Global path planning algorithm and system for agv with steering constraint

Abstract

The application relates to the technical field of path planning, in particular to an AGV global path planning algorithm and system with steering constraints, which comprises the following steps: acquiring planning data to determine candidate path nodes and a candidate steering relationship table, acquiring vehicle constraint data to determine a starting point state and a maximum steering angle, screening out candidate steering states that do not satisfy the maximum steering angle, calculating a steering sweep area, generating environment occupation data, road selectivity constraints and road unobstructed values, constructing a state path graph and outputting global path instructions. In the application, the vehicle heading is limited, the vehicle body turning sweep space and the road topology unobstructed state are incorporated into the global search state, the non-steering unobstructed state and the non-selectable road section are excluded in the global stage, the candidate paths are sorted according to the steering constraint result, the road unobstructed value and the road length, the consistency of the global path and the position heading control is improved, and the turning re-planning and parking waiting are reduced.

Description

Technical Field

[0001] This invention relates to the field of path planning technology, and in particular to a global path planning algorithm and system for AGVs with steering constraints. Background Technology

[0002] Path planning technology involves determining routes for mobile devices in known or perceptible environments based on starting point, target location, environmental map, road conditions, obstacle distribution, and vehicle movement conditions. It primarily supports autonomous driving, navigation control, task execution, and traffic management for mobile devices, and is widely applied in industries such as warehousing and logistics, intelligent manufacturing, factory handling, inspection operations, port transportation, and unmanned delivery. Traditional AGV global path planning algorithms refer to path planning methods that determine the overall driving route of automated guided vehicles within a work area, identifying a global path between the starting point and target point that meets the driving conditions of the automated guided vehicle.

[0003] Existing automated guided vehicles (AGVs) rely primarily on road node connectivity and road length for global path planning within the operational area. Changes in vehicle orientation and sweep range at turning nodes are not included in the search process. Temporary road occupancy and predicted obstacle occupancy are difficult to incorporate into path selection. Once the path reaches the turning area, insufficient turning space and discontinuous heading adjustments can easily occur. Frequent slowdowns or replanning in local control can lead to increased vehicle stops and scheduling resource consumption, resulting in decreased turnaround efficiency and impacting the continuity of mission passage and scheduling stability. Summary of the Invention

[0004] To address the technical problems existing in the prior art, embodiments of the present invention provide an AGV global path planning algorithm and system with steering constraints. The technical solution is as follows: On the one hand, a global path planning algorithm for AGVs with steering constraints is provided, including the following steps: S1: Obtain planning data, determine candidate path nodes based on the planning data, and generate a candidate turning relationship table according to the connection direction of the candidate path nodes; S2: Obtain vehicle constraint data, determine the starting state, entry direction, departure direction and maximum steering angle based on the vehicle constraint data and the candidate steering relationship table, calculate the heading amplitude for the candidate steering states, and filter out the candidate steering states whose heading amplitude is greater than the maximum steering angle; S3: For the candidate turning states after screening, calculate the turning sweep area based on the vehicle constraint data, the entering direction and the leaving direction, generate environmental occupancy data based on the planning data, determine the turning margin and passage mark according to the occupancy relationship between the turning sweep area and the environmental occupancy data, and form the turning constraint result from the turning margin and the passage mark; S4: Generate road traffic attributes based on the planning data, determine road selectivity constraints and road smoothness values ​​based on the road traffic attributes, construct a stateful path graph by combining the candidate path nodes, the turning constraint results and the road smoothness values, eliminate the states that cannot be turned and road segments that do not meet the road selectivity constraints, sort by the turning constraint results, the road smoothness values ​​and the road length, and output global path instructions.

[0005] On the other hand, the process of acquiring the planning data and generating the candidate turning relationship table is as follows: Acquire road centerline data, road boundary data, road intersection location data, road direction change location data, and task docking location data, and form an initial node set based on the endpoints of the road centerline data and the road intersection location data; Nodes in the initial node set that overlap with the road direction change location data and the task stop location data are retained. Duplicate nodes below the node spacing threshold are merged to determine the candidate path nodes. The candidate turning relationship table is established according to the access direction and exit direction of the center line of the adjacent road of the candidate path node.

[0006] On the other hand, the process of determining the candidate turning state is as follows: The vehicle's current position, heading, geometric parameters, and maximum steering angle are obtained. The vehicle's current position is mapped to the starting position in the candidate path nodes. The vehicle's heading is compared with the entry direction in the candidate steering relationship table. The entry direction with the smallest direction difference is determined as the entry direction corresponding to the starting state. The entry direction is combined with the departure direction corresponding to the same candidate path node to generate the candidate steering state.

[0007] On the other hand, the calculation process for the heading amplitude is as follows: Read the entry and exit direction angles in the candidate turning state, adjust the entry and exit direction angles to the same angle measurement range, calculate the minimum direction change required for the entry direction angle to turn to the exit direction angle, and determine the minimum direction change as the heading amplitude.

[0008] On the other hand, the calculation process for the turning sweep area is as follows: Obtain the entry direction and exit direction corresponding to the candidate turning state after filtering. Determine the offset relationship between the vehicle's outer corner point and the vehicle's center point based on the vehicle constraint data. Calculate the trajectory of the vehicle's outer corner point according to the heading change process from the entry direction to the exit direction. Connect the outer boundary and inner boundary of the trajectory of the vehicle's outer corner point to generate the turning sweep area.

[0009] On the other hand, the calculation process of the trajectory of the vehicle's outer corner points is as follows: The lower limit of the turning radius is determined based on the wheelbase and maximum steering angle in the vehicle constraint data. The turning sampling angle interval is determined according to the heading amplitude corresponding to the candidate steering state. The position of the vehicle's outer corner point is calculated point by point along the arc direction corresponding to the lower limit of the turning radius. The continuous positions of the vehicle's outer corner points are associated with the trajectory of the vehicle's outer corner points.

