Artificially guided vehicle planning method, device, equipment and storage medium

Abstract

The present application relates to the technical field of automatic driving, and discloses a vehicle planning method and device based on artificial guidance, equipment and a storage medium, which are used for improving the operation efficiency of an automatic driving vehicle. The vehicle planning method based on artificial guidance comprises the following steps: obtaining at least one passing point data corresponding to a passing point marked by an artificial on a supervision platform; based on the at least one passing point data, combining map data, real-time road condition data and vehicle state data, and using a space-time data fusion algorithm to generate a constraint node with a target priority; performing mixed path planning processing according to the constraint node to obtain an initial path; based on the initial path, performing differential adjustment on a must-arrive point and an avoidance point, and performing local re-planning through a dynamic optimization algorithm to obtain an optimized path; and generating a control instruction executable by a target automatic driving vehicle based on the optimized path to control the target automatic driving vehicle to operate.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, and in particular to a vehicle planning method, apparatus, device, and storage medium based on human guidance. Background Technology

[0002] In today's era of rapid development in automation technology, logistics and operations in closed environments such as mines and ports are accelerating towards intelligence and automation. Autonomous vehicles, with their advantages of high efficiency, stability, and the ability to operate continuously for extended periods, are finding increasingly widespread application in these closed environments.

[0003] Currently, in enclosed environments such as mines and ports, most existing vehicle planning methods are based on pre-defined fixed rules and simple path planning algorithms. These methods typically pre-define basic information such as vehicle routes and stopping points in the system, combining static map data within the environment with the vehicle's initial state data to generate the vehicle's path. Existing methods lack effective integration of human guidance. In actual operations, operators may require manual intervention and adjustments, but existing planning methods struggle to respond quickly and accurately to these manual instructions, failing to meet the complex and ever-changing operational needs of enclosed environments, thus leading to low vehicle operation efficiency. Summary of the Invention

[0004] This invention provides a vehicle planning method, apparatus, device, and storage medium based on human guidance to solve the problem of low vehicle operation efficiency caused by the lack of human guidance in the prior art.

[0005] The first aspect of this invention provides a vehicle planning method based on manual guidance, comprising: acquiring route point data corresponding to at least one route point manually marked on a monitoring platform; generating constraint nodes with target priorities using a spatiotemporal data fusion algorithm based on the at least one route point data, combined with map data, real-time traffic data, and vehicle status data; performing hybrid path planning processing based on the constraint nodes to obtain an initial path; making differentiated adjustments to must-reach points and detour points based on the initial path, and performing local replanning through a dynamic optimization algorithm to obtain an optimized path; and generating control commands executable by the target autonomous vehicle based on the optimized path to control the operation of the target autonomous vehicle.

[0006] In one feasible implementation, each waypoint data includes the geographical coordinates, initial priority, and traffic constraint attributes of the waypoint. The step of generating priority-bearing constraint nodes based on at least one waypoint data, combined 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 surrounding area of ​​the at least one waypoint and vehicle status data of the target autonomous vehicle; preprocessing the map data using a spatiotemporal data fusion algorithm to construct a spatiotemporal map model; performing data fusion analysis using a spatiotemporal graph neural network (ST-GNN) based on the at least one waypoint data, the real-time traffic data, the vehicle status data, and the spatiotemporal map model, respectively extracting the influence parameters of road features, traffic condition influencing factors, and the current vehicle status on the accessibility of each waypoint; and assigning priority weights to the nodes corresponding to each waypoint according to the initial priority and corresponding influence parameters in the waypoint data, thereby generating priority-bearing constraint nodes.

[0007] In one feasible implementation, the preprocessing of the map data using a spatiotemporal data fusion algorithm to construct a spatiotemporal map model includes: extracting road node information and edge information from the map data, wherein the node information includes the location of road intersections, and the edge information includes the connection relationships and lengths of roads; dividing a day into multiple time periods based on a preset time period division rule, and obtaining historical traffic data of roads in each time period; and constructing a spatiotemporal map model using a spatiotemporal data fusion algorithm based on the road node information, edge information, and historical communication data of each time period, wherein the spatiotemporal map model includes a road topology structure, and the road topology structure includes dynamic features of roads across multiple time periods.

[0008] In one feasible implementation, the step of performing data fusion analysis using a spatiotemporal graph neural network (ST-GNN) based on at least one waypoint data, the real-time traffic data, the vehicle status data, and the spatiotemporal map model to extract road features, traffic condition influencing factors, and parameters affecting the accessibility of each waypoint includes: integrating the at least one waypoint data, the real-time traffic data, the vehicle status data, and relevant data from the spatiotemporal map model to construct an input dataset; inputting the input dataset into the ST-GNN, processing the input dataset layer by layer through its multi-layer network structure to obtain an output structure; and extracting parameters affecting the accessibility of each waypoint based on the output structure, including road features, traffic condition influencing factors, and parameters affecting the accessibility of each waypoint.

[0009] In one feasible implementation, priority weights are assigned to the nodes corresponding to each waypoint based on their priority and influence parameters, generating constraint nodes with target priorities and determining the initial weight values ​​corresponding to different initial priorities. The influence parameters of each waypoint are analyzed. For unfavorable factors affecting the accessibility of waypoints, the initial weight values ​​of the corresponding waypoints are adjusted by decreasing their impact on accessibility. For favorable factors, the initial weight values ​​of the corresponding waypoints are adjusted by increasing their improvement on accessibility. The adjusted weights are assigned to the nodes corresponding to each waypoint, generating constraint nodes with target priorities, where a higher priority weight indicates a higher importance of the node in path planning.

[0010] In one feasible implementation, the step of performing hybrid path planning based on the constraint nodes to obtain an initial path includes: using a graph search-based path planning algorithm, constructing a search graph starting from the current position of the target autonomous vehicle and combining it with the constraint nodes; performing path search based on the search graph according to the target priority and traffic constraint attributes of each constraint node; and determining the initial path based on the nodes traversed during the search process when a target point that satisfies the conditions of all constraint nodes is found.

[0011] In one feasible implementation, the step of making differentiated adjustments to the mandatory destinations and detour points based on the initial path, and performing local replanning through a dynamic optimization algorithm to obtain an optimized path, includes: identifying mandatory destinations and detour points in the initial path, wherein the mandatory destinations are manually marked mandatory route points, and the detour points are manually marked prohibited areas; dynamically adjusting the arrival time window of the mandatory destinations based on the target priority of the mandatory destinations and the current vehicle status, and expanding or shrinking the detour range of the detour points according to real-time traffic conditions; and using a dynamic optimization algorithm, with the adjusted arrival time window of the mandatory destinations and the detour range of the detour points as constraints, performing local replanning on the initial path to obtain the optimized path.

[0012] In one feasible implementation, the dynamic adjustment of the arrival time window of the must-reach point based on its target priority and the vehicle's current status, and the expansion or contraction of the detour range of the bypass point according to real-time traffic conditions, includes: for a high-priority first must-reach point, if the distance from the vehicle's current position to the first must-reach point is greater than a preset distance and the estimated arrival time is greater than a preset time range, then the arrival time window is extended; if the distance from the vehicle's current position to the first must-reach point is not greater than a preset distance and the estimated arrival time is less than a preset time range, then the arrival time window is compressed; for a low-priority second must-reach point, the arrival time window is dynamically widened or tightened according to the vehicle's current travel progress and remaining path length; for each bypass point, when real-time traffic conditions show traffic congestion around the bypass point, its bypass range is expanded; when real-time traffic conditions show traffic flow is smooth around the bypass point, its bypass range is narrowed.

