatest impact on accessibility. After processing in the hidden layers, the data is transformed into a more representative and discriminative feature representation. Finally, the output layer integrates and outputs the feature data processed by the hidden layer to form an output structure. Based on this output structure, a specific analytical algorithm is used to extract the impact parameters of road characteristics, road condition influencing factors, and the vehicle's current state on the accessibility of each waypoint. The impact parameters of road characteristics include curvature, slope, and width. The impact parameters of road condition influencing factors include congestion level and accident status. The impact parameters of the vehicle's current state include remaining battery / fuel level and current speed.

[0049] Compared with traditional solutions that rely solely on a single data source or simply stack different data, this solution deeply integrates waypoints, real-time road conditions, vehicle status data, and spatiotemporal map model-related data to construct a structured input data set, allowing different types of data to complement and verify each other. It can comprehensively characterize traffic scenarios from multiple dimensions such as spatial location, road characteristics, real-time traffic, and vehicle status. It then uses the multi-layer structure of the spatiotemporal graph neural network (ST-GNN) to reasonably allocate data, capture spatiotemporal patterns, highlight key features, and explore the inherent correlations of the data. Based on the output structure, it accurately extracts the influencing parameters of each factor on the accessibility of the waypoints, greatly improving the accuracy and comprehensiveness of the analysis results.

[0050] 205. Assign priority weights to nodes corresponding to each waypoint based on the initial priority and corresponding influencing parameters in the waypoint data, and generate constraint nodes with target priorities. Determine the initial weight values ​​corresponding to different initial priorities; analyze the influencing parameters of each waypoint; for unfavorable factors affecting the accessibility of the waypoint, reduce the initial weight value of the corresponding waypoint according to the degree of its impact on accessibility; for favorable factors, increase the initial weight value of the corresponding waypoint according to the degree of its improvement on accessibility; assign the adjusted weight to the node corresponding to each waypoint to generate a constraint node with the target priority, where the higher the priority weight, the more important the node in path planning.

[0051] Map the initial priority to the initial weight value, for example, map high, medium and low priorities to 0.8, 0.5 and 0.2; analyze the influencing parameters of each waypoint. In terms of road characteristics, if the curvature of the road where the waypoint is located is too large and the slope is extremely steep, these unfavorable factors will seriously hinder vehicle traffic. According to the degree to which the vehicle speed is reduced and the energy consumption is increased, the initial weight value corresponding to the waypoint is reduced by a certain proportion; conversely, if the road is flat and wide, it is a favorable factor. According to the degree to which it improves vehicle driving efficiency and safety, the initial weight value is increased in a corresponding proportion; for road condition influencing factors, if the road sections around the waypoint are severely congested and accidents occur frequently, the initial weight value is reduced according to the duration of congestion and the impact of the accident impact range on accessibility; if the road condition is unobstructed, the initial weight value is increased according to the improved traffic efficiency. Regarding the vehicle's current status, if the vehicle has sufficient remaining battery power / fuel to ensure smooth arrival at the waypoint, the initial weight value is increased according to the increased driving security level; if the battery power / fuel level is insufficient, the initial weight value is reduced according to the risk level that may affect arrival; finally, the adjusted weight is assigned to the node corresponding to each waypoint to generate a constraint node with target priority.

[0052] Compared with the existing solutions with fixed weights or simple hierarchical priorities, this solution dynamically quantifies and analyzes multi-dimensional influencing parameters such as road characteristics, real-time road conditions, and vehicle status, and establishes a coupling mechanism between initial priority and dynamic adjustment factors, so that the weights of waypoints can adapt to changes in actual traffic scenarios. In tests, it significantly improved the arrival rate of high-priority nodes and the rationality of path planning.

[0053] 206. Perform hybrid path planning processing according to the constraint nodes to obtain an initial path; A graph-search-based path planning algorithm is used. The current position of the target autonomous driving vehicle is used as the starting point, and a search graph is constructed in combination with constraint nodes. Based on the search graph, a path search is performed according to the target priority and access constraint attributes of each constraint node. When the target point that meets the conditions of all constraint nodes is found, the initial path is determined based on the nodes passed during the search process.