[0010] On the other hand, the process of forming the steering constraint result is as follows: The turning sweep area is spatially overlapped with the road boundary occupancy area, fixed obstacle occupancy area, and predicted moving obstacle occupancy area in the environmental occupancy data. When the turning sweep area crosses the road boundary occupancy area or overlaps with the fixed obstacle occupancy area, a non-turning passage mark is determined. When the turning sweep area overlaps with the predicted moving obstacle occupancy area, a restricted passage mark is determined. The turning margin is determined based on the distance between adjacent boundaries to form the turning constraint result.

[0011] On the other hand, the process of determining the road selectability constraint is as follows: Extract road segment status records connected to the candidate path nodes from the planning data, read the road segment occupancy status, traffic direction, available road width, predicted occupancy status, and number of task vehicles, compare the road segment occupancy status with the occupancy threshold to determine occupancy restrictions, compare the available road width with the vehicle passage width to determine width restrictions, compare the predicted occupancy status with the predicted occupancy threshold to determine prediction restrictions, and generate the road selectivity constraints from the occupancy restrictions, width restrictions, and prediction restrictions.

[0012] On the other hand, the process of determining the road traffic flow value is as follows: Obtain the status records of the road segments that satisfy the road selectivity constraints; determine the road occupancy evaluation value based on the road segment vacancy ratio; determine the direction consistency evaluation value based on the matching relationship between the traffic direction and the vehicle driving direction; determine the width margin evaluation value based on the difference between the available road width and the vehicle traffic width; determine the task occupancy evaluation value based on the number of task vehicles; and perform a weighted summation of the road occupancy evaluation value, the direction consistency evaluation value, the width margin evaluation value, and the task occupancy evaluation value to generate the road smoothness value. The output process of the global path instruction is as follows: Obtain the candidate path ranked first in the sorted candidate global path sequence, extract the candidate path node, the entry direction, the departure direction, the steering margin, the passage marker, and the road accessibility value corresponding to the candidate path, convert the candidate path node into a position control point, convert the departure direction into a heading control variable, convert the steering margin and the passage marker into a speed control condition, associate the position control point, the heading control variable, and the speed control condition, and output the global path instruction.

[0013] On the other hand, a global path planning system for AGVs with steering constraints is provided. This system is applied to the global path planning algorithm for AGVs with steering constraints, including: The node table building module obtains planning data, determines candidate path nodes based on the planning data, and generates a candidate turning relationship table according to the connection direction of the candidate path nodes. The steering filtering module acquires vehicle constraint data, determines the starting state, entry direction, departure direction and maximum steering angle based on the vehicle constraint data and the candidate steering relationship table, calculates the heading amplitude for candidate steering states, and filters out candidate steering states whose heading amplitude is greater than the maximum steering angle. The sweep judgment module calculates the turning sweep area based on the vehicle constraint data, the entry direction and the departure direction for the candidate turning states after screening, generates environmental occupancy data based on the planning data, determines the turning margin and passage mark according to the occupancy relationship between the turning sweep area and the environmental occupancy data, and forms the turning constraint result from the turning margin and the passage mark. The path output module generates road traffic attributes based on the planning data, determines road selectivity constraints and road accessibility values ​​based on the road traffic attributes, constructs a stateful path graph by combining the candidate path nodes, the turning constraint results, and the road accessibility values, removes the states that cannot be turned and road segments that do not meet the road selectivity constraints, sorts the path by the turning constraint results, the road accessibility values, and the road length, and outputs global path instructions.

[0014] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: By mapping planning data and vehicle constraint data to a candidate steering relationship table, the heading amplitude is first compared with the maximum steering angle constraint, and the steering sweep area is then compared with the environmental occupancy data for occupancy judgment, forming a steering constraint result expressed by steering margin and traffic markers. Road traffic attributes are transformed into road selectability constraints and road accessibility values, so that the state-based path map simultaneously reflects vehicle steerability, road space accessibility, and topological accessibility. During the path search stage, non-steerable traffic states and unselectable road segments can be eliminated in advance. The candidate global path sequence is sorted according to the steering constraint result, road accessibility value, and road length. The output global path command fits the position and heading control requirements, reducing replanning at turns and parking waiting. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart of the main steps of the present invention; Figure 2 This is a detailed flowchart of the candidate path nodes and candidate turning relationship table of the present invention; Figure 3 This is a detailed flowchart of the candidate turning state screening process for the present invention. Figure 4 This is a detailed flowchart of the process for determining the occupancy of the sweeping area in this invention; Figure 5 This is a detailed flowchart of the state-based path graph construction and global path instruction output of the present invention; Figure 6 This is a system flowchart of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0019] This invention provides an AGV global path planning algorithm and system with steering constraints. During the actual operation of the AGV performing handling tasks along the warehouse, production line or factory transportation road, the road centerline, road boundary, intersection position, turning position, parking position and road occupancy status continuously participate in the global path generation. Before entering the road node, the vehicle needs to meet the road passability condition and the vehicle turning capability condition at the same time, so that the generated global path can not only connect the starting point and the ending point, but also form executable entry direction, exit direction, turning margin and speed control conditions at each turning position.

[0020] In AGV global path planning, when paths are generated solely based on road node connectivity, the paths are geometrically connected. However, when vehicles turn at intersections, curves, or near obstacles, issues arise such as turning angles exceeding the vehicle's turning capabilities, vehicle outlines sweeping road boundaries, spatial overlap with obstacle-occupied areas, and entry into congested or inverse road sections. This embodiment integrates candidate path nodes, candidate turning relationships, vehicle turning capabilities, turning sweep areas, environmental occupancy data, road traffic attributes, and road flow values ​​into a stateful path graph. This ensures that each path state includes node position and direction status, and performs joint elimination of turning feasibility and road selectivity before global path output.

[0021] Please see Figure 1 and Figure 2 S1: Obtain planning data, determine candidate path nodes based on the planning data, and generate a candidate turning relationship table according to the connection direction of the candidate path nodes.

[0022] The planning data refers to the dataset used to describe the roads that AGVs can travel on and their traffic status. This planning data includes road centerline data, road boundary data, road intersection data, road direction change data, task stopping location data, and road segment status records. Road centerline data describes the reference travel line when the AGV travels along the road; road boundary data describes the lateral road restrictions that the vehicle's outline cannot cross; intersection data describes the node positions formed by the connection of multiple roads; direction change data describes the locations where the direction changes on straight roads, curved roads, or bends; task stopping location data describes the task positions of the AGV for picking up and placing goods, waiting, or aligning; and road segment status records describe the current or predicted traffic status of the road segment.

