Closed park unmanned vehicle path planning method based on intelligent optimization algorithm

Abstract

The present application relates to the technical field of unmanned vehicle path planning, and discloses a closed park unmanned vehicle path planning method based on an intelligent optimization algorithm, comprising the following steps: step 1, acquiring park road, obstacle and unmanned vehicle start and end point data, determining a directed corridor graph, a dynamic no-entry time window and start and end state parameters; step 2, defining a composite candidate solution, determining a lateral bias range, a berthing convergence start point and a terminal berthing length; step 3, generating a candidate corridor skeleton, an initial population and a candidate detour sub-path library; step 4, constructing a unified total fitness function and completing initial fitness evaluation; step 5, determining a convergence slope and a corridor entropy and switching a squirrel search mode; step 6, replacing a high-cost sub-section and updating fitness during jump search; step 7, updating continuous variables and locking an optimal corridor sequence; and step 8, generating a deterministic executable trajectory. The present application realizes efficient planning of a closed park unmanned vehicle path and generation of a deterministic executable trajectory.

Description

Technical Field

[0001] This invention belongs to the field of unmanned vehicle path planning technology, specifically relating to a method for unmanned vehicle path planning in closed parks based on intelligent optimization algorithms. Background Technology

[0002] With the increasing application of unmanned vehicles in scenarios such as closed parks, factories, campuses, and logistics stations, path planning, as a key link in the autonomous control of unmanned vehicles, directly affects vehicle traffic efficiency, obstacle avoidance safety, and terminal docking accuracy. Although closed parks have the characteristics of clear road boundaries, fixed stations, and relatively controllable traffic organization, there are still problems in actual operation, such as limited road width, local one-way traffic, dynamic changes of temporary obstacles, and strict requirements for the docking posture at the destination.

[0003] In existing technologies, common path planning methods often focus on finding a shortest path based on grid maps or geometric space searches, or use a single heuristic algorithm for global optimization. While these methods are generally applicable in static environments, they have shortcomings when applied to autonomous vehicles in closed parks: First, they often prioritize shortest path search, neglecting constraints such as road width, boundary safety margins, driving curvature, speed limits, and road risks, resulting in insufficient matching between the obtained path and actual road conditions and vehicle kinematic constraints. Second, the handling of dynamic obstacles is mostly limited to spatial avoidance, lacking a unified consideration of obstacle occupancy periods, making it difficult to address dynamic traffic safety over time. Third, some solutions separate global path selection, local obstacle avoidance, and endpoint docking, easily leading to uneven end-point trajectories, untrackability, or difficulty in meeting target attitude and zero-speed docking requirements. Summary of the Invention

[0004] This invention provides a path planning method for unmanned vehicles in closed parks based on intelligent optimization algorithms. It solves the technical problems in related technologies where the path planning of unmanned vehicles in closed parks fails to adequately consider road topology constraints, dynamic obstacle timing constraints, and the need for precise docking at the end point, resulting in poor executability of the planned path, insufficient traffic safety, and low docking accuracy.

[0005] This invention provides a path planning method for unmanned vehicles in closed parks based on intelligent optimization algorithms, including the following steps:

[0006] Step 1: Obtain basic geographic data of park roads, attribute data of park roads, real-time monitoring data of dynamic obstacles, basic status of the starting point of the unmanned vehicle and basic status of the ending point of the unmanned vehicle, and determine the directed corridor map of the park, dynamic restricted time window and start and end status parameters of the unmanned vehicle.

[0007] Step 2: Define composite candidate solutions based on the directed corridor graph of the park and the start and end state parameters of the unmanned vehicles, and determine the feasible range of lateral offset, the starting point parameter of docking convergence and the end docking length parameter.

[0008] Step 3: Based on the directed corridor map of the park, the feasible range of lateral offset, and the composite candidate solution, determine the candidate corridor skeleton, the initial population, and the candidate bypass sub-path library;

[0009] Step 4: Based on the dynamic restricted access time window, the initial population, and the basic parameters of the unmanned vehicle, construct a unified overall fitness function and determine the fitness evaluation value of each candidate solution in the initial population;

[0010] Step 5: Determine the algorithm convergence slope and corridor entropy based on the fitness evaluation value in the most recent preset number of iterations, and determine the squirrel search mode switching result based on the algorithm convergence slope and corridor entropy;

[0011] Step 6: When the squirrel search mode switching result is a jump search, based on the unified overall fitness function and the candidate detour sub-path library, the high-cost corridor sub-segments in the candidate solutions are replaced to obtain updated candidate solutions, and the fitness evaluation value of the updated candidate solutions is determined.