[0054] Taking the current position of the target autonomous vehicle as the starting point for path planning, the constraint nodes are integrated into the graph structure, and a complete search graph is constructed based on the actual road connectivity. Each edge in the graph represents a drivable road and is assigned a corresponding travel cost, such as distance, estimated travel time, etc. At the same time, the travel constraint attributes of each constraint node are clarified, such as whether U-turns are allowed and whether it is drivable during a specific time period. A graph-search-based path planning algorithm is used. During the search process, the target priority of each constraint node is considered. Nodes with higher priorities are assigned higher exploration weights during the search. At the same time, combined with the travel cost of the edges, a comprehensive evaluation value of each node is dynamically calculated to guide the search direction. The search range is gradually expanded according to this evaluation mechanism, and possible paths are continuously explored. When a target point that meets all constraint node conditions is found, the sequence of nodes passed through during the search process is backtracked and these nodes are connected in the search order to determine the initial path. Constraint node conditions include having to pass through specific priority nodes and meeting the travel constraints of each node.

[0055] Existing path planning technologies often fail to fully consider the priority differences of key nodes in actual traffic scenarios and the complex and changeable traffic constraints, resulting in the planned paths possibly failing to meet actual driving needs and causing problems with safety and efficiency. This solution, however, utilizes a graph-search-based path planning algorithm to construct a search graph incorporating constraint nodes starting from the current position of the autonomous driving vehicle. This approach accurately fits the actual environment, searches for paths based on node priorities and traffic constraint attributes, and rationally weighs key factors to determine an initial path that meets all conditions, effectively improving the accuracy, rationality, and practicality of autonomous driving path planning.

[0056] 207. Based on the initial path, differentiated adjustments are made to the must-reach points and avoidance points, and local replanning is performed using a dynamic optimization algorithm to obtain the optimized path; 208. Generate control instructions executable by the target autonomous driving vehicle based on the optimized path to control the target autonomous driving vehicle to operate.

[0057] In an embodiment of the present invention, manually annotated waypoint data is obtained and integrated with multi-source data such as maps, real-time road conditions, and vehicle status. A spatiotemporal map model is constructed using a spatiotemporal data fusion algorithm. The parameters affecting the accessibility of each waypoint are accurately extracted with the help of a spatiotemporal graph neural network (ST-GNN), and constraint nodes with reasonable priorities are generated. This allows path planning to fully consider multiple key factors in actual operations. Through hybrid path planning, differentiated adjustment, and local replanning using a dynamic optimization algorithm, it can flexibly respond to complex and changing operating environments, continuously optimize the path, and ultimately generate executable control instructions. This effectively improves the accuracy and rationality of the target autonomous driving vehicle's path planning, enhances its adaptability to different scenarios, and thereby improves operational efficiency and safety.

[0058] See also Figure 3 Another embodiment of the vehicle planning method based on manual guidance in the embodiment of the present invention includes: 301. Obtaining waypoint data corresponding to at least one waypoint manually marked on the supervision platform; 302. Based on at least one waypoint data, combined with map data, real-time traffic data, and vehicle status data, a spatiotemporal data fusion algorithm is used to generate a constraint node with a target priority; 303. Perform hybrid path planning processing according to the constraint nodes to obtain an initial path; 304. Identify the must-reach points and avoidance points in the initial path. The must-reach points are manually marked mandatory waypoints, and the avoidance points are manually marked prohibited areas. The information related to the manually marked waypoints is read from the storage structure to clarify the respective identification features of the must-reach points and the avoidance points. For example, the must-reach points have specific mandatory waypoint signs, and the avoidance points have no-entry area signs; the initial path is converted into a sequence consisting of a series of continuous coordinate points, and the coordinate matching algorithm is used to compare the path coordinate points with the geographic coordinate range marked by the must-reach points. If the path coordinate point falls within the coordinate range of the must-reach point, the point is identified as the must-reach point and its accompanying attributes are recorded; for the avoidance points, the polygon area judgment algorithm is used to check whether the path coordinate point sequence has an intersection with the polygon area formed by the boundary coordinates of the avoidance points. If there is an intersection, the area is identified as the avoidance point, and its related attributes are recorded, thereby completing the identification of the must-reach points and avoidance points in the initial path.