[0023] Candidate path nodes are location nodes that can serve as the global path search status. Their sources include road centerline endpoints, road intersections, road direction change locations, and task stopping locations. Candidate path nodes are not arbitrary sampling points on the road centerline, but rather nodes that can influence path topology, vehicle direction changes, or task execution. When determining candidate path nodes, road segment endpoints are first read from the road centerline data, and then these endpoints are merged with road intersection data to form an initial node set. Nodes in the initial node set that overlap with road direction change location data are retained, allowing road direction changes to be identified in subsequent steering decisions. Nodes that overlap with task stopping location data are also retained, enabling task start and end points to participate in the path search status output. Nodes with duplicate locations or locations with a distance less than the node spacing threshold are merged. After merging, the node positions that represent road connectivity are retained, and the road connectivity relationships of the merged nodes are aggregated into the same candidate path node.

[0024] In one embodiment, when the road centerline data is generated from road boundary data or passable area data, the passable area of ​​the AGV is first determined based on the road boundary. Then, a road skeleton located in the middle of the passable area is generated using a Voronoi diagram, and the adjacency relationship between the road skeleton and adjacent road boundaries is determined through Delaunay triangulation. Subsequently, the skeleton points in the road skeleton corresponding to road intersections, road direction change locations, and task stopping locations are determined as candidate path nodes, so that the candidate path nodes can reflect the road topology connection relationship and retain the road geometric changes that affect vehicle steering.

[0025] The node spacing threshold is a preset criterion used to determine whether two candidate nodes are duplicate nodes. It is derived from the accuracy of road map production, the allowable deviation of AGV positioning, and road topology editing rules. This threshold is written into the map rule item of the planning data as a configuration field. The field content includes the node merging criterion, merging priority, and inheritance rules of the connection relationship after merging. When there are multiple nodes in the map library that are close in location but connected by different roads, it is first checked whether their road connection directions can be merged. If the connection directions are consistent or can be inherited, they are merged into one candidate path node. If the connection directions conflict and correspond to different road levels or different traffic areas, they are retained as different candidate path nodes, and their optional status is handled in the subsequent road optionality constraints.

[0026] The candidate turning relationship table is a data table that records the combination relationships of entry and exit directions using candidate path nodes as indexes. When generating the candidate turning relationship table, for each candidate path node, the centerline segments of the roads adjacent to it are read. An entry direction is generated based on the direction the road centerline segments connect to the node, and an exit direction is generated based on the direction the road centerline segments detach from the node. Both entry and exit directions are determined by the tangential driving direction of the road centerline at the node or the pointing relationship of adjacent road segments. If a road segment has one-way traffic, only entry and exit directions that match the traffic direction are retained; if a road segment allows two-way traffic, entry and exit relationships for each direction are generated separately. For each set of entry and exit directions under the same candidate path node, the node identifier, connecting road segment identifier, exit road segment identifier, entry direction angle, exit direction angle, traffic direction attribute, and initial available state are written to form the candidate turning relationship table for subsequent turning filtering.

[0027] In one embodiment, the planning data undergoes validity verification before being used to generate candidate path nodes. Validity verification includes checking for continuous endpoints of the road centerline, whether the road boundary can enclose the corresponding road segment, whether the intersection falls within the traversable road area, whether the direction change location can be associated with at least one road centerline segment, and whether the task stopping location can be mapped to a candidate path node or an adjacent road segment. When a road segment lacks a road boundary, the road segment is assigned a boundary pending confirmation state and its selection priority is reduced in subsequent road selectability constraints. When the task stopping location cannot be mapped to a candidate path node, a candidate path node located on the same road segment and satisfying the travel direction is selected as an alternative start / end state, and the positional offset description between the stopping location and the alternative node is retained for final path instruction conversion.

[0028] In this embodiment, candidate path nodes and candidate turning relationship tables are generated by planning data, so that the road connectivity relationship is transformed from a simple location connection to a turning connection with directional status, thereby enabling subsequent vehicle turning capability screening and sweep judgment to directly apply to the combination of entry and exit directions at the node.

[0029] Please see Figure 1 and Figure 3 S2: Obtain vehicle constraint data, determine the starting state, entry direction, departure direction and maximum steering angle based on the vehicle constraint data and candidate steering relationship table, calculate the heading amplitude for candidate steering states, and filter out candidate steering states with heading amplitude greater than the maximum steering angle.

[0030] Vehicle constraint data refers to a set of data describing the AGV's motion capabilities and geometric constraints. This data includes the vehicle's current position, heading, geometric parameters, wheelbase, passage width, the offset of the vehicle's corner points relative to its center point, and the maximum steering angle. The current position determines the starting point; the heading determines the vehicle's directional state upon entering the starting node; the geometric parameters calculate the steering sweep area; the wheelbase and maximum steering angle determine the turning boundaries the vehicle can perform under the current constraints; and the passage width is used to determine road width limitations.

[0031] The starting state refers to the initial search state formed after the vehicle's current position is mapped to candidate path nodes. When determining the starting state, the vehicle's current position is first matched with candidate path nodes. If the vehicle's current position falls within the matching range of a candidate path node, that candidate path node is determined as the starting position. If the vehicle's current position is located in a road segment connecting two candidate path nodes, the candidate path node ahead of the vehicle's travel direction and conforming to the road's traffic direction is taken as the starting position, and the road segment relationship from the vehicle's current position to that candidate path node is recorded as the initial access relationship. Subsequently, the vehicle's heading is read and compared with the entry direction registered in the candidate steering relationship table. The entry direction with the smallest direction difference and conforming to the road's traffic direction is determined as the entry direction corresponding to the starting state.