[0013] In one feasible implementation, the step of employing a dynamic optimization algorithm, using the adjusted time window of the must-reach point and the avoidance range of the bypass point as constraints, performs local replanning on the initial path to obtain an optimized path. This includes: using the adjusted time window of the must-reach point and the avoidance range of the bypass point as constraints of the model, with the shortest travel time as the primary optimization objective and the shortest path length as the secondary optimization objective, to construct a dynamic optimization model; using the dynamic optimization algorithm to solve the dynamic optimization model, and finding the path scheme that optimizes the objectives while satisfying the constraints; and based on the solved path scheme, locally adjusting the initial path to generate an optimized path, wherein the optimized path satisfies the arrival requirements of the must-reach point and the avoidance requirements of the bypass point.

[0014] A second aspect of the present invention provides a vehicle planning device based on human guidance, comprising: an acquisition module for acquiring route point data corresponding to at least one route point manually marked on a monitoring platform; a generation module for generating constraint nodes with target priorities based on the at least one route point data, combined with map data, real-time traffic data, and vehicle status data, using a spatiotemporal data fusion algorithm; a processing module for performing hybrid path planning processing based on the constraint nodes to obtain an initial path; an optimization module for making differentiated adjustments to must-reach points and detour points based on the initial path, and performing local replanning through a dynamic optimization algorithm to obtain an optimized path; and a control module for generating control commands executable by the target autonomous vehicle based on the optimized path to control the operation of the target autonomous vehicle.

[0015] In one feasible implementation, the generation module includes: an acquisition unit, configured to read map data related to the area where the at least one waypoint is located from a preset map database, and acquire real-time traffic data and vehicle status data of the target autonomous vehicle in the surrounding area of ​​the at least one waypoint; a construction unit, configured to preprocess the map data using a spatiotemporal data fusion algorithm to construct a spatiotemporal map model; an extraction unit, configured to extract data based on at least one waypoint data, the real-time traffic data, the vehicle status data, and the spatiotemporal map model, and perform data fusion analysis using a spatiotemporal graph neural network (ST-GNN) to extract road features, traffic condition influencing factors, and the influence parameters of the current vehicle status on the accessibility of each waypoint; and a generation unit, configured to assign priority weights to the nodes corresponding to each waypoint according to the initial priority and corresponding influence parameters in the data of each waypoint, and generate priority-bound constraint nodes.

[0016] In one feasible implementation, the construction unit is specifically used to: extract road node information and edge information 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; divide a day into multiple time periods based on a preset time period division rule, and obtain historical traffic data of roads in each time period; construct a spatiotemporal map model using a spatiotemporal data fusion algorithm based on the road node information, edge information, and historical communication data of each time period, wherein the spatiotemporal map model includes a road topology structure, and the road topology structure includes dynamic features of roads in multiple time periods.

[0017] In one feasible implementation, the extraction unit is specifically used to: 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 dataset; input the input dataset into a spatiotemporal graph neural network (ST-GNN), and process the input dataset layer by layer through its multi-layer network structure to obtain an output structure; and extract road features, traffic condition influencing factors, and the influence parameters of the vehicle's current status on the reachability of each waypoint based on the output structure. In one feasible implementation, the generation unit is specifically used to: determine the initial weight values ​​corresponding to different initial priorities; analyze the influence parameters of each waypoint; for unfavorable factors affecting the accessibility of waypoints, adjust the initial weight values ​​of the corresponding waypoints by decreasing them according to their degree of influence on accessibility; for favorable factors, adjust the initial weight values ​​of the corresponding waypoints by increasing them according to their degree of improvement on accessibility; assign the adjusted weights to the nodes corresponding to each waypoint, generating constraint nodes with target priorities, wherein the higher the priority weight, the higher the importance of the node in path planning. In one feasible implementation, the processing module is specifically used to: construct a search graph using a graph search-based path planning algorithm, starting from the current position of the target autonomous vehicle and combining it with the constraint nodes; perform path search based on the search graph according to the target priority and traffic constraint attributes of each constraint node; and determine the initial path based on the nodes traversed during the search process when a target point that satisfies the conditions of all constraint nodes is found.

[0018] In one feasible implementation, the optimization module includes: an identification unit, used to identify mandatory reachable points and detour points in the initial path, wherein the mandatory reachable points are manually marked mandatory route points and the detour points are manually marked prohibited areas; an adjustment unit, used to dynamically adjust the arrival time window of the mandatory reachable points based on the target priority of the mandatory reachable points and the current state of the vehicle, and to expand or shrink the detour range of the detour points according to real-time traffic conditions; and a replanning unit, used to perform local replanning of the initial path using a dynamic optimization algorithm, with the adjusted time window of the mandatory reachable points and the detour range of the detour points as constraints, to obtain an optimized path.

[0019] In one feasible implementation, the adjustment unit is specifically used to: 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 a preset distance and the estimated arrival time is greater than a preset time range, then extend the arrival time window; if the distance from the vehicle's current location to the first must-reach point is not greater than a preset distance and the estimated arrival time is less than a preset time range, then compress the arrival time window forward; for a low-priority second must-reach point, dynamically widen or tighten the arrival time window according to the vehicle's current travel progress and remaining path length; for each bypass point, when real-time traffic conditions show that the area around the bypass point is congested, expand its bypass range; when real-time traffic conditions show that the area around the bypass point is clear, narrow its bypass range.

[0020] In one feasible implementation, the replanning unit is specifically used to: use the adjusted time window of the must-reach point and the detour range of the avoidance point as constraints of the model, with the shortest travel time as the primary optimization objective and the shortest path length as the secondary optimization objective, to construct a dynamic optimization model; use a dynamic optimization algorithm to solve the dynamic optimization model, and find the path scheme that optimizes the objective while satisfying the constraints; based on the solved path scheme, locally adjust the initial path to generate an optimized path, wherein the optimized path satisfies the arrival requirements of the must-reach point and the detour 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 invokes the instructions in the memory to cause the electronic device to execute the above-described vehicle planning method based on human guidance.

[0022] A fourth aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the above-described human-guided vehicle planning method.

[0023] The technical solution provided by this invention involves acquiring path point data corresponding to at least one path point manually marked on a monitoring platform; based on the at least one path point 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; hybrid path planning is performed according to the constraint nodes to obtain an initial path; based on the initial path, must-reach points and detour points are differentiated and adjusted, and local replanning is performed through a dynamic optimization algorithm to obtain an optimized path; based on the optimized path, executable control commands for the target autonomous vehicle are generated to control the operation of the target autonomous vehicle. In this embodiment of the invention, by acquiring manually marked path point data and integrating it into path planning, it can quickly respond to and accurately execute manual guidance. By combining multi-source data and using a spatiotemporal data fusion algorithm to generate constraint nodes and perform hybrid path planning, followed by differentiated adjustment and local replanning through a dynamic optimization algorithm, it can flexibly adapt to complex and ever-changing operational needs, plan more accurate and reasonable paths, and finally generate executable control commands, effectively improving the operational efficiency of the target autonomous vehicle. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of an embodiment of the vehicle planning method based on manual guidance in this invention. Figure 2 This is a schematic diagram of another embodiment of the vehicle planning method based on manual guidance in this invention. Figure 3 This is a schematic diagram of another embodiment of the vehicle planning method based on manual guidance in the present invention. Figure 4 This is a schematic diagram of an embodiment of the vehicle planning device based on manual guidance in this invention. Figure 5 This is a schematic diagram of another embodiment of the vehicle planning device based on manual guidance in this invention. Figure 6 This is a schematic diagram of one embodiment of the electronic device in this invention. Detailed Implementation

[0025] This invention provides a vehicle planning method, apparatus, device, and storage medium based on manual guidance. By acquiring route point data marked by humans on a monitoring platform and effectively integrating it into route planning, vehicle planning can quickly respond to and accurately execute human intervention adjustments based on actual operating conditions, thereby improving vehicle operation efficiency.