[0059] 305. Based on the target priority of the must-reach point and the current status of the vehicle, the arrival time window of the must-reach point is dynamically adjusted, and the avoidance range of the avoidance point is expanded or reduced according to the real-time road conditions; For the first must-reach point with high priority, if the distance from the current position of the vehicle to the first must-reach point is greater than the preset distance and the expected arrival time is greater than the preset time range, the arrival time window will be extended backward; if the distance from the current position of the vehicle to the first must-reach point is not greater than the preset distance and the expected arrival time is less than the preset time range, the arrival time window will be compressed forward; for the second must-reach point with low priority, the arrival time window will be dynamically relaxed or tightened according to the current driving progress and the remaining path length of the vehicle. Specifically, for the second must-reach point with low priority, the current position of the vehicle is obtained in real time through the positioning device, and the current driving progress is determined by calculating the proportion of the traveled distance in combination with the pre-planned path. At the same time, the remaining path length is obtained according to the path planning algorithm, and a correlation model of driving progress, remaining path length and time consumption is established based on big data analysis. According to the model and the real-time driving speed of the vehicle, a dynamic estimate is made. The remaining time to reach the second must-reach point; compare and analyze the estimated remaining time with the preset initial arrival time window. If the current driving progress is lower than the preset reasonable progress threshold, and the remaining path length is greater than the remaining length that should be calculated at this progress based on historical data and current road conditions, it is judged that the vehicle is unlikely to arrive within the initial time window. In this case, the arrival time window will be extended backward to relax it according to the preset dynamic adjustment ratio. Conversely, if the driving progress is higher than the preset reasonable progress threshold, and the remaining path length is less than the remaining length that should be, it indicates that the vehicle is likely to arrive early. In this way, the arrival time window will be compressed forward to tighten it according to the corresponding ratio, thereby achieving flexible adjustment of the arrival time window of the second must-reach point; for each avoidance point, when the real-time road conditions show that the traffic around the avoidance point is congested, the avoidance range will be expanded; when the real-time road conditions show that the traffic around the avoidance point is smooth, the avoidance range will be reduced.

[0060] For example, in a mining operation scenario, a target autonomous mining truck is tasked with transporting mined ore from a mining site to a crushing station. Crushing Station A, which is close to the main production area and has an urgent need for ore, is designated as the first destination with high priority. The preset distance is 3 kilometers, and the preset arrival time is 10-15 minutes. After loading the ore, the target autonomous driving vehicle starts and sets off. After driving for a while, it relies on its own positioning system and the positioning base station deployed inside the mine to calculate, and concludes that the distance between the target driving vehicle and crushing station A is 4 kilometers, which is greater than the preset distance. At the same time, the target autonomous driving vehicle obtains information through on-board sensors and real-time interaction with the mine dispatching center, and learns that there are other vehicles intersecting on the roads in the mine and there are many bends. Combined with its own driving speed, the estimated arrival time is 18 minutes, which is greater than the preset time range. At this time, the time window for arriving at crushing station A is extended to 20 minutes. As the target autonomous driving vehicle continues to move towards crushing station A, it is found during recalculation that the distance from the vehicle to this point becomes 2 kilometers, which is no greater than the preset distance, and the estimated arrival time becomes 8 minutes, which is less than the preset time range. At this time, the arrival time window is compressed forward to 9 minutes.

[0061] Additionally, the mine has designated areas for avoidance due to safety hazards or ongoing equipment maintenance. When the autonomous vehicle's real-time monitoring system indicates that large equipment is moving materials around one of these avoidance points, causing traffic congestion and traffic jams in the area, the vehicle's control system immediately expands the radius of this avoidance point from 100 meters to 200 meters. Similarly, when the safety hazard around another avoidance point is resolved, equipment maintenance is completed, and the road is clear again, the control system reduces the radius of that avoidance point from 100 meters to 50 meters.

[0062] Existing technologies mostly use fixed modes when processing the arrival time windows of must-reach points and the avoidance ranges of avoidance points. They cannot flexibly adapt to the complex and changeable actual traffic conditions and vehicle status, which can easily lead to unreasonable planning and affect driving efficiency and safety. However, this solution dynamically adjusts the arrival time windows of must-reach points based on the target priority of the must-reach points and the current status of the vehicle. It can more accurately fit the actual driving conditions of the vehicle, ensure that high-priority tasks are completed on time, and reasonably plan the time of low-priority tasks. It can expand or reduce the avoidance range of avoidance points according to real-time road conditions, effectively avoid congested sections, and make full use of unobstructed roads, greatly improving the flexibility and adaptability of path planning, and enhancing the efficiency and safety of autonomous driving vehicles.