[0032] A candidate turning state refers to the state to be judged at the same candidate path node, formed by the combination of an entry direction and a departure direction. When generating a candidate turning state, the candidate path node in the candidate turning relationship table is used as an index to read the available entry direction set and departure direction set under that node. For the starting position, the entry direction is based on the vehicle's heading; for non-starting positions, the entry direction comes from the direction state when the previous road segment arrived at the node. Each entry direction is combined with a departure direction that can be connected to the same candidate path node to form a candidate turning state, and the node identifier, entry direction, departure direction, connecting road segment, exiting road segment, and state source are recorded.

[0033] Heading amplitude refers to the minimum directional change required for a vehicle to turn from the entering direction to the leaving direction. When calculating heading amplitude, the entering and leaving direction angles of the candidate turning state are read. First, the two angles are adjusted to the same angular measurement range. Then, it is determined which directional change is smaller, the clockwise or counterclockwise change, when turning from the entering direction to the leaving direction. The smaller directional change is taken as the heading amplitude for that candidate turning state. If the entering and leaving directions correspond to the same straight-ahead road condition, the heading amplitude is recorded as the straight-ahead state. If the leaving direction requires the vehicle to make a U-turn or cross a prohibited road direction, the heading amplitude is recorded as the restricted state and determined by the maximum turning angle.

[0034] The maximum steering angle is a configuration field in the vehicle constraint data that represents the upper limit of the AGV's steering execution capability. It originates from vehicle motion control parameters, vehicle steering mechanism calibration results, or vehicle operation strategy configuration. When filtering candidate steering states, the heading amplitude of each candidate steering state is compared with the allowable steering conditions corresponding to the maximum steering angle. When the heading amplitude exceeds the allowable steering conditions, the candidate steering state is deleted and will not proceed to the sweep judgment. When the heading amplitude is within the allowable steering conditions, the candidate steering state is retained and added to the set of states to be swept. For straight-ahead states, they can be directly added to the retained states. For U-turn states, they will only proceed to the subsequent judgment if both the vehicle constraint data and road traffic attributes allow U-turns.

[0035] In one embodiment, when the vehicle's current position, heading, or geometric parameters are missing, the system first checks whether the vehicle's direction state in the previous valid path instruction is still consistent with the road segment corresponding to the current position. If they are consistent, the previous valid direction state is used as the direction input for the starting point state. If they are inconsistent, the entry direction under the candidate starting point node that is consistent with the road traffic direction and has the least steering restriction is used as the initialization direction, and this starting point state is marked as a pending confirmation state. The pending confirmation state can still enter path generation, but it is converted into a restricted traffic condition in the speed control conditions when the path instruction is output, so that the vehicle reduces the steering execution requirements and waits for position and orientation updates before entering the next control stage.

[0036] In this embodiment, by mapping the vehicle's current position and heading to the starting state, and filtering out candidate turning states with the maximum steering angle, the global path search excludes direction combinations that the vehicle's movement capability cannot execute before entering the road passage judgment, thereby reducing the number of objects to be processed in subsequent sweep judgments and making the output path directionally executable.

[0037] Please see Figure 1 and Figure 4 S3: For the candidate steering states after screening, calculate the steering sweep area based on vehicle constraint data, entry direction and exit direction, generate environmental occupancy data based on planning data, determine the steering margin and passage mark according to the occupancy relationship between the steering sweep area and the environmental occupancy data, and form the steering constraint result from the steering margin and passage mark.

[0038] The steering sweep area refers to the spatial area traversed by the outer corner points of the AGV when it turns from the entry direction to the exit direction at a candidate path node. The steering sweep area is used to determine whether the vehicle will cross the road boundary or overlap with the area occupied by obstacles when performing this steering state. When calculating the steering sweep area, the entry and exit directions corresponding to the candidate steering state are first read, and then the offset relationship between the vehicle's outer corner points and the vehicle's center point is read from the vehicle constraint data. The offset relationship between the vehicle's outer corner points and the vehicle's center point refers to the positional relationship of the vehicle's front, rear, left, and right outer corner points in the vehicle's own coordinate system. This relationship is determined by the vehicle's geometric parameters and participates in the sweep generation as the vehicle's outer boundary during path planning.

[0039] The vehicle's outer corner trajectory refers to the set of positions traversed sequentially by each outer corner point during a turn. When determining the vehicle's outer corner trajectory, the lower limit of the vehicle's usable turning radius is determined based on the wheelbase and maximum steering angle in the vehicle constraint data. This lower limit of the turning radius originates from the vehicle motion constraint configuration and represents the turning capability boundary that the vehicle must not be less than under the current turning capability. Subsequently, the turning sampling angle interval is determined according to the heading amplitude corresponding to the candidate turning state. The turning sampling angle interval originates from the path planning accuracy configuration and the road map geometric accuracy configuration and is used to control the dispersion of the outer corner trajectory. Along the arc direction corresponding to the lower limit of the turning radius, the vehicle's center point attitude is determined point by point according to the heading change sequence from the entry direction to the exit direction. Under each attitude, the position of the vehicle's outer corner points is determined based on the outer corner point offset relationship. The continuous outer corner point positions are then associated sequentially according to the corresponding corner points to form the trajectory of each outer corner point.

[0040] When generating the turning sweep area, the trajectories located outside the vehicle's turning point are connected to form the outer boundary, and the trajectories located inside the vehicle's turning point are connected to form the inner boundary. The starting outer boundary of the entering direction and the ending outer boundary of the leaving direction are included as part of the closed boundary, forming a complete sweep area. For straight-ahead driving, the sweep area is generated according to the outer envelope of the vehicle's outline along the road centerline entering and leaving nodes. For small-turn driving, the outer and inner trajectories are connected according to the actual heading change process. For restricted U-turn driving, the sweep area is only generated if the previous steps are preserved and the road attributes allow it.

[0041] Environmental occupancy data refers to the spatial occupancy set converted from planning data, including at least road boundary occupancy areas, fixed obstacle occupancy areas, and predicted mobile obstacle occupancy areas. Road boundary occupancy areas are generated from road boundary data and describe the outer areas of the road beyond which vehicles are not permitted to cross. Fixed obstacle occupancy areas are generated from fixed obstacle records, road facility occupancy records, or impassable area records in the planning data. Predicted mobile obstacle occupancy areas are formed from the predicted occupancy status in the road segment status records and describe areas where other vehicles, temporary obstructions, or dynamic occupancy objects may affect turning traffic during the path planning cycle. Environmental occupancy data is aligned with the road boundaries of candidate path nodes during generation to ensure that the turning sweep area and occupancy area use the same map coordinate system.