[0012] Step 7: Based on the initial population or updated candidate solutions, update the continuous variables in each candidate solution according to the unified total fitness function to obtain the fine-tuned composite candidate solutions and their corresponding fitness evaluation values, and determine the optimal corridor sequence according to the preset corridor locking conditions.

[0013] Step 8: Generate a deterministic executable trajectory based on the fine-tuned composite candidate solution, the optimal corridor sequence, the dynamic no-entry time window, the berthing convergence start point parameters, and the end berthing length parameters.

[0014] The beneficial effects of this invention are as follows: For the path planning scenario of unmanned vehicles in closed parks, this invention unifies the modeling of park road topology, road attribute constraints, dynamic obstacle temporal information, and endpoint docking requirements. Based on this, an intelligent optimization algorithm is used to perform global route selection and local fine-tuning collaborative optimization of candidate paths. Compared to solutions that only focus on the shortest distance or static obstacle avoidance, this invention enables the planning results to simultaneously consider travel direction, boundary safety, curvature constraints, dynamic drivability, and endpoint docking accuracy, improving the executability and control adaptability of the path. Furthermore, by adaptively switching search modes, the algorithm's ability to escape local optima is enhanced, improving search efficiency and overall optimization effect. The final output trajectory has better smoothness, trackability, and zero-speed docking stability, making it more suitable for the actual operation and control of unmanned vehicles in closed parks. Attached Figure Description

[0015] Figure 1This is a flowchart of the unmanned vehicle path planning method for closed parks based on intelligent optimization algorithms of the present invention. Detailed Implementation

[0016] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0017] It should be noted that, unless otherwise defined, the technical or scientific terms used in one or more embodiments of the present invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in one or more embodiments of the present invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0018] like Figure 1 As shown, the path planning method for unmanned vehicles in a closed park based on intelligent optimization algorithms includes the following steps:

[0019] Step 1: Obtain basic geographic data of park roads, attribute data of park roads, real-time monitoring data of dynamic obstacles, basic status of the starting point of the unmanned vehicle and basic status of the ending point of the unmanned vehicle, and determine the directed corridor map of the park, dynamic restricted time window and start and end status parameters of the unmanned vehicle.

[0020] Step 2: Define composite candidate solutions based on the directed corridor graph of the park and the start and end state parameters of the unmanned vehicles, and determine the feasible range of lateral offset, the starting point parameter of docking convergence and the end docking length parameter.

[0021] Step 3: Based on the directed corridor map of the park, the feasible range of lateral offset, and the composite candidate solution, determine the candidate corridor skeleton, the initial population, and the candidate bypass sub-path library;

[0022] Step 4: Based on the dynamic restricted access time window, the initial population, and the basic parameters of the unmanned vehicle, construct a unified overall fitness function and determine the fitness evaluation value of each candidate solution in the initial population;

[0023] Step 5: Determine the algorithm convergence slope and corridor entropy based on the fitness evaluation value in the most recent preset number of iterations, and determine the squirrel search mode switching result based on the algorithm convergence slope and corridor entropy;

[0024] Step 6: When the squirrel search mode switching result is a jump search, based on the unified overall fitness function and the candidate detour sub-path library, the high-cost corridor sub-segments in the candidate solutions are replaced to obtain updated candidate solutions, and the fitness evaluation value of the updated candidate solutions is determined.

[0025] Step 7: Based on the initial population or updated candidate solutions, update the continuous variables in each candidate solution according to the unified total fitness function to obtain the fine-tuned composite candidate solutions and their corresponding fitness evaluation values, and determine the optimal corridor sequence according to the preset corridor locking conditions.

[0026] Step 8: Generate a deterministic executable trajectory based on the fine-tuned composite candidate solution, the optimal corridor sequence, the dynamic no-entry time window, the berthing convergence start point parameters, and the end berthing length parameters.

[0027] In one embodiment of the present invention, basic geographic data of park roads, park road attribute data, real-time monitoring data of dynamic obstacles, basic starting state of unmanned vehicles (UAVs), and basic ending state of UAVs are first acquired. Based on the acquired data, a directed corridor map of the park, a dynamic restricted time window, and UAV start-stop state parameters are determined. The basic geographic data of park roads refers to a set of basic data used to characterize the topological and spatial distribution relationships of roads within a closed park, including at least intersection locations, task station locations, road boundaries, and spatial distribution information of road segments. The park road attribute data refers to a set of data used to characterize the traffic capacity and vehicle motion constraints of each road segment, including at least road width, road boundary safety margin, maximum allowable driving curvature, maximum allowable driving speed, and road risk coefficient. The real-time monitoring data of dynamic obstacles refers to dynamic target occupancy information obtained in real time through vehicle-mounted sensing devices, roadside sensing devices, or the park dispatch system, used to reflect the spatiotemporal distribution of other vehicles, pedestrians, work equipment, or temporary obstacles in the park roads. The basic starting state and basic ending state of UAVs are used to characterize the kinematic constraints of the UAVs at the planning start time and target stopping position, respectively.