[0063] 306. Using a dynamic optimization algorithm, with the adjusted time window of the must-reach point and the avoidance range of the avoidance point as constraints, the initial path is locally replanned to obtain the optimized path.

[0064] The adjusted time window of the must-reach point and the avoidance range of the avoidance point are used as the constraints of the model, with the shortest driving time as the main optimization goal and the shortest path length as the secondary optimization goal, to construct a dynamic optimization model; the dynamic optimization model is solved using a dynamic optimization algorithm, and a path plan that optimizes the optimization goal is found while satisfying the constraints; based on the solved path plan, the initial path is partially adjusted to generate an optimized path that meets the arrival requirements of the must-reach points and the avoidance requirements of the avoidance points.

[0065] The adjusted time window of the must-reach point is converted into time upper and lower limit constraints, and the avoidance range of the avoidance point is clearly defined as a spatial constraint in the form of a set of geographic coordinates; the shortest driving time is taken as the primary optimization goal, and its priority is reflected by assigning it a higher weight coefficient, and the shortest path length is set as the secondary optimization goal and assigned a relatively low weight, while the above-mentioned time and space constraints are integrated into the model; an improved genetic algorithm is used as the dynamic optimization algorithm, and when initializing the population, multiple mutant path individuals are generated based on the initial path; in the iterative process, individuals are screened according to the fitness function, and new individuals are continuously generated using crossover and mutation operations, and an elite retention strategy is used to ensure that high-quality individuals are not eliminated; finally, under the premise of satisfying all constraints, a path plan that makes the optimization goal optimal is solved, and the initial path is partially adjusted based on this plan, such as replanning the order of passing through the must-reach points, connecting new road sections around the avoidance points, etc., so as to generate an optimized path that meets both the arrival time requirements of the must-reach points and the avoidance requirements of the avoidance points.

[0066] Existing path planning technologies usually use fixed settings for the time window of the required destination and the range of avoidance points. The optimization objectives are single and lack dynamic consideration of actual traffic conditions, resulting in the generated path being difficult to balance efficiency and actual needs. However, this solution uses these dynamically adjusted parameters as constraints, constructs a dynamic optimization model with the shortest driving time as the primary focus and the shortest path length as the secondary focus, and solves it with a dynamic algorithm. This can generate an optimized path that is more realistic, meets the requirements of required destination and avoidance, and is efficient and reasonable, greatly improving the quality of path planning.

[0067] 307. Generate control instructions executable by the target autonomous driving vehicle based on the optimized path to control the target autonomous driving vehicle to operate.

[0068] In an embodiment of the present invention, by obtaining manually annotated waypoint data, combining multi-source data to generate constraint nodes with target priorities and plan the initial path, identifying must-reach points and avoidance points, dynamically adjusting the arrival time window based on the must-reach point priority and vehicle status, flexibly changing the avoidance point range based on real-time road conditions, and then using this as a constraint to locally re-plan the path using a dynamic optimization algorithm, and finally generating executable instructions to control the vehicle. This method can fully consider manual guidance and dynamic changing factors in actual operations, so that path planning meets actual needs, effectively improving the ability of autonomous driving vehicles to cope with complex scenarios, enhancing the rationality and flexibility of path planning, and thereby improving operational efficiency and safety.

[0069] The above describes the vehicle planning method based on manual guidance in the embodiment of the present invention. The following describes the vehicle planning method device based on manual guidance in the embodiment of the present invention. Figure 4 In one embodiment of the present invention, a vehicle planning method and apparatus based on manual guidance includes: An acquisition module 401 is configured to acquire waypoint data corresponding to at least one waypoint manually marked on a supervision platform; A generation module 402 is configured to generate a constraint node with a target priority using a spatiotemporal data fusion algorithm based on at least one waypoint data, combined with map data, real-time traffic data, and vehicle status data; Processing module 403, configured to perform hybrid path planning processing according to the constraint nodes to obtain an initial path; Optimization module 404 is used to make differential adjustments to the required points and avoidance points based on the initial path, and perform local replanning using a dynamic optimization algorithm to obtain an optimized path; The control module 405 is used to generate control instructions executable by the target autonomous driving vehicle based on the optimized path to control the target autonomous driving vehicle to operate.