[0042] In one embodiment, the predicted occupancy area for moving obstacles is generated based on the obstacle's historical location, current direction of movement, road segment, and predicted occupancy status of the road segment. For obstacles moving continuously along a road segment, an LSTM-based trajectory prediction rule is used to read their historical trajectory sequence, and the road areas they may subsequently pass through are determined according to the trajectory change trend. When multiple moving obstacles exist around the same candidate turning state, a spatiotemporal attention processing rule is used to distinguish the degree of influence of each moving obstacle on the current turning state, and obstacles with a stronger spatiotemporal correlation with the current turning sweep area are preferentially written into the predicted occupancy area for moving obstacles. The thus generated predicted occupancy area for moving obstacles is used as part of the environmental occupancy data for subsequent access marking and turning margin determination.

[0043] Traffic markers are status fields that record the turning traffic attributes of candidate turning states. Traffic markers include permitted turning, restricted turning, and prohibited turning. When determining traffic markers, the turning sweep area is spatially overlapped with the road boundary occupancy area, the fixed obstacle occupancy area, and the predicted moving obstacle occupancy area. When the turning sweep area crosses the road boundary occupancy area, a prohibited turning traffic marker is determined; when the turning sweep area overlaps with the fixed obstacle occupancy area, a prohibited turning traffic marker is determined; when the turning sweep area overlaps with the predicted moving obstacle occupancy area but does not cross the road boundary and does not overlap with the fixed obstacle occupancy area, a restricted turning traffic marker is determined; when the turning sweep area does not form a prohibitive overlap with the aforementioned occupancy areas, a permitted turning traffic marker is determined.

[0044] Steering margin refers to the remaining available space between the steering sweep area and the adjacent road boundary or obstacle-occupied area. When determining steering margin, the spacing relationships between adjacent boundaries in the outer and inner boundaries of the steering sweep area and the environmental occupancy data are read, and the spacing relationships are converted into margin levels according to the margin level mapping rules in the road map rules. The sources of margin levels include vehicle passage width, road boundary safety distance configuration, obstacle avoidance strategy, and steering execution stability configuration. Steering margin levels are used for subsequent path ranking and speed control condition generation: when the margin level meets normal passage conditions, subsequent path ranking continues to compare based on road accessibility and road length; when the margin level is under restricted passage conditions, the corresponding candidate path can be retained but its ranking priority is reduced, and speed limit or cautious passage conditions are generated in the final path instruction; when the margin level is under prohibited passage conditions, the corresponding candidate steering state is eliminated.

[0045] In one embodiment, when a candidate turning state is determined to be a restricted passage marker, the speed obstacle method is used to determine the unsuitable speed range for the vehicle in this turning state based on the relative positional relationship between the predicted occupied area of ​​the moving obstacle and the turning sweep area. Then, combining the vehicle's current speed state, maximum steering angle, steering margin level, and road traffic attributes, a dynamic window method is used to select a speed state that meets the turning passage conditions within the allowable speed range. The selected speed state does not change the connection relationship of the candidate path nodes; instead, it is written as a speed control condition into the turning constraint result and associated with the position control point and heading control quantity when the global path command is output.

[0046] The steering constraint result refers to the state judgment result composed of steering margin and passage marker. When forming the steering constraint result, a node identifier, entry direction, exit direction, sweep area index, passage marker, steering margin level, and anomaly description are written for each candidate steering state. If the road boundary data is missing, causing the sweep area to be unable to complete the boundary judgment, the candidate steering state is marked as restricted passage, and the boundary missing state is written into the anomaly description; if the predicted occupied area of ​​a moving obstacle is missing, the basic passage marker is determined only based on the road boundary occupied area and the fixed obstacle occupied area, and the dynamic occupancy unconfirmed state is written into the anomaly description; if the candidate steering state has been filtered out by the maximum steering angle, no steering constraint result is generated.

[0047] In this embodiment, a steering sweep area is generated by the trajectory of the vehicle's outer corner points, and its occupancy relationship is determined with the predicted occupancy areas of the road boundary, fixed obstacles, and moving obstacles. This allows the steering action at the path node to form a state result of passable, restricted, or impassable during the global path generation stage, so that the subsequent state-based path map can eliminate unexecutable steering and retain steering states that are restricted but can be passed at adjustable speed.

[0048] Please see Figure 1 and Figure 5 S4: Generate road traffic attributes based on planning data, determine road selectivity constraints and road smoothness values ​​based on road traffic attributes, construct a stateful path graph by combining candidate path nodes, turning constraint results and road smoothness values, eliminate states that cannot be turned and road segments that do not meet road selectivity constraints, sort by turning constraint results, road smoothness values ​​and road length, and output global path instructions.

[0049] Among them, road access attributes refer to the set of attributes extracted from planning data and used to determine whether road segments are available for path search and how they participate in ranking. Road access attributes include road segment occupancy status, traffic direction, available road width, predicted occupancy status, and number of task vehicles. Road segment occupancy status describes whether the road segment is currently occupied or partially occupied; traffic direction describes the direction in which vehicles are allowed to travel on the road segment; available road width describes the effective width within the road boundary that allows AGVs to pass; predicted occupancy status describes the dynamic occupancy status that the road segment may experience during path execution; and the number of task vehicles describes the task load status of vehicles currently using or about to use the road segment.