[0028] Specifically, in step 11, the basic geographic data of the park's roads is parsed to obtain intersection location data, task station location data, road segment boundary data, and actual spatial location data of road segments. Intersection location data refers to nodal spatial data representing the connection, branching, or merging locations of roads within the park; task station location data refers to nodal spatial data representing the locations of operational targets such as loading / unloading points, stopping points, inspection points, or delivery points; road segment boundary data refers to geometric boundary data used to define the left and right boundaries or traversable edge range of each road segment; and actual spatial location data of road segments refers to data representing the centerline, extension direction, spatial length, and actual location distribution of each road segment in the park's coordinate system. In practice, basic geographic information can be first read from high-precision maps, park electronic maps, pre-calibrated road models, or map databases stored on the scheduling platform; then, the read road network data is split and standardized to extract the location descriptions of intersections, task stations, road boundaries, and road segment entities. For example, for a curved road connecting two task stations, the starting and ending points, the sequence of centerline coordinates, and the corresponding left and right boundary lines can be extracted to form the basic geometric objects required for subsequent road structure modeling. By performing layered analysis on the basic geographic data of the park's roads, the originally continuously distributed road space within the closed park can be transformed into standardized road elements suitable for subsequent graph structure modeling and trajectory constraint calculation, providing support for the establishment of a basic path planning model that combines road topological relationships and geometric constraints in this invention.

[0029] Step 12 involves parsing the park's road attribute data and mapping road width, road boundary safety margin, maximum permissible curvature, maximum permissible speed, and road risk coefficient to the road segments represented by their actual spatial location data. Specifically, road width refers to the effective width of the corresponding road segment that allows unmanned vehicles to pass laterally; road boundary safety margin refers to the safe buffer distance reserved at the road edge to prevent collisions between the vehicle body, load, or control errors and the road boundary; maximum permissible curvature refers to the maximum allowable trajectory curvature of the unmanned vehicle on that road segment while meeting steering capability, tire adhesion, and vehicle posture constraints; maximum permissible speed refers to the highest safe driving speed set under the road segment's attribute conditions; and the road risk coefficient is a quantitative parameter used to describe the potential risk level of the road segment, comprehensively reflecting factors such as line-of-sight obstruction, frequency of oncoming traffic, intensity of cross-interference, frequency of operation vehicle entry and exit, or road surface complexity. In practice, based on park management rules, historical road traffic statistics, road environment labeling information, and the vehicle platform's own motion performance parameters, corresponding road attribute values ​​are assigned to each road segment, and a one-to-one correspondence is established between these attribute values ​​and the actual spatial location data of the road segments. By binding road attribute data to specific road segments, the subsequent path planning process can no longer rely solely on the shortest distance or single connectivity for searching. Instead, it can simultaneously consider road width limitations, boundary safety constraints, vehicle curvature constraints, speed constraints, and risk constraints during the search process. This makes the path obtained by the present invention more in line with the control and guidance requirements of unmanned vehicles in closed parks.

[0030] Step 13: Construct a directed corridor map of the park based on intersection location data, task station location data, road segment boundary data, actual spatial location data of road segments, and the park road attribute data represented by the actual spatial location data of each road segment. The directed corridor map of the park is a directed graph structure model used to represent the traffic direction, connection relationship, and attribute constraints of roads within a closed park; a corridor refers to abstracting a road segment with a relatively clear traffic direction and boundary constraints within the park as a continuous traffic channel along which unmanned vehicles can travel; "directed" means assigning directional attributes to road segments that allow one-way traffic or have different constraints in different directions, enabling the path planning process to distinguish between forward and reverse traffic. The directed corridor map of the park includes intersection nodes determined based on intersection location data, task station nodes determined based on task station location data, corridor edges determined based on actual spatial location data of road segments, and the corresponding road width, road boundary safety margin, maximum allowable driving curvature, maximum allowable driving speed, and road risk coefficient for each corridor edge. In practical implementation, each intersection and task station in the park can be abstracted as a node in the graph, and each road segment connecting the nodes can be abstracted as one or more directed edges in the graph. For road segments with bidirectional traffic capacity, two directed corridor edges in opposite directions can be constructed. For road segments with one-way traffic organization, scheduling priority direction, or differences in traffic risk in different directions, different directed edge attributes are assigned according to the actual traffic rules. After construction, the directed corridor graph of the park retains both the topological connection relationship between roads and the spatial and attribute constraints corresponding to each traffic edge. By constructing the directed corridor graph of the park, the complex park road network can be transformed into a structured traffic model suitable for optimization search algorithms. This enables the improved squirrel search algorithm of this invention to generate and filter candidate paths within the framework of directed channels and constraint attributes, thereby improving the adaptability of path search to the actual road organization form of the park.