[0026] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0027] It is understood that the executing entity of this invention can be a vehicle planning device based on manual guidance, or it can be a terminal or a server; no specific limitation is made here. This embodiment of the invention will be described using a server as the executing entity as an example.

[0028] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 One embodiment of the vehicle planning method based on manual guidance in this invention includes: 101. Obtain route data corresponding to at least one route point manually marked on the monitoring platform; Each waypoint can have additional attributes, such as dwell time and initial priority. The monitoring platform provides an interactive map interface, allowing operators to mark key points on the map and set relevant parameters by clicking or dragging with the mouse. The system stores the geographic coordinates, type, and additional attributes of these waypoints in a structured manner, such as using JSON or Protobuf format. Simultaneously, the system performs validity checks on the input data, such as checking whether the coordinates are within a reasonable range and whether the radius of the detour point is too large, to avoid path planning anomalies due to incorrect input.

[0029] For example, staff members operate on the mining area map through the user interface of the monitoring platform. The staff marked three points of passage: point A (temporary ore storage area), which is set as high priority and must be reached; point B (dangerous road section with a slope greater than 15°), which is marked as an avoidance point and set as low priority; and point C (maintenance station), which is set as medium priority and requires 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 constraint nodes with target priority. Based on the geographic coordinates of at least one transit point, the original map data of the corresponding area is obtained from a preset map database; the original map data is parsed to extract the road elements, and the road elements are reorganized according to a preset data structure; based on the reorganized road element data, roads are constructed to obtain road network topology information.

[0031] Acquire real-time traffic data of the area surrounding at least one route point, and extract traffic information of each road segment from the real-time traffic data; based on the preset traffic condition and traffic capacity influence relationship model, calculate the traffic condition influence factor of each road segment using the traffic condition information of each road segment. The traffic condition influence factor reflects the degree to which traffic conditions reduce road traffic capacity.

[0032] The process involves acquiring vehicle status data of the target autonomous vehicle, combining it with at least one waypoint data, road network topology information, and road condition influencing factors, and generating constraint nodes with target priorities according to a preset priority algorithm. Specifically, devices such as speed sensors, battery sensors, and load sensors can be used to collect real-time vehicle status data such as vehicle speed, remaining battery power, and load status of the target autonomous vehicle. This vehicle status data, at least one waypoint data, road network topology information, and road condition influencing factors calculated based on real-time road condition data are all input into a preset priority algorithm model. This model first uses the preset initial priority of the waypoints as a basis, comprehensively considering the vehicle status (e.g., when the vehicle battery is low, priority is given to ensuring access to necessary and nearby waypoints). Simultaneously, it incorporates road condition influencing factors, avoiding road sections surrounding low-priority waypoints with poor road conditions and non-necessary access. The priority of each waypoint is re-evaluated through a weighted calculation method, ultimately generating constraint nodes with target priorities.

[0033] For example, based on the coordinates of the marked route points A (temporary ore storage area), B (dangerous road section), and C (maintenance station), road data for the corresponding areas is extracted from the mining area map database. This data is analyzed to construct a topological network containing key attributes such as road slope and width. Real-time traffic information is obtained, revealing that the congestion index around point A reaches 0.8 due to the dense concentration of ore transport vehicles. According to the traffic condition assessment model, the traffic capacity of this road section has decreased by 60%. Considering the current status of the mining trucks (load of 35 tons and battery level of 28%), the priority algorithm dynamically adjusts, raising the priority of the must-reach point A from the preset 8 to 9. Due to limited battery power, this critical task needs to be completed first. The detour point B... Priority 2 remains unchanged, and the bypass radius is set at 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 battery level is below 30%. Finally, three constraint nodes are output: point A, with coordinates (x1, y1), target priority 9 and must be reached; point B, with coordinates (x2, y2), target priority 2 and bypass radius of 50 meters; and point C, with coordinates (x3, y3), target priority 4, requires a stay of 10 minutes and meets the battery level trigger condition.

[0034] 103. Perform hybrid path planning based on the constraint nodes to obtain the initial path; Based on the constraint nodes, a graph search algorithm, such as A* or Dijkstra's algorithm, is used. The appropriate algorithm is selected according to the specific scenario requirements. If dynamic factors such as real-time traffic conditions need to be considered, an improved A* algorithm can be used. A search graph is constructed based on the road network topology information, with constraint nodes serving as key nodes. The weights of each road segment in the search graph are dynamically adjusted using real-time traffic data; for example, a higher traffic condition impact factor results in a higher segment weight and higher travel cost. Simultaneously, vehicle status data, such as the impact of load and battery level on vehicle driving capacity, are incorporated as additional constraints. These factors are comprehensively considered during the search process. The algorithm iteratively searches for the optimal path from the starting point to the ending point, passing through all constraint nodes and satisfying both vehicle driving capacity and real-time traffic condition requirements. This path serves as the initial path. The algorithm traverses based on target priority.

[0035] For example, based on the constraint nodes A, B, and C, 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. Combining real-time traffic data, the weights of each road segment in the search graph are dynamically adjusted. Since the road segment near point A is congested and has a large impact factor on traffic conditions, its weight is increased, indicating high travel costs. At the same time, the vehicle status data of the mining truck, such as load and battery level, are considered. Load affects the vehicle's acceleration and braking performance, while battery level limits the driving range. These factors are incorporated into the search process as additional constraints. The algorithm traverses according to the target priority, starting from the current position of the mining truck, and 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 traffic requirements, serving as the initial path.

[0036] 104. Based on the initial path, differentiate the must-reach points and detour points, and perform local replanning through a dynamic optimization algorithm to obtain the optimized path; For guaranteed destinations, the system checks whether the initial path ensures the vehicle can reach them accurately. If a guaranteed destination is not reasonably included in the path (e.g., bypassing a severely congested section during path planning), the path is readjusted. This readjustment involves re-searching for feasible paths within a certain range around the guaranteed destination, taking into account real-time traffic conditions and road topology to ensure the vehicle reaches the guaranteed destination in the shortest time and at the lowest cost. For detour points, the system continuously monitors the distance between the vehicle and the detour point. If, due to changes in road conditions or other reasons, the vehicle tends to approach the detour point, the dynamic optimization algorithm immediately initiates local replanning. This algorithm uses the vehicle's current position as a new starting point, combining remaining constraint nodes, real-time traffic conditions, and vehicle status to replan a local path while avoiding the detour point. During local replanning, the system fully considers the continuity and smoothness of the path to avoid instability caused by sudden path changes. Meanwhile, the dynamic optimization algorithm evaluates the merits of different path schemes in real time. Through continuous iterative optimization, it selects the optimal local path scheme, gradually corrects and improves the initial path, and finally obtains an optimized path that meets the requirement of reaching the destination and effectively avoids detour points.

[0037] For example, if it is found that due to changes in real-time traffic conditions, the route originally planned to pass through point A is becoming increasingly congested, and continuing along the original route may result in failure to reach point A on time, then adjustments need to be made to the route to the essential point A. A feasible route is re-searched within a certain range around point A, taking into account the updated real-time traffic conditions and road topology to ensure the mining truck can reach point A in the shortest time and at the lowest cost. Simultaneously, if it is detected that the mining truck's direction of travel is trending towards the avoidance point B, this is because the road ahead is partially closed due to construction, affecting the originally planned route. At this point, the dynamic optimization algorithm immediately initiates local replanning, using the mining truck's current position as the new starting point, and combining the remaining constraint nodes, such as points A and C, real-time traffic conditions (avoiding congested and construction sections), and vehicle status (e.g., 28% battery charge, full load), a new local route is planned while avoiding point B. During the planning process, the continuity and smoothness of the route are considered to avoid unstable driving situations such as sudden braking or sharp turns caused by sudden route changes. Through continuous iterative optimization, the optimal local path scheme is selected, and the initial path is corrected and improved. Finally, an optimized path is obtained that ensures accurate arrival at the must-reach point A, passes through point C for maintenance at the appropriate time, and effectively avoids the detour point B.