[0050] Road selectivity constraints refer to the rules for determining whether a road segment is allowed to participate in the global path search. When determining road selectivity constraints, the status records of road segments connected to candidate path nodes are first extracted from the planning data. Then, the road segment occupancy status, traffic direction, available road width, predicted occupancy status, and number of task vehicles are read in a preset order. The occupancy status of a road segment is compared with the occupancy judgment conditions corresponding to the occupancy threshold to form an occupancy restriction. The occupancy threshold is derived from road management rules and scheduling strategy configurations and is used to distinguish between road segments that are accessible, have restricted access, and are prohibited from entering. The available road width is compared with the width judgment conditions corresponding to the vehicle passage width to form a width restriction. The vehicle passage width is derived from vehicle constraint data, and the width judgment conditions are derived from road boundary safety configurations. The predicted occupancy status is compared with the predicted occupancy judgment conditions corresponding to the predicted occupancy threshold to form a prediction restriction. The predicted occupancy threshold is derived from a dynamic occupancy prediction strategy and is used to determine whether there is an unsuitable occupancy trend for a road segment during path execution. The occupancy restriction, width restriction, and prediction restriction together generate the road selectivity constraints. When any restriction reaches the prohibited entry state, the road segment is not included in the state-based path graph; when the restriction is the restricted entry state, the road segment is retained but written into the restricted access attribute; when all restrictions meet the entry conditions, the road segment participates in the path search as an optional road segment.

[0051] The road accessibility value is an evaluation field used to describe the relative smoothness of traffic flow on road segments that meet road selectivity constraints. When generating the road accessibility value, the status records of road segments that meet the road selectivity constraints are first read, and then road occupancy evaluation value, direction consistency evaluation value, width margin evaluation value, and task occupancy evaluation value are generated respectively. The road occupancy evaluation value is determined by the level rules corresponding to the road segment's idle ratio; the more complete the idle state, the more favorable the road occupancy evaluation is for path ranking. The direction consistency evaluation value is determined by the matching relationship between the traffic direction and the vehicle's driving direction; the evaluation is more favorable when the vehicle's driving direction is consistent with the road's allowed direction, and becomes restricted when there are directional restrictions or when waiting for right-of-way is required. The width margin evaluation value is determined by the remaining traffic status between the available road width and the vehicle's traffic width; the evaluation is more favorable when the remaining traffic status meets the road boundary safety configuration, and becomes restricted when it approaches the restriction conditions. The task occupancy evaluation value is determined by the occupancy level corresponding to the number of task vehicles; the more task vehicles a road segment undertakes, the more likely it is to select other road segments that meet the conditions during ranking. The weights of each evaluation value are derived from the path planning strategy configuration. The weight configuration is set in the order of road safety priority, traffic direction matching priority, congestion avoidance priority, and path length coordination priority, and is uniformly maintained by road management rules. When generating road traffic flow values, the above evaluation results are summarized according to the weight configuration to obtain the road traffic flow status used for ranking.

[0052] In one embodiment, the road traffic flow value can also be corrected by combining the path traffic flow prediction results of the graph neural network. During processing, candidate path nodes are treated as graph nodes, adjacent road segments are treated as graph connections, and road segment occupancy status, traffic direction, available road width, predicted occupancy status, and number of vehicles on duty are used as input fields for nodes or connections. The graph neural network propagates the traffic impact of adjacent road segments according to the road topology connections, enabling road segments adjacent to congested road segments, predicted occupancy road segments, or road segments with concentrated vehicle traffic to obtain corresponding traffic flow correction results. The corrected road traffic flow value is entered into the stateful path graph and serves as one of the criteria for ranking candidate paths.

[0053] A stateful path graph is a path search data structure that simultaneously includes candidate path nodes, road segment connections, entry and exit directions, turning constraint results, and road accessibility values. When constructing a stateful path graph, candidate path nodes are treated as position states, entry and exit directions from the candidate turning relationship table are treated as direction states, road segments between adjacent candidate path nodes are treated as connecting edges, turning constraint results are written to the node turning state, and road accessibility values ​​are written to the road segment connection state. For turning constraint results representing states where turning is not permitted, these are directly deleted from the stateful path graph. For road segments that do not meet road selectivity constraints, their connections are deleted. Restricted passage markers and restricted road segments are retained in the stateful path graph but set to low priority during sorting and converted into speed control conditions when outputting path instructions.

[0054] In one embodiment, when generating a candidate global path sequence in the stateful path graph, an improved A* algorithm is used to perform global path search. During the search process, candidate path nodes are treated as position states, entry and exit directions as direction states, and road length, road accessibility, and turning constraints are used as evaluation criteria for path expansion. A dynamic weighting heuristic adjusts the search priority based on the road connectivity between the current search state and the target state, road accessibility, and turning constraints. When the search state is close to the target state and road accessibility is stable, the expansion priority towards the target direction is increased. When there are restricted access markers, predicted occupancy states, or insufficient turning margins around the search state, the expansion priority of detour directions or accessible road segments is increased. This ensures that path search not only progresses based on road distance but also simultaneously considers turning feasibility and road accessibility.

[0055] In one embodiment, when there are multiple road intersections between the starting and ending states, a bidirectional A* search rule is used to expand the path from both the starting and ending states. When the search states on both sides meet at the same candidate path node, the same road segment, or a location with a connectable turning relationship, the search results from both sides are concatenated according to the entry and exit directions, and the turning constraint results are called again to verify the turnable traffic status of the concatenated location. For road segments with continuous centerlines, minimal direction changes, and no task stopping locations, a JPS jump point search rule is used to skip intermediate locations that do not affect road selection and turning judgment, retaining only road intersections, road direction change locations, task stopping locations, road traffic flow value change locations, and turning constraint result change locations as path expansion nodes.

[0056] When sorting and outputting global path instructions, a sequence of candidate global paths is first generated in the stateful path graph. Each candidate global path consists of a candidate path node, a corresponding road segment, an entry direction, an exit direction, a steering constraint result, and a road accessibility value. Sorting is performed by comparing the steering constraint result, road accessibility value, and road length in that order. The steering constraint result is prioritized to exclude paths that are unexecutable or highly restricted; the road accessibility value is used to select road segments with more stable traffic conditions among candidate paths where steering is feasible; and the road length is used to select a path with a more suitable travel distance after the steering and road accessibility conditions are met. If multiple traffic restriction markers exist in a candidate path, the path with fewer traffic restriction markers and a more stable steering margin level is prioritized. If a candidate path has a short road length but contains non-steering traffic conditions, it is not included in the output path. If a candidate path has a good road accessibility value but its starting state is inconsistent with the vehicle's current heading, the starting state matching is re-executed or a pending confirmation state is output.