[0070] In an embodiment of the present invention, by acquiring manually annotated waypoint data and integrating it into path planning, it can quickly respond to and accurately execute manual guidance. By combining multi-source data with a spatiotemporal data fusion algorithm to generate constraint nodes and perform hybrid path planning, and then undergoing differentiated adjustment and local replanning using a dynamic optimization algorithm, it can flexibly adapt to complex and changing operational requirements, plan a more accurate and reasonable path, and ultimately generate executable control instructions, effectively improving the operating efficiency of the target autonomous driving vehicle.

[0071] See also Figure 5 Another embodiment of the vehicle planning device based on manual guidance in the embodiment of the present invention includes: An acquisition module 401 is configured to acquire waypoint data corresponding to at least one waypoint manually marked on a supervision platform; A generation module 402 is configured to generate a constraint node with a target priority using a spatiotemporal data fusion algorithm based on at least one waypoint data, combined with map data, real-time traffic data, and vehicle status data; Processing module 403, configured to perform hybrid path planning processing according to the constraint nodes to obtain an initial path; Optimization module 404 is used to make differential adjustments to the required points and avoidance points based on the initial path, and perform local replanning using a dynamic optimization algorithm to obtain an optimized path; The control module 405 is used to generate control instructions executable by the target autonomous driving vehicle based on the optimized path to control the target autonomous driving vehicle to operate.

[0072] Optionally, the generating module 402 includes: an acquisition unit 4021 configured to read map data related to an area where at least one waypoint is located from a preset map database, and to acquire real-time traffic data of an area surrounding the at least one waypoint and vehicle status data of a target autonomous driving vehicle; A construction unit 4022 is used to pre-process the map data using a spatiotemporal data fusion algorithm to construct a spatiotemporal map model; Extraction unit 4023, configured to extract parameters affecting the accessibility of each waypoint based on at least one waypoint data, real-time road condition data, vehicle status data, and a spatiotemporal map model, using a spatiotemporal graph neural network (ST-GNN) to perform data fusion analysis, and extract parameters affecting the accessibility of each waypoint based on road characteristics, road condition influencing factors, and the current vehicle status; The generating unit 4024 is configured to assign priority weights to nodes corresponding to each waypoint according to the initial priority and corresponding influencing parameters in the waypoint data, and generate constraint nodes with priorities.

[0073] Optionally, the construction unit 4022 may be specifically configured to: Node information and edge information of roads are extracted from map data, where node information includes the location of road intersections, and edge information includes the connection relationship and length of roads. Based on the preset time period division rules, a day is divided into multiple time periods, and the historical traffic data of roads in each time period is obtained. Based on the node information, edge information of the road and the historical communication data of each time period, a spatiotemporal map model is constructed using a spatiotemporal data fusion algorithm. The spatiotemporal map model includes the road topology structure, and the road topology structure includes the dynamic characteristics of the road in multiple time periods.

[0074] Optionally, the extraction unit 4023 may be specifically configured to: The input dataset is constructed by integrating at least one waypoint data, real-time road condition data, vehicle status data and relevant data in the spatiotemporal map model. The input dataset is input into the spatiotemporal graph neural network ST-GNN, and the input dataset is processed layer by layer through its multi-layer network structure to obtain an output structure. Based on the output structure, the road characteristics, road condition influencing factors and the impact parameters of the vehicle's current status on the accessibility of each waypoint are extracted respectively.

[0075] Optionally, the generating unit 4024 may be specifically configured to: Determine the initial weight values ​​corresponding to different initial priorities; analyze the influencing parameters of each waypoint; for unfavorable factors affecting the accessibility of the waypoint, reduce the initial weight value of the corresponding waypoint according to the degree of its impact on accessibility; for favorable factors, increase the initial weight value of the corresponding waypoint according to the degree of its improvement on accessibility; assign the adjusted weight to the node corresponding to each waypoint to generate a constraint node with the target priority, where the higher the priority weight, the more important the node in path planning. Optionally, the processing module 403 may be specifically configured to: A graph-search-based path planning algorithm is used. The current position of the target autonomous driving vehicle is used as the starting point, and a search graph is constructed in combination with constraint nodes. Based on the search graph, a path search is performed according to the target priority and access constraint attributes of each constraint node. When the target point that meets the conditions of all constraint nodes is found, the initial path is determined based on the nodes passed during the search process.