[0038] 105. Generate control commands that can be executed by the target autonomous vehicle based on the optimized path to control the operation of the target autonomous vehicle.

[0039] The optimized path is discretized, dividing the continuous path curve into multiple path points at certain time or distance intervals. Each path point contains key information such as the vehicle's geographic coordinates, speed, acceleration, and steering angle at that location. Based on the vehicle's dynamics and kinematics models, combined with current vehicle status data such as current speed, steering angle, and battery level, the control parameters corresponding to each path point are further optimized and adjusted. For example, based on the vehicle's maximum acceleration and braking capacity, the speed parameters at the path points are reasonably limited to ensure that the vehicle's driving safety and comfort are not affected by excessively rapid speed changes. The optimized path point information is encapsulated into control commands according to specific communication protocols and interface specifications. These control commands are transmitted in real time to various control modules of the vehicle, such as the engine control module, braking control module, and steering control module, through the onboard communication system, such as CAN bus or Ethernet. After receiving the control commands, each control module controls the vehicle's driving state according to the parameter information in the commands, enabling the vehicle to accurately travel along the optimized path.

[0040] In this embodiment of the invention, by acquiring manually marked waypoint data and integrating it into path planning, it can quickly respond to and accurately execute manual guidance. By combining multi-source data and using spatiotemporal data fusion algorithms to generate constraint nodes and perform hybrid path planning, and then using differentiated adjustment and dynamic optimization algorithms for local replanning, it can flexibly adapt to complex and ever-changing operational needs, plan more accurate and reasonable paths, and finally generate executable control commands, effectively improving the operational efficiency of the target autonomous vehicle.

[0041] Please see Figure 2 Another embodiment of the vehicle planning method based on manual guidance in this invention includes: 201. Obtain route data corresponding to at least one route point manually marked on the monitoring platform; 202. Read map data related to the area where at least one waypoint is located from the preset map database, and obtain real-time traffic data and vehicle status data of the target autonomous vehicle in the area surrounding at least one waypoint. Map data provides basic information about regional roads, real-time traffic data reflects the current road conditions, and vehicle status data includes the vehicle's own operating status.

[0042] 203. Use spatiotemporal data fusion algorithms to preprocess map data in order to construct a spatiotemporal map model; The system extracts road node and edge information from map data. Node information includes the location of road intersections, and edge information includes the road's connectivity and length. Based on a preset time period division rule, the day is divided into multiple time periods, and historical traffic data for each time period is obtained. Based on the road node and edge information and historical communication data for each time period, a spatiotemporal map model is constructed using a spatiotemporal data fusion algorithm. The spatiotemporal map model includes the road topology, which contains the dynamic characteristics of roads across multiple time periods. Using vector analysis tools, road layers in the map are identified and extracted, transforming roads into vector lines composed of a series of discrete points. For road intersection locations in the node information, GIS spatial analysis functions are used to detect the intersection points of vector lines. Combined with the map's coordinate system, the latitude and longitude 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, for the acquired vector road lines, GIS distance measurement tools are used to calculate the actual length between adjacent nodes along the road centerline to ensure the accuracy of the length data. Graph theory algorithms are used to treat each node as a vertex in the graph and road lines as edges. By analyzing the connections between vertices, a road connection matrix is ​​constructed to present the connection status of each road with other roads, thereby completely extracting the node and edge information of the roads.

[0043] Based on preset time-segmentation rules, a day is divided into multiple time periods. Specifically, considering the cyclical patterns of traffic flow and actual needs, the day can be initially divided into 24 time periods, with each hour as the basic unit. For special periods with drastic traffic flow changes, such as morning and evening rush hours, further subdivisions into 15-minute or 30-minute segments are possible. When acquiring historical road traffic data for each time period, on the one hand, a legitimate, secure, and stable interface connection can be established with the traffic management department's data system to retrieve key data such as vehicle flow statistics, average speed, number of congestion occurrences, and duration from its massive database. On the other hand, intelligent traffic sensors deployed at key road nodes, such as important intersections and bridges, can be used to collect real-time information on vehicle license plates and passage times. By filtering, organizing, and analyzing this raw data by time period, historical communication data such as the actual number of vehicles passing through and traffic density can be obtained for each time period.

[0044] An initial static road topology is constructed using road node information as vertices and edge information, represented as a graph. Nodes record the coordinates of key locations such as road intersections, while edges record road connections and fixed lengths. For historical traffic data across different time periods, cluster analysis from data mining is used to classify the data according to characteristics such as traffic flow and average speed, extracting typical traffic patterns for each time period, such as congestion patterns during peak hours and smooth traffic patterns during off-peak hours. A spatiotemporal interpolation algorithm is then employed to interpolate traffic patterns between adjacent time periods based on the time intervals and traffic pattern variations. Data interpolation is performed to fill data gaps, making the traffic data continuous and smooth in the time dimension. Then, the processed traffic data for each time period is integrated with the static topology structure. By assigning dynamic attributes for different time periods to each edge, such as travel time and congestion probability in different time periods, the static road topology structure is expanded into a spatiotemporal topology structure containing dynamic features of multiple time periods. Finally, 3D visualization technology is used to display the road topology in a 2D 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, constructing an intuitive spatiotemporal map model with rich information.

[0045] Existing technologies often rely solely on static map data for route planning, failing to reflect the dynamic characteristics of roads over time, leading to 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. By rationally dividing time periods and acquiring historical traffic data for each period, it comprehensively understands the traffic patterns of roads at different times. Furthermore, using a spatiotemporal data fusion algorithm, it deeply integrates the static road structure with dynamic traffic data, constructing 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 traffic data, vehicle status data and spatiotemporal map model, use the spatiotemporal graph neural network ST-GNN to perform data fusion analysis, and extract the influence parameters of road features, traffic condition influencing factors and vehicle current status on the accessibility of each waypoint. The input dataset is constructed by integrating at least one waypoint data, real-time traffic data, vehicle status data, and relevant data from the spatiotemporal map model. The input dataset is then fed into the spatiotemporal graph neural network ST-GNN, which processes the input dataset layer by layer through its multi-layer network structure to obtain the output structure. Based on the output structure, road features, traffic condition influencing factors, and the influence parameters of the current vehicle status on the accessibility of each waypoint are extracted.

[0047] The data from at least one waypoint is preprocessed and converted into a unified coordinate format. For real-time traffic data, key indicators such as congestion levels, average vehicle speeds, and accident occurrences are parsed for each road segment. Vehicle status data includes the vehicle's current location, speed, remaining battery / fuel level, and direction of travel. Data such as road topology and traffic patterns at different times related to the area where at least one waypoint is located are extracted from the spatiotemporal map model. These data from different sources are then concatenated and correlated according to a specific data structure to construct a structured input dataset, ensuring a clear correspondence between each data element.

[0048] The constructed input dataset is fed into a pre-trained spatiotemporal graphical neural network (ST-GNN). This network has a multi-layered structure, including an input layer, multiple hidden layers, and an output layer. The input layer receives the input dataset and, based on the functional characteristics of the neurons, rationally distributes data such as waypoint coordinates, real-time traffic indicators, vehicle status information, road topology, and traffic patterns to each neuron. The hidden layers receive the data distributed by the input layer and perform deep processing on the data using nonlinear transformations. Utilizing spatiotemporal convolution operations, based on the characteristics of the input data—including the spatial relationships between different road segments, the relative positions of waypoints and roads, and the temporal changes in real-time traffic conditions and differences in traffic patterns across different time periods—it captures the correlation between the data in both spatiotemporal dimensions, uncovering the implicit spatiotemporal patterns and rules. Furthermore, through an attention mechanism, different weights are assigned to each feature based on its importance to the accessibility of each waypoint, highlighting key features and allowing the network to focus more on factors with a significant impact on accessibility. After processing by the hidden layers, the data is transformed into more representative and discriminative feature representations. Finally, the output layer integrates the feature data processed by the hidden layer to produce the output structure. Based on the output structure, specific analytical algorithms are used to extract the road features, road condition influencing factors, and parameters affecting the accessibility of each waypoint using the vehicle's current state. The influencing parameters corresponding to road features include road curvature, slope, and width; the influencing parameters corresponding to road condition influencing factors include congestion level and accident status; and the influencing parameters corresponding to the vehicle's current state include remaining battery / fuel level and current speed.