[0057] Global path instructions refer to the path execution data that the AGV uses for subsequent navigation control. When outputting global path instructions, the first candidate path in the sorted global path sequence is retrieved, and the candidate path node, entry direction, exit direction, turning margin, passage marker, and road accessibility value corresponding to that path are extracted. The candidate path nodes are converted into position control points, which record the road node positions the vehicle should reach in the path sequence. The exit direction is converted into a heading control variable, which indicates the target driving direction when the vehicle leaves the current candidate path node. The turning margin and passage marker are converted into speed control conditions, which constrain the vehicle's speed at turning nodes or restricted road sections. Finally, the position control points, heading control variable, and speed control conditions are correlated and output, enabling subsequent navigation control to continuously execute the global path according to node positions, exit directions, and passage restrictions.

[0058] In one embodiment, when no output candidate paths are available after the stateful path graph has been eliminated, a path failure state is generated according to the anomaly handling sequence. First, it is checked whether the lack of candidate paths is due to a failure in matching the starting point state direction. If so, the process returns to S2 to redetermine the starting point's entry direction. Second, it is checked whether all road segments are eliminated due to overly strict road selectivity constraints. If so, restricted road segments are read and a waiting or detour state is generated. Third, it is checked whether the path is restricted because the turning sweep area overlaps with the predicted occupied area of ​​a moving obstacle. If so, the original path search result is retained and a waiting for replanning state is output. If the stateful path graph cannot be generated due to missing road boundary, road centerline, or vehicle constraint data, a planning data correction state is output. None of the above abnormal states are output as normal global path instructions, but rather as trigger conditions for replanning or waiting for scheduling.

[0059] In this embodiment, road segments unsuitable for search are eliminated by road selectability constraints, and the selectable road segments are sorted by road accessibility values. The steering constraint results and road lengths are used together in path selection, so that the global path command simultaneously includes position control points, heading control quantities, and speed control conditions. This ensures that the global planning results are consistent with the AGV's steering execution capability, road accessibility, and dynamic occupancy status.

[0060] Please see Figure 6 A global path planning system for AGVs with steering constraints, the system includes: The node table creation module acquires planning data, determines candidate path nodes based on the planning data, and generates a candidate steering relationship table according to the connection direction of the candidate path nodes. The input content of the node table creation module corresponds to the planning data in S1 of the embodiment, and its output content is candidate path nodes and a candidate steering relationship table, which is then passed to the steering filtering module. The node table creation module does not undertake the vehicle steering capability judgment; its functional boundaries are limited to the node-based processing of road centerlines, road boundaries, intersections, direction change locations, and task stopping locations, as well as the recording of relationships between entry and exit directions.

[0061] The steering filtering module acquires vehicle constraint data and determines the starting point state, entry direction, departure direction, and maximum steering angle based on the vehicle constraint data and the candidate steering relationship table. It calculates the heading amplitude for each candidate steering state and filters out those with a heading amplitude greater than the maximum steering angle. The input to the steering filtering module includes the candidate steering relationship table output by the node table building module, as well as the vehicle's current position, heading, vehicle geometry parameters, and maximum steering angle. Its output is the filtered candidate steering states, which is then passed to the sweep judgment module. The steering filtering module does not generate road accessibility values ​​or output final path instructions.

[0062] The sweep judgment module calculates the sweeping area based on vehicle constraint data, entry direction, and exit direction for the filtered candidate turning states. It generates environmental occupancy data based on planning data, determines the turning margin and passage markers according to the occupancy relationship between the sweeping area and the environmental occupancy data, and forms the turning constraint result from the turning margin and passage markers. The inputs of the sweep judgment module correspond to the candidate turning states, vehicle outline corner point offset relationships, road boundary occupancy areas, fixed obstacle occupancy areas, and predicted moving obstacle occupancy areas in S3 of the embodiment. Its output is the turning constraint result, which is then passed to the path output module. The sweep judgment module does not change the topological connection relationships of the candidate path nodes; it only determines whether the turning state at the node is passable, restricted, or impassable.

[0063] The path output module generates road traffic attributes based on planning data, determines road selectivity constraints and road accessibility values ​​based on these attributes, constructs a stateful path graph by combining candidate path nodes, turning constraint results, and road accessibility values, eliminates states where turning is not possible and road segments that do not meet road selectivity constraints, sorts the path by turning constraint results, road accessibility values, and road length, and outputs global path instructions. The input to the path output module includes candidate path nodes, turning constraint results, road segment status records, and road traffic attributes; its output is a global path instruction containing position control points, heading control values, and speed control conditions. The path output module follows the processing rules of S4 in the aforementioned embodiment, enabling the various modules of the system to collaboratively achieve AGV global path planning with turning constraints in the order of node table creation, turning filtering, sweep judgment, and path output.

[0064] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A global path planning algorithm for AGVs with steering constraints, characterized by the following steps: include: Obtain planning data, determine candidate path nodes based on the planning data, and generate a candidate turning relationship table according to the connection direction of the candidate path nodes; Obtain vehicle constraint data, determine the starting state, entry direction, departure direction and maximum steering angle based on the vehicle constraint data and the candidate steering relationship table, calculate the heading amplitude for the candidate steering states, and filter out the candidate steering states whose heading amplitude is greater than the maximum steering angle. For the candidate turning states after screening, the turning sweep area is calculated based on the vehicle constraint data, the entering direction and the leaving direction, the environmental occupancy data is generated based on the planning data, the turning margin and the passage mark are determined according to the occupancy relationship between the turning sweep area and the environmental occupancy data, and the turning constraint result is formed by the turning margin and the passage mark. Based on the planning data, road traffic attributes are generated. Based on the road traffic attributes, road selectivity constraints and road smoothness values ​​are determined. A stateful path graph is constructed by combining the candidate path nodes, the turning constraint results, and the road smoothness values. The states that cannot be turned and road segments that do not meet the road selectivity constraints are eliminated. The path is sorted by the turning constraint results, the road smoothness values, and the road length, and a global path instruction is output.