[0076] Optionally, the optimization module 404 includes: Identification unit 4041, used to identify the must-reach points and avoidance points in the initial path, the must-reach points are manually marked mandatory waypoints, and the avoidance points are manually marked prohibited areas; Adjustment unit 4042, for dynamically adjusting the arrival time window of the must-reach point based on the target priority of the must-reach point and the current state of the vehicle, and expanding or reducing the avoidance range of the avoidance point according to the real-time road conditions; The replanning unit 4043 is used to adopt a dynamic optimization algorithm to locally replan the initial path with the adjusted time window of the must-reach point and the avoidance range of the avoidance point as constraints to obtain an optimized path.

[0077] Optionally, the adjusting unit 4042 may be specifically configured to: For the first must-reach point with high priority, if the distance from the vehicle's current position to the first must-reach point is greater than the preset distance and the expected arrival time is greater than the preset time range, the arrival time window will be extended backward; if the distance from the vehicle's current position to the first must-reach point is not greater than the preset distance and the expected arrival time is less than the preset time range, the arrival time window will be compressed forward; for the second must-reach point with low priority, the arrival time window will be dynamically relaxed or tightened based on the vehicle's current driving progress and the remaining path length; for each avoidance point, when the real-time traffic conditions show that the traffic around the avoidance point is congested, the avoidance range will be expanded; when the real-time traffic conditions show that the traffic around the avoidance point is smooth, the avoidance range will be reduced.

[0078] Optionally, the re-planning unit 4043 may be specifically configured to: The adjusted time window of the must-reach point and the avoidance range of the avoidance point are used as the constraints of the model, with the shortest driving time as the main optimization goal and the shortest path length as the secondary optimization goal, to construct a dynamic optimization model; the dynamic optimization model is solved using a dynamic optimization algorithm, and a path plan that optimizes the optimization goal is found while satisfying the constraints; based on the solved path plan, the initial path is partially adjusted to generate an optimized path that meets the arrival requirements of the must-reach points and the avoidance requirements of the avoidance points.

[0079] In an embodiment of the present invention, by obtaining manually annotated waypoint data, combining multi-source data with a spatiotemporal data fusion algorithm to construct a spatiotemporal map model, and using a spatiotemporal graph neural network to extract key influencing parameters, constraint nodes with reasonable priorities are generated. This fully considers the various complex factors in actual operations, and uses a graph search-based algorithm for hybrid path planning. It also identifies must-reach points and avoidance points, dynamically adjusts the must-reach point time window and avoidance point range based on a variety of conditions, and uses this as a constraint to locally replan the path through a dynamic optimization algorithm, making the path planning more practical and flexible to adapt to changes. Ultimately, executable instructions are generated to control vehicle operations, effectively improving the accuracy, rationality, and flexibility of autonomous vehicle path planning, enhancing the ability to cope with complex scenarios, and improving operational efficiency and safety.

[0080] above Figure 4 and Figure 5 The vehicle planning device based on manual guidance in the embodiment of the present invention is described in detail from the perspective of modular functional entities, and the electronic device in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0081] See also Figure 6As shown, the electronic device includes a processor 600 and a memory 601 , wherein the memory 601 stores machine-executable instructions that can be executed by the processor 600 , and the processor 600 executes the machine-executable instructions to implement the above-mentioned vehicle planning method based on manual guidance.

[0082] Further, Figure 6 The electronic device shown further includes a bus 602 and a communication interface 603 , and the processor 600 , the communication interface 603 and the memory 601 are connected via the bus 602 .

[0083] The memory 601 may include a high-speed random access memory (RAM) and may also include a non-volatile memory (non-volatile memory), such as at least one disk storage. The communication connection between the system network element and at least one other network element is achieved through at least one communication interface 603 (which may be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. may be used. The bus 602 may be an ISA bus, a PCI bus, or an EISA bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 6 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or one type of bus.