[0049] Compared to traditional solutions that rely on a single data source or simply pile up different data, this solution deeply integrates waypoint, real-time traffic, vehicle status data, and spatiotemporal map model-related data to construct a structured input dataset. This allows different types of data to complement and corroborate each other, comprehensively depicting traffic scenarios from multiple dimensions such as spatial location, road characteristics, real-time traffic, and vehicle status. Furthermore, by utilizing the multi-layer structure of the spatiotemporal graph neural network ST-GNN, the solution rationally allocates data, captures spatiotemporal patterns, highlights key features, and mines the inherent correlations within the data. Based on the output structure, it accurately extracts the parameters affecting the accessibility of waypoints, greatly improving the accuracy and comprehensiveness of the analysis results.

[0050] 205. Based on the initial priority and corresponding influence parameters in the data of each waypoint, assign priority weights to the nodes corresponding to each waypoint and generate constraint nodes with target priorities. Determine the initial weight values ​​corresponding to different initial priorities; analyze the influence parameters of each waypoint; for unfavorable factors affecting the accessibility of waypoints, adjust the initial weight values ​​of the corresponding waypoints by decreasing their impact on accessibility; for favorable factors, adjust the initial weight values ​​of the corresponding waypoints by increasing their improvement on accessibility; assign the adjusted weights to the nodes corresponding to each waypoint to generate constraint nodes with target priorities, where a higher priority weight indicates a higher importance of the node in path planning.

[0051] The initial priorities are mapped to initial weight values, for example, high, medium, and low priorities are mapped to 0.8, 0.5, and 0.2, respectively. The influence parameters of each waypoint are analyzed. Regarding road characteristics, if the road curvature of a waypoint is too large or the slope is extremely steep, these unfavorable factors will seriously hinder vehicle passage. Based on the degree to which these factors reduce vehicle speed and increase energy consumption, the initial weight value corresponding to that waypoint is reduced by a certain proportion. Conversely, if the road is flat and wide, it is considered a favorable factor, and based on its degree of improvement in vehicle driving efficiency and safety, the initial weight value is increased accordingly. For road condition influencing factors, if the road sections surrounding a waypoint are severely congested and accidents are frequent, the initial weight value is reduced based on the impact of congestion duration and accident impact range on accessibility. If the road conditions are unobstructed, the initial weight value is increased based on the improved traffic efficiency. Regarding the vehicle's current status, if the vehicle has sufficient remaining battery / fuel to ensure a smooth arrival at the waypoint, the initial weight value is increased based on the enhanced driving safety. If the battery / fuel is insufficient, the initial weight value is decreased based on the degree of risk that may affect arrival. Finally, the adjusted weights are assigned to the nodes corresponding to each waypoint, generating constraint nodes with target priorities.

[0052] Compared to existing technologies that use fixed weights or simple hierarchical priorities, this solution uses dynamic quantitative analysis of multi-dimensional influencing parameters such as road characteristics, real-time traffic conditions, and vehicle status to establish a coupling mechanism between initial priorities and dynamic adjustment factors. This allows waypoint weights to adapt to changes in actual traffic scenarios, significantly improving the arrival rate of high-priority nodes and the rationality of path planning in testing.

[0053] 206. Perform hybrid path planning based on the constraint nodes to obtain the initial path; Using a graph search-based path planning algorithm, a search graph is constructed starting from the current position of the target autonomous vehicle and combining it with constraint nodes. Based on the search graph, a path search is performed according to the target priority and traffic constraint attributes of each constraint node. When a target point that satisfies the conditions of all constraint nodes is found, the initial path is determined based on the nodes traversed during the search process.

[0054] Using the current position of the target autonomous vehicle as the starting point for path planning, constraint nodes are integrated into a graph structure. A complete search graph is constructed based on the actual road connectivity, where each edge represents a passable road and is assigned a corresponding travel cost, such as distance or estimated travel time. The travel constraint attributes of each constraint node are also defined, such as whether U-turns are allowed or whether passage is permitted during specific time periods. 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 priority are given higher exploration weights. The comprehensive evaluation value of each node is dynamically calculated based on the edge travel costs to guide the search direction. The search range is gradually expanded according to this evaluation mechanism, continuously exploring possible paths. When a target point satisfying all constraint node conditions is found, the sequence of nodes traversed during the search is backtracked, and these nodes are connected in the search order to determine the initial path. Constraint node conditions include, for example, the requirement to pass through specific priority nodes and satisfying the travel constraints of each node.

[0055] Existing path planning technologies often fail to fully consider the priority differences of key nodes and the complex and ever-changing traffic constraints in real-world traffic scenarios. This results in planned paths that may not meet actual driving needs, leading to insufficient safety and efficiency. In contrast, this solution utilizes a graph search-based path planning algorithm. Starting from the current position of the autonomous vehicle, it constructs a search graph incorporating constraint nodes. This algorithm accurately reflects the actual environment and searches for paths based on node priorities and traffic constraint attributes. It can reasonably balance key factors and 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, differentiate the must-reach points and detour points, and perform local replanning through a dynamic optimization algorithm to obtain the optimized path; 208. Generate control commands that can be executed by the target autonomous vehicle based on the optimized path to control the operation of the target autonomous vehicle.

[0057] In this embodiment of the invention, manually labeled waypoint data is acquired and integrated with multi-source data such as maps, real-time traffic conditions, and vehicle status. A spatiotemporal map model is constructed using a spatiotemporal data fusion algorithm. The spatiotemporal graph neural network ST-GNN is used to accurately extract parameters affecting the accessibility of each waypoint, thereby generating constraint nodes with reasonable priorities. This allows path planning to fully consider various key factors in actual operations. Through hybrid path planning, differentiated adjustment, and local replanning using dynamic optimization algorithms, the system can flexibly cope with complex and ever-changing operating environments, continuously optimize the path, and ultimately generate executable control commands. This effectively improves the accuracy and rationality of path planning for the target autonomous vehicle, enhances its adaptability to different scenarios, and thus improves operational efficiency and safety.