2. The AGV global path planning algorithm with steering constraints according to claim 1, characterized in that, The specific processes for obtaining the planning data and generating the candidate turning relationship table are as follows: Acquire road centerline data, road boundary data, road intersection location data, road direction change location data, and task docking location data, and form an initial node set based on the endpoints of the road centerline data and the road intersection location data; Nodes in the initial node set that overlap with the road direction change location data and the task stop location data are retained. Duplicate nodes below the node spacing threshold are merged to determine the candidate path nodes. The candidate turning relationship table is established according to the access direction and exit direction of the center line of the adjacent road of the candidate path node.

3. The AGV global path planning algorithm with steering constraints according to claim 1, characterized in that, The process of determining the candidate turning state is as follows: The vehicle's current position, heading, geometric parameters, and maximum steering angle are obtained. The vehicle's current position is mapped to the starting position in the candidate path nodes. The vehicle's heading is compared with the entry direction in the candidate steering relationship table. The entry direction with the smallest direction difference is determined as the entry direction corresponding to the starting state. The entry direction is combined with the departure direction corresponding to the same candidate path node to generate the candidate steering state.

4. The AGV global path planning algorithm with steering constraints according to claim 1, characterized in that, The calculation process for the heading amplitude is as follows: Read the entry and exit direction angles in the candidate turning state, adjust the entry and exit direction angles to the same angle measurement range, calculate the minimum direction change required for the entry direction angle to turn to the exit direction angle, and determine the minimum direction change as the heading amplitude.

5. The AGV global path planning algorithm with steering constraints according to claim 1, characterized in that, The calculation process for the turning sweep area is as follows: Obtain the entry direction and exit direction corresponding to the candidate turning state after filtering. Determine the offset relationship between the vehicle's outer corner point and the vehicle's center point based on the vehicle constraint data. Calculate the trajectory of the vehicle's outer corner point according to the heading change process from the entry direction to the exit direction. Connect the outer boundary and inner boundary of the trajectory of the vehicle's outer corner point to generate the turning sweep area.

6. The AGV global path planning algorithm with steering constraints according to claim 5, characterized in that, The calculation process for the trajectory of the vehicle's outer corner points is as follows: The lower limit of the turning radius is determined based on the wheelbase and maximum steering angle in the vehicle constraint data. The turning sampling angle interval is determined according to the heading amplitude corresponding to the candidate steering state. The position of the vehicle's outer corner point is calculated point by point along the arc direction corresponding to the lower limit of the turning radius. The continuous positions of the vehicle's outer corner points are associated with the trajectory of the vehicle's outer corner points.

7. The AGV global path planning algorithm with steering constraints according to claim 6, characterized in that, The specific process for forming the steering constraint result is as follows: The turning sweep area is spatially overlapped with the road boundary occupancy area, fixed obstacle occupancy area, and predicted moving obstacle occupancy area in the environmental occupancy data. When the turning sweep area crosses the road boundary occupancy area or overlaps with the fixed obstacle occupancy area, a non-turning passage mark is determined. When the turning sweep area overlaps with the predicted moving obstacle occupancy area, a restricted passage mark is determined. The turning margin is determined based on the distance between adjacent boundaries to form the turning constraint result.

8. The AGV global path planning algorithm with steering constraints according to claim 1, characterized in that, The process for determining the road selectability constraints is as follows: Extract road segment status records connected to the candidate path nodes from the planning data, read the road segment occupancy status, traffic direction, available road width, predicted occupancy status, and number of task vehicles, compare the road segment occupancy status with the occupancy threshold to determine occupancy restrictions, compare the available road width with the vehicle passage width to determine width restrictions, compare the predicted occupancy status with the predicted occupancy threshold to determine prediction restrictions, and generate the road selectivity constraints from the occupancy restrictions, width restrictions, and prediction restrictions.

9. The AGV global path planning algorithm with steering constraints according to claim 8, characterized in that, The process for determining the road traffic flow value is as follows: Obtain the status records of the road segments that satisfy the road selectivity constraints; determine the road occupancy evaluation value based on the road segment vacancy ratio; determine the direction consistency evaluation value based on the matching relationship between the traffic direction and the vehicle driving direction; determine the width margin evaluation value based on the difference between the available road width and the vehicle traffic width; determine the task occupancy evaluation value based on the number of task vehicles; and perform a weighted summation of the road occupancy evaluation value, the direction consistency evaluation value, the width margin evaluation value, and the task occupancy evaluation value to generate the road smoothness value. The output process of the global path instruction is as follows: Obtain the candidate path ranked first in the sorted candidate global path sequence, extract the candidate path node, the entry direction, the departure direction, the steering margin, the passage marker, and the road accessibility value corresponding to the candidate path, convert the candidate path node into a position control point, convert the departure direction into a heading control variable, convert the steering margin and the passage marker into a speed control condition, associate the position control point, the heading control variable, and the speed control condition, and output the global path instruction.

10. An AGV global path planning system with steering constraints, the system being used to implement the AGV global path planning algorithm with steering constraints as described in any one of claims 1-9, characterized in that, The system includes: The node table building module obtains planning data, determines candidate path nodes based on the planning data, and generates a candidate turning relationship table according to the connection direction of the candidate path nodes. The steering filtering module acquires vehicle constraint data, determines the starting state, entry direction, departure direction and maximum steering angle based on the vehicle constraint data and the candidate steering relationship table, calculates the heading amplitude for candidate steering states, and filters out candidate steering states whose heading amplitude is greater than the maximum steering angle. The sweep judgment module calculates the turning sweep area based on the vehicle constraint data, the entry direction and the departure direction for the candidate turning states after screening, generates environmental occupancy data based on the planning data, determines the turning margin and passage mark according to the occupancy relationship between the turning sweep area and the environmental occupancy data, and forms the turning constraint result from the turning margin and the passage mark. The path output module generates road traffic attributes based on the planning data, determines road selectivity constraints and road accessibility values ​​based on the road traffic attributes, constructs a stateful path graph by combining the candidate path nodes, the turning constraint results, and the road accessibility values, removes the states that cannot be turned and road segments that do not meet the road selectivity constraints, sorts the path by the turning constraint results, the road accessibility values, and the road length, and outputs global path instructions.

Images