[0084] The processor 600 may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method can be completed by hardware integrated logic circuits or software instructions in the processor 600. The above processor 600 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of the present disclosure. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in conjunction with the embodiments of the present disclosure can be directly implemented and executed by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium mature in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. The storage medium is located in the memory 601 , and the processor 600 reads the information in the memory 601 and completes the method steps of the aforementioned embodiment in combination with its hardware.

[0085] The present invention also provides an electronic device, wherein the computer device includes a memory and a processor, wherein the memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor executes the steps of the vehicle planning method based on manual guidance in the above-mentioned embodiments.

[0086] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the steps of the vehicle planning method based on manual guidance.

[0087] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0088] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0089] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features thereof can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A vehicle planning method based on manual guidance, characterized in that: The vehicle planning method based on manual guidance includes: Obtaining waypoint data corresponding to at least one waypoint manually marked on the supervision platform; Based on at least one waypoint data, combined with map data, real-time traffic data and vehicle status data, a spatiotemporal data fusion algorithm is used to generate constraint nodes with target priorities; Perform hybrid path planning processing according to the constraint nodes to obtain an initial path; Based on the initial path, differential adjustments are made to the must-reach points and avoidance points, and local replanning is performed using a dynamic optimization algorithm to obtain an optimized path; Based on the optimized path, control instructions executable by the target autonomous driving vehicle are generated to control the target autonomous driving vehicle to operate.

2. The vehicle planning method based on manual guidance according to claim 1, characterized in that: Each waypoint data includes the geographical coordinates, initial priority and traffic constraint attributes of the waypoint. The method of generating a constraint node with a priority based on at least one waypoint data in combination with map data, real-time traffic data and vehicle status data includes: Reading map data related to the area where the at least one waypoint is located from a preset map database, and obtaining real-time traffic data for the area surrounding the at least one waypoint and vehicle status data of the target autonomous driving vehicle; Preprocessing the map data using a spatiotemporal data fusion algorithm to construct a spatiotemporal map model; Based on at least one waypoint data, the real-time road condition data, the vehicle status data, and the spatiotemporal map model, a spatiotemporal graph neural network (ST-GNN) is used to perform data fusion analysis to extract parameters affecting the accessibility of each waypoint, including road characteristics, road condition influencing factors, and the current state of the vehicle; According to the initial priority and corresponding influencing parameters in the data of each waypoint, priority weights are assigned to the nodes corresponding to each waypoint to generate constraint nodes with priorities.

3. The vehicle planning method based on manual guidance according to claim 2, characterized in that: The method of preprocessing the map data using a spatiotemporal data fusion algorithm to construct a spatiotemporal map model includes: Extracting node information and edge information of roads from the map data, wherein the node information includes the location of road intersections, and the edge information includes the connection relationship and length of the roads; Based on the preset time period division rules, a day is divided into multiple time periods, and historical traffic data of roads in each time period is obtained; Based on the node information, edge information and historical communication data of each time period of the road, a spatiotemporal map model is constructed using a spatiotemporal data fusion algorithm. The spatiotemporal map model includes a road topology structure, and the road topology structure includes multi-time period dynamic characteristics of the road.

4. The vehicle planning method based on manual guidance according to claim 2, characterized in that: The method uses a spatiotemporal graph neural network (ST-GNN) to perform data fusion analysis based on at least one waypoint data, the real-time road condition data, the vehicle status data, and the spatiotemporal map model to extract parameters affecting the accessibility of each waypoint, including: Integrate the at least one waypoint data, the real-time traffic data, the vehicle status data, and relevant data in the spatiotemporal map model to construct an input data set; Input the input data set into the spatiotemporal graph neural network ST-GNN, and process the input data set layer by layer through its multi-layer network structure to obtain an output structure; Based on the output structure, the road characteristics, road condition influencing factors and the influencing parameters of the vehicle's current state on the accessibility of each waypoint are extracted respectively.

5. The vehicle planning method based on manual guidance according to claim 2, characterized in that: According to the priority and influencing parameters of each waypoint, the nodes corresponding to each waypoint are assigned priority weights to generate constraint nodes with target priorities. Determine the initial weight values ​​corresponding to different initial priorities; Analyze the influencing parameters of each waypoint. For unfavorable factors affecting the accessibility of the waypoint, reduce the initial weight of the corresponding waypoint according to the degree of its impact on accessibility. For favorable factors, increase the initial weight of the corresponding waypoint according to the degree of its improvement on accessibility. The adjusted weights are assigned to the nodes corresponding to each waypoint to generate constraint nodes with target priorities. The higher the priority weight, the more important the node is in path planning.