[0058] Please see Figure 3 Another embodiment of the vehicle planning method based on manual guidance in this invention includes: 301. Obtain route data corresponding to at least one route point manually marked on the monitoring 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 constraint nodes with target priority. 303. Perform hybrid path planning based on the constraint nodes to obtain the initial path; 304. Identify the mandatory and detour points in the initial path. Mandatory points are manually marked mandatory route points, and detour points are manually marked prohibited areas. The system reads manually marked waypoint information from the storage structure to identify the distinctive features of mandatory and detour points. For example, mandatory points have specific mandatory route markers, while detour points have prohibited area markers. The initial path is transformed into a sequence of continuous coordinate points. Using a coordinate matching algorithm, the path coordinate points are compared with the geographic coordinate range marked by the mandatory points. If a path coordinate point falls within the coordinate range of a mandatory point, it is identified as a mandatory point, and its associated attributes are recorded. For detour points, a polygon region judgment algorithm is used to check whether the sequence of path coordinate points intersects with the polygon region formed by the boundary coordinates of the detour points. If there is an intersection, the region is identified as a detour point, and its related attributes are recorded. This completes the identification of mandatory and detour 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 detour range of the detour point is expanded or reduced according to the real-time traffic conditions. For high-priority first-reachable points, if the distance from the vehicle's current location to the first-reachable point is greater than a preset distance and the estimated arrival time is greater than a preset time range, the arrival time window is extended. If the distance from the vehicle's current location to the first-reachable point is not greater than a preset distance and the estimated arrival time is less than a preset time range, the arrival time window is compressed. For low-priority second-reachable points, the arrival time window is dynamically widened or tightened based on the vehicle's current travel progress and remaining path length. Specifically, for low-priority second-reachable points, the vehicle's current location is obtained in real time through the positioning device, and the percentage of distance already traveled is calculated based on the pre-planned path to determine the current travel progress. Simultaneously, the remaining path length is calculated based on the path planning algorithm. A correlation model between travel progress, remaining path length, and time consumption is established based on big data analysis. Based on this model and the vehicle's real-time speed, the arrival time window is dynamically estimated. The remaining time to reach the second mandatory destination is compared 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 expected at this progress based on historical data and current road conditions, it is determined that the vehicle is unlikely to arrive within the initial time window. In this case, the arrival time window is extended later to widen 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 expected remaining length, it indicates that the vehicle is likely to arrive earlier. The arrival time window is compressed forward according to the corresponding ratio to tighten it, thereby achieving flexible adjustment of the arrival time window for the second mandatory destination. For each detour point, when the real-time traffic conditions show that the traffic around the detour point is congested, its detour range is expanded; when the real-time traffic conditions show that the traffic around the detour point is smooth, its detour range is narrowed.

[0060] For example, in a mining operation scenario, a target autonomous mining transport vehicle is tasked with transporting mined ore from the mining site to the crushing station. Among them, crushing station A, which is close to the main production area and has an extremely urgent need for ore, is set as the high-priority first destination. The preset distance is 3 kilometers, and the preset arrival time range is 10-15 minutes. After loading ore, the target autonomous vehicle starts and departs. After driving for a period of time, relying on its own positioning system and the positioning base stations deployed inside the mine, it calculates that the distance from the target vehicle to crushing station A is 4 kilometers, which is greater than the preset distance. At the same time, the target autonomous vehicle obtains information through onboard sensors and real-time interaction with the mine dispatch center. It learns that there are other vehicles intersecting and many curves on the roads inside the mine. Based on its own driving speed, the estimated arrival time is 18 minutes, which is greater than the preset time range. Therefore, the time window for arriving at crushing station A is extended to 20 minutes. As the target autonomous vehicle continues to move towards crushing station A, the calculation is performed again and it is found that the distance from the vehicle to the point has become 2 kilometers, which is not greater than the preset distance, and the estimated arrival time has become 8 minutes, which is less than the preset time range. Therefore, the arrival time window is compressed to 9 minutes.

[0061] In addition, some areas within the mine need to be avoided due to safety hazards or ongoing equipment maintenance. When the real-time monitoring system of the autonomous vehicle shows that large equipment is moving materials around one of the avoidance points, causing traffic difficulties and a "congestion" situation, the vehicle's control system immediately expands the range of this avoidance point, increasing the original avoidance radius from 100 meters to 200 meters. Conversely, when the safety hazard around another avoidance point is eliminated, the equipment maintenance is completed, and the road is reopened, the control system reduces the avoidance range of that point from 100 meters to 50 meters.

[0062] Existing technologies often employ fixed patterns when handling arrival time windows for must-reach points and detour ranges for avoidance points. This makes them inflexible and unable to adapt to complex and ever-changing real-world traffic conditions and vehicle status, potentially leading to unreasonable planning and impacting driving efficiency and safety. In contrast, this solution dynamically adjusts the arrival time window for must-reach points based on their target priority and the vehicle's current status. This more accurately reflects the actual driving conditions, ensuring high-priority tasks are completed on time while rationally planning the time for low-priority tasks. By expanding or shrinking the detour range based on real-time traffic conditions, it effectively avoids congested road sections and makes full use of unobstructed roads. This significantly improves the flexibility and adaptability of path planning, enhancing the efficiency and safety of autonomous vehicles.

[0063] 306. Using a dynamic optimization algorithm, with the adjusted time window of the reachable 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 for the must-reach point and the detour range for the avoidance point are used as constraints for the model. The primary optimization objective is to minimize travel time, while the secondary optimization objective is to minimize path length, thus constructing a dynamic optimization model. The dynamic optimization algorithm is used to solve the dynamic optimization model, and the optimal path scheme is found under the constraint conditions. Based on the solved path scheme, the initial path is locally adjusted to generate the optimized path, which satisfies the arrival requirements of the must-reach point and the detour requirements of the avoidance point.

[0065] The adjusted time window for reachable points is transformed into upper and lower time limits, and the avoidance range for detour points is defined as a spatial constraint in the form of a set of geographic coordinates. The shortest travel time is the primary optimization objective, assigned a high weight coefficient to reflect its priority. The shortest path length is set as a secondary optimization objective with a relatively low weight. The aforementioned time and spatial constraints are integrated into the model. An improved genetic algorithm is used as the dynamic optimization algorithm. During population initialization, multiple mutated path individuals are generated based on the initial path. During iteration, individuals are selected based on the fitness function, and crossover and mutation operations are used to continuously generate new individuals. An elite retention strategy ensures that high-quality individuals are not eliminated. Finally, under the premise of satisfying all constraints, the optimal path scheme that achieves the optimization objective is obtained. Based on this scheme, the initial path is locally adjusted, such as replanning the order of passing through reachable points and connecting new road segments that bypass detour points, thereby generating an optimized path that satisfies both the reachable point arrival time requirement and the detour point avoidance requirement.

[0066] Existing route planning technologies typically use fixed settings for the time window of the must-reach point and the range of detour points. The optimization objective is singular and lacks dynamic consideration of actual traffic conditions, resulting in generated routes that fail to balance efficiency and actual needs. In contrast, this solution uses these dynamically adjusted parameters as constraints, constructs a dynamic optimization model with the shortest travel time as the primary goal and the shortest route length as a secondary goal, and solves the model using a dynamic algorithm. This generates more realistic, efficient, and reasonable optimized routes that meet both the must-reach and detour requirements, significantly improving the quality of route planning.

[0067] 307. Generate control commands that can be executed by the target autonomous vehicle based on the optimized path to control the operation of the target autonomous vehicle.

[0068] In this embodiment of the invention, manually labeled waypoint data is acquired, and constraint nodes with target priorities are generated by combining multi-source data to plan an initial path. Must-reach points and detour points are identified, and their arrival time windows are dynamically adjusted based on the priority of must-reach points and vehicle status. The range of detour points is flexibly changed according to real-time road conditions. Then, using these as constraints, a dynamic optimization algorithm is used to locally replan the path, and finally, executable instructions are generated to control the vehicle. This fully considers the dynamic changes in human guidance and actual operation, making the path planning fit the actual needs, effectively improving the ability of autonomous vehicles to cope with complex scenarios, enhancing the rationality and flexibility of path planning, and thus improving operational efficiency and safety.

[0069] The above describes the vehicle planning method based on manual guidance in the embodiments of the present invention. The following describes the apparatus for the vehicle planning method based on manual guidance in the embodiments of the present invention. Please refer to [link / reference]. Figure 4 One embodiment of the vehicle planning method apparatus based on manual guidance in this invention includes: The acquisition module 401 is used to acquire route point data corresponding to at least one route point marked by a person on the monitoring platform; The generation module 402 is used to generate constraint nodes with target priority based on at least one waypoint data, combined with map data, real-time traffic data and vehicle status data, using a spatiotemporal data fusion algorithm. Processing module 403 is used to perform hybrid path planning based on constraint nodes to obtain an initial path; The optimization module 404 is used to differentiate the must-reach points and detour points based on the initial path, and to perform local replanning through a dynamic optimization algorithm to obtain the optimized path; The control module 405 is used to generate control commands executable by the target autonomous vehicle based on the optimized path to control the operation of the target autonomous vehicle.

[0070] In this embodiment of the invention, by acquiring manually marked waypoint data and integrating it into path planning, it can quickly respond to and accurately execute manual guidance. By combining multi-source data and using spatiotemporal data fusion algorithms to generate constraint nodes and perform hybrid path planning, and then using differentiated adjustment and dynamic optimization algorithms for local replanning, it can flexibly adapt to complex and ever-changing operational needs, plan more accurate and reasonable paths, and finally generate executable control commands, effectively improving the operational efficiency of the target autonomous vehicle.

[0071] Please see Figure 5 Another embodiment of the vehicle planning device based on manual guidance in this invention includes: The acquisition module 401 is used to acquire route point data corresponding to at least one route point marked by a person on the monitoring platform; The generation module 402 is used to generate constraint nodes with target priority based on at least one waypoint data, combined with map data, real-time traffic data and vehicle status data, using a spatiotemporal data fusion algorithm. Processing module 403 is used to perform hybrid path planning based on constraint nodes to obtain an initial path; The optimization module 404 is used to differentiate the must-reach points and detour points based on the initial path, and to perform local replanning through a dynamic optimization algorithm to obtain the optimized path; The control module 405 is used to generate control commands executable by the target autonomous vehicle based on the optimized path to control the operation of the target autonomous vehicle.

[0072] Optionally, the generation module 402 includes: The acquisition unit 4021 is used to read map data related to the area where at least one waypoint is located from a preset map database, and to acquire real-time traffic data and vehicle status data of the target autonomous vehicle in the area surrounding at least one waypoint. Construction unit 4022 is used to preprocess map data using a spatiotemporal data fusion algorithm to construct a spatiotemporal map model; Extraction unit 4023 is used to extract data based on at least one waypoint data, real-time traffic data, vehicle status data and spatiotemporal map model, and to perform data fusion analysis using spatiotemporal graph neural network ST-GNN to extract road features, traffic influencing factors and vehicle current status on the accessibility of each waypoint. The generation unit 4024 is used to assign priority weights to the nodes corresponding to each path point based on the initial priority and corresponding influence parameters in the data of each path point, and generate constraint nodes with priority.

[0073] Optionally, building block 4022 can be specifically used for: The system extracts road node and edge information from map data. Node information includes the location of road intersections, and edge information includes the road's connectivity and length. Based on a preset time period division rule, the day is divided into multiple time periods, and historical traffic data for each time period is obtained. Based on the road node and edge information and historical communication data for each time period, a spatiotemporal map model is constructed using a spatiotemporal data fusion algorithm. The spatiotemporal map model includes the road topology, which contains the dynamic characteristics of roads across multiple time periods.

[0074] Optionally, the extraction unit 4023 can be specifically used for: The input dataset is constructed by integrating at least one waypoint data, real-time traffic data, vehicle status data, and relevant data from the spatiotemporal map model. The input dataset is then fed into the spatiotemporal graph neural network ST-GNN, which processes the input dataset layer by layer through its multi-layer network structure to obtain the output structure. Based on the output structure, road features, traffic condition influencing factors, and the influence parameters of the current vehicle status on the accessibility of each waypoint are extracted.

[0075] Optionally, the generating unit 4024 can be specifically used for: Determine the initial weight values ​​corresponding to different initial priorities; analyze the influence parameters of each waypoint; for unfavorable factors affecting the accessibility of waypoints, adjust the initial weight values ​​of the corresponding waypoints by decreasing their impact on accessibility; for favorable factors, adjust the initial weight values ​​of the corresponding waypoints by increasing their improvement on accessibility; assign the adjusted weights to the nodes corresponding to each waypoint to generate constraint nodes with target priorities, where a higher priority weight indicates a higher importance of the node in path planning. Optionally, the processing module 403 can be specifically used for: Using a graph search-based path planning algorithm, a search graph is constructed starting from the current position of the target autonomous vehicle and combining it with constraint nodes. Based on the search graph, a path search is performed according to the target priority and traffic constraint attributes of each constraint node. When a target point that satisfies the conditions of all constraint nodes is found, the initial path is determined based on the nodes traversed during the search process.

[0076] Optionally, optimization module 404 includes: The identification unit 4041 is used to identify the mandatory points and detour points in the initial path. The mandatory points are manually marked mandatory route points, and the detour points are manually marked prohibited areas. The adjustment unit 4042 is used to dynamically adjust the arrival time window of the must-reach point based on the target priority of the must-reach point and the current status of the vehicle, and to expand or shrink the detour range of the detour point according to the real-time road conditions. Replanning unit 4043 is used to perform local replanning on the initial path using a dynamic optimization algorithm, with the adjusted time window of the reachable point and the avoidance range of the avoidance point as constraints, to obtain the optimized path.

[0077] Optionally, the adjustment unit 4042 can be specifically used for: For high-priority first must-reach point, if the distance from the vehicle's current location to the first must-reach point is greater than a preset distance and the estimated arrival time is greater than a preset time range, the arrival time window will be extended; if the distance from the vehicle's current location to the first must-reach point is not greater than a preset distance and the estimated arrival time is less than a preset time range, the arrival time window will be compressed. For low-priority second must-reach point, the arrival time window will be dynamically widened or tightened based on the vehicle's current travel progress and remaining path length. For each detour point, when real-time traffic conditions show that the area around the detour point is congested, its detour range will be expanded; when real-time traffic conditions show that the area around the detour point is clear, its detour range will be narrowed.

[0078] Optionally, the replanning unit 4043 can be specifically used for: The adjusted time window for the must-reach point and the detour range for the avoidance point are used as constraints for the model. The primary optimization objective is to minimize travel time, while the secondary optimization objective is to minimize path length, thus constructing a dynamic optimization model. The dynamic optimization algorithm is used to solve the dynamic optimization model, and the optimal path scheme is found under the constraint conditions. Based on the solved path scheme, the initial path is locally adjusted to generate the optimized path, which satisfies the arrival requirements of the must-reach point and the detour requirements of the avoidance point.

[0079] In this embodiment of the invention, manually labeled waypoint data is acquired, and a spatiotemporal map model is constructed using a spatiotemporal data fusion algorithm combined with multi-source data. Key influencing parameters are extracted using a spatiotemporal graph neural network, thereby generating constraint nodes with reasonable priorities. This fully considers various complex factors in actual operations, employs a graph search-based algorithm for hybrid path planning, and identifies must-reach points and detour points. The time window for must-reach points and the range of detour points are dynamically adjusted based on various conditions, and the path is locally replanned using a dynamic optimization algorithm as a constraint. This makes the path planning more realistic and flexible in adapting to changes, and finally generates executable instructions to control vehicle operation. This effectively improves the accuracy, rationality, and flexibility of path planning for autonomous vehicles, enhances the ability to cope with complex scenarios, and improves operational efficiency and safety.

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

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

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

[0083] The memory 601 may include high-speed random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 603 (which can be wired or wireless), such as the Internet, wide area network, local area network, metropolitan area network, etc. The bus 602 may be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 6 The symbol is represented by a single double-headed arrow, 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. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of the processor 600 or by instructions in software form. The 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 methods, steps, and logic block diagrams disclosed in the embodiments of this disclosure. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this disclosure can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules may reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory 601. Processor 600 reads the information in memory 601 and, in conjunction with its hardware, completes the method steps of the aforementioned embodiment.