6. The vehicle planning method based on manual guidance according to claim 1, characterized in that: The performing hybrid path planning processing according to the constraint nodes to obtain an initial path includes: Using a graph-based path planning algorithm, taking the current position of the target autonomous driving vehicle as a starting point and combining the constraint nodes to construct a search graph; Based on the search graph, a path search is performed according to the target priority and the access constraint attribute of each constraint node; When the target point that meets all the constraint node conditions is found, the initial path is determined based on the nodes passed during the search process.

7. The vehicle planning method based on manual guidance according to claim 1, characterized in that: Based on the initial path, differential adjustments are made to the must-reach points and the avoidance points, and local replanning is performed using a dynamic optimization algorithm to obtain an optimized path, including: Identifying mandatory points and avoidance points in the initial path, wherein the mandatory points are manually marked mandatory waypoints and the avoidance points are manually marked prohibited areas; Based on the target priority of the must-reach point and the current state of the vehicle, the arrival time window of the must-reach point is dynamically adjusted, and the avoidance range of the avoidance point is expanded or reduced according to the real-time road conditions; A dynamic optimization algorithm is used to locally replan the initial path with the adjusted time window of the must-reach point and the avoidance range of the avoidance point as constraints to obtain an optimized path.

8. The vehicle planning method based on manual guidance according to claim 7, characterized in that: The method of dynamically adjusting the arrival time window of the must-reach point based on the target priority of the must-reach point and the current state of the vehicle, and expanding or reducing the detour range of the detour point according to the real-time road conditions, includes: For a high-priority first-must-reach point, if the distance from the vehicle's current location to the first must-reach point is greater than the preset distance and the estimated arrival time is greater than the preset time range, the arrival time window will be extended backward; if the distance from the vehicle's current location to the first must-reach point is not greater than the preset distance and the estimated arrival time is less than the preset time range, the arrival time window will be compressed forward; For the second destination with a low priority, the arrival time window is dynamically relaxed or tightened based on the vehicle's current driving progress and remaining path length. For each detour point, when the real-time traffic conditions show that the traffic around the detour point is congested, the detour range will be expanded; when the real-time traffic conditions show that the traffic around the detour point is unobstructed, the detour range will be reduced.

9. The vehicle planning method based on manual guidance according to claim 7, characterized in that: The dynamic optimization algorithm is used to locally replan the initial path with the adjusted time window of the required point and the avoidance range of the avoidance point as constraints to obtain an optimized path, including: The adjusted time window of the must-reach point and the avoidance range of the avoidance point are used as the constraints of the model, with the shortest travel time as the primary optimization goal and the shortest path length as the secondary optimization goal to build a dynamic optimization model. Solving the dynamic optimization model using a dynamic optimization algorithm, and finding a path solution that optimizes the optimization goal while satisfying the constraints; According to the path solution obtained by the solution, the initial path is locally adjusted to generate an optimized path, and the optimized path meets the arrival requirements of the must-reach points and the avoidance requirements of the avoidance points.

10. A vehicle planning device based on manual guidance, characterized in that: The vehicle planning device based on manual guidance includes: An acquisition module, configured to acquire waypoint data corresponding to at least one waypoint manually marked on the supervision platform; A generation module, configured to generate a constraint node with a target priority using a spatiotemporal data fusion algorithm based on at least one waypoint data, combined with map data, real-time traffic data, and vehicle status data; A processing module, configured to perform hybrid path planning processing according to the constraint nodes to obtain an initial path; An optimization module is used to make differential adjustments to the required points and avoidance points based on the initial path, and perform local replanning through a dynamic optimization algorithm to obtain an optimized path; A control module is used to generate control instructions executable by the target autonomous driving vehicle based on the optimized path to control the operation of the target autonomous driving vehicle.

11. An electronic device, characterized in that: The electronic device comprises: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory to enable the electronic device to execute the vehicle planning method based on manual guidance as described in any one of claims 1 to 9.

12. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by a processor, the vehicle planning method based on manual guidance as described in any one of claims 1 to 9 is implemented.

Images