[0085] The present invention also provides an electronic device, the computer device including a memory and a processor, the memory storing computer-readable instructions, which, when executed by the processor, cause the processor to perform the steps of the vehicle planning method based on human guidance in the above 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, wherein the computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steps of the human-guided vehicle planning method.

[0087] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing 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, in essence, or the part 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 to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0089] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for vehicle planning based on artificial guidance, characterized in that, The vehicle planning method based on human guidance includes: Obtain route point data corresponding to at least one route point manually marked on the regulatory 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 priority. Based on the constraint nodes, perform hybrid path planning to obtain the initial path; Based on the initial path, the must-reach points and detour points are adjusted differently, and local replanning is performed through a dynamic optimization algorithm to obtain the optimized path; Based on the optimized path, control commands executable by the target autonomous vehicle are generated to control the operation of the target autonomous vehicle; Each waypoint data includes the waypoint's geographic coordinates, initial priority, and traffic constraint attributes. The step of generating priority-bearing constraint nodes based on at least one waypoint data, combined 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 surrounding area of ​​the at least one waypoint and vehicle status data of the target autonomous vehicle; preprocessing the map data using a spatiotemporal data fusion algorithm to construct a spatiotemporal map model; performing data fusion analysis using a spatiotemporal graph neural network (ST-GNN) based on the at least one waypoint data, the real-time traffic data, the vehicle status data, and the spatiotemporal map model, extracting the influence parameters of road features, traffic condition influencing factors, and the current vehicle status on the accessibility of each waypoint; and assigning priority weights to the nodes corresponding to each waypoint according to the initial priority and corresponding influence parameters in the waypoint data, thereby generating priority-bearing constraint nodes.

2. The human-guided vehicle planning method of claim 1, wherein, The step of preprocessing the map data using a spatiotemporal data fusion algorithm to construct a spatiotemporal map model includes: The node information and edge information of the roads are extracted 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 preset time period division rules, a day is divided into multiple time periods, and historical traffic data of roads within each time period is obtained; Based on the node information, edge information, and 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, which contains the dynamic features of the road over multiple time periods.

3. The vehicle planning method based on manual guidance according to claim 1, characterized in that, The method, based on at least one waypoint data, the real-time traffic data, vehicle status data, and the spatiotemporal map model, utilizes a spatiotemporal graph neural network (ST-GNN) for data fusion analysis to extract road features, traffic condition influencing factors, and parameters affecting the accessibility of each waypoint, including: The input dataset is constructed by integrating the data of at least one waypoint, the real-time traffic data, the vehicle status data, and the relevant data in the spatiotemporal map model. The input dataset is fed into the spatiotemporal graphical neural network ST-GNN, and its multi-layer network structure processes the input dataset layer by layer to obtain the output structure. Based on the output structure, road features, road condition influencing factors, and the influence parameters of the vehicle's current state on the accessibility of each waypoint are extracted.

4. The vehicle planning method based on manual guidance according to claim 1, characterized in that, Based on the priority and influence parameters of each waypoint, priority weights are assigned to the nodes corresponding to each waypoint, generating 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 waypoints, adjust the initial weight value of the corresponding waypoints by decreasing their impact on accessibility. For favorable factors, adjust the initial weight value of the corresponding waypoints by increasing their improvement on accessibility. The adjusted weights are assigned to the nodes corresponding to each waypoint, generating constraint nodes with target priorities. The higher the priority weight, the more important the node is in the path planning.

5. The vehicle planning method based on manual guidance according to claim 1, characterized in that, The process of performing hybrid path planning based on the constraint nodes to obtain the initial path includes: Using a graph search-based path planning algorithm, a search graph is constructed starting from the current position of the target autonomous vehicle and combining it with the constraint nodes; Based on the search graph, path search is performed according to the target priority and access constraint attributes of each constraint node; When a target point that satisfies all the constraints of the nodes is found, the initial path is determined based on the nodes traversed during the search.

6. The vehicle planning method based on manual guidance according to claim 1, characterized in that, Based on the initial path, the required points and detour points are differentiated and adjusted, and a local replanning is performed through a dynamic optimization algorithm to obtain the optimized path, including: Identify the must-reach points and detour points in the initial path. The must-reach points are manually marked mandatory route points, and the detour points are manually marked prohibited areas. 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 detour range of the detour point is expanded or reduced according to real-time road conditions. A dynamic optimization algorithm is used to perform local replanning on the initial path, with the adjusted time window of the reachable point and the avoidance range of the avoidance point as constraints, to obtain the optimized path.

7. The vehicle planning method based on manual guidance according to claim 6, characterized in that, The method of dynamically adjusting the arrival time window of the must-reach point based on the target priority and the current state of the vehicle, and expanding or shrinking the detour range of the bypass point according to real-time traffic conditions, includes: For the first must-reach point with high priority, if the distance from the vehicle's current location to the first must-reach point is greater than a preset distance and the estimated arrival time is greater than a 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 a preset distance and the estimated arrival time is less than a preset time range, the arrival time window will be compressed forward. For low-priority second must-reach points, the arrival time window is dynamically widened or tightened based on the vehicle's current travel progress and remaining path length. For each detour point, when real-time traffic conditions show that the area around the detour point is congested, the detour range is expanded; when real-time traffic conditions show that the area around the detour point is clear, the detour range is reduced.

8. The vehicle planning method based on manual guidance according to claim 6, characterized in that, The method employs a dynamic optimization algorithm, using the adjusted time window for reachable points and the avoidance range for avoidance points as constraints, to perform local replanning on the initial path, resulting in an optimized path, including: The adjusted time window for the must-reach point and the bypass range for the bypass point are used as constraints of the model. The shortest travel time is the primary optimization objective, and the shortest path length is the secondary optimization objective, in order to construct a dynamic optimization model. The dynamic optimization model is solved using a dynamic optimization algorithm, and the optimal path scheme is found under the condition that the constraints are satisfied. Based on the solution obtained, the initial path is locally adjusted to generate an optimized path. The optimized path satisfies the requirements for reaching the must-reach point and avoiding the detour point.

9. A vehicle planning device based on manual guidance, characterized in that, The vehicle planning device based on human guidance includes: The acquisition module is used to acquire route point data corresponding to at least one route point marked by humans on the regulatory platform; The generation module is used to generate constraint nodes with target priorities based on at least one waypoint data, combined with map data, real-time traffic data and vehicle status data, using a spatiotemporal data fusion algorithm. The processing module is used to perform hybrid path planning based on the constraint nodes to obtain an initial path; The optimization module is used to differentiate the must-reach points and detour points based on the initial path, and to perform local replanning through a dynamic optimization algorithm to obtain the optimized path; The control module is used to generate control commands executable by the target autonomous vehicle based on the optimized path to control the operation of the target autonomous vehicle. Each waypoint data includes the geographical coordinates, initial priority, and traffic constraint attributes of the waypoint. The generation module includes: an acquisition unit, used to read map data related to the area where the at least one waypoint is located from a preset map database, and to acquire real-time traffic data and vehicle status data of the target autonomous vehicle in the surrounding area of ​​the at least one waypoint; a construction unit, used to preprocess the map data using a spatiotemporal data fusion algorithm to construct a spatiotemporal map model; an extraction unit, used to extract the influence parameters of road features, traffic condition influencing factors, and vehicle current status on the accessibility of each waypoint based on the 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 for data fusion analysis; and a generation unit, used to assign priority weights to the nodes corresponding to each waypoint according to the initial priority and corresponding influence parameters in the data of each waypoint, and generate constraint nodes with priority.

10. An electronic device, characterized in that, The electronic device includes: a memory and at least one processor, wherein the memory stores instructions; The at least one processor invokes the instructions in the memory to cause the electronic device to execute the human-guided vehicle planning method as described in any one of claims 1-8.

11. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement the vehicle planning method based on human guidance as described in any one of claims 1-8.

Images