[0031] Step 14: Determine the start and end state parameters of the unmanned vehicle (UAV) based on its starting and ending basic states. These parameters include the starting position, starting heading angle, starting speed, ending position, ending target heading angle, and ending speed, with the ending speed set to zero. The starting and ending positions are used to determine the starting and target spatial coordinates for path planning, respectively. The starting heading angle refers to the orientation of the UAV's longitudinal axis relative to the park's reference coordinate system at the start of planning. The starting speed refers to the actual or expected speed of the UAV at the start of planning. The ending target heading angle refers to the vehicle's orientation constraint that the UAV must meet after reaching the target parking position. Setting the ending speed to zero means that, in scenarios involving parking or precise positioning within a closed park, the UAV must ultimately complete its arrival, docking, or work docking in a stationary state. In practice, the current position, orientation, and speed information of the UAV can be obtained through an onboard positioning and attitude measurement device, and the ending position and ending target heading angle can be read from task scheduling instructions, station parking rules, or workstation constraints. By explicitly determining the state parameters of the start and end points, especially by constraining the end point speed to zero, subsequent path planning can generate not only a connected path, but also a control target path that meets the requirements of the starting attitude and the ending docking attitude.

[0032] Step 15: Discretize each corridor edge in the directed corridor map of the park into corridor units according to a preset length interval. Determine the occupancy time interval of each dynamic obstacle in each corridor unit based on real-time monitoring data of dynamic obstacles. Merge the occupancy time intervals corresponding to the same corridor unit to obtain the dynamic prohibition time window corresponding to each corridor unit. Here, a corridor unit refers to the smallest traffic analysis unit formed by discretizing a continuous corridor edge along its length direction at a fixed distance or a preset resolution; the occupancy time interval refers to the time range within which a dynamic obstacle enters and leaves a corridor unit in the prediction time domain; the dynamic prohibition time window refers to the time interval in a specific corridor unit where it is unsuitable for autonomous vehicles to pass due to the presence of one or more dynamic obstacles. In practical implementation, each corridor edge can first be divided into multiple corridor units of equal or approximately equal length according to the planning accuracy requirements. Then, based on real-time monitoring data of dynamic obstacles, combined with the current position, speed, heading, movement trend of dynamic obstacles, or the target trajectory given by the upper-level scheduling, the future occupancy position of obstacles in the road can be predicted temporally. Subsequently, the prediction results are mapped onto the corresponding corridor units to obtain the start and end times of each dynamic obstacle's occupation of each corridor unit. When multiple dynamic obstacle occupancy intervals exist in the same corridor unit within the same prediction time domain, these occupancy time intervals are combined along the time axis to obtain a unified dynamic no-entry time window for the corridor unit. By introducing a dynamic no-entry time window, this invention can elevate dynamic obstacle avoidance from a traditional pure spatial obstacle avoidance problem to a traffic control problem that also considers the time dimension. This allows the improved squirrel search algorithm to not only avoid spatial conflicts when searching for paths but also avoid obstacle occupancy periods, thereby improving the dynamic drivability of the planning results and the actual driving safety.

[0033] This embodiment constructs a directed corridor graph of the park by performing structured analysis on the basic geographic data and road attribute data of the park's roads. This graph includes intersection nodes, task station nodes, and corridor edge attribute constraints. Combined with the start and end state parameters of the unmanned vehicle and the dynamic prohibition time windows corresponding to dynamic obstacles, a unified expression of the closed park's road topology, vehicle motion constraints, and dynamic temporal conflict constraints is achieved. This provides an accurate and complete input basis for path search, enabling the planning results to simultaneously consider traffic directionality, boundary safety, curvature tracing, zero-speed stopping requirements at the destination, and dynamic obstacle avoidance needs. This improves the executability, stability, and control adaptability of unmanned vehicle path planning in closed parks.

[0034] In one embodiment of the present invention, a composite candidate solution is defined based on the directed corridor graph of the park and the start and end state parameters of the unmanned vehicle, and the feasible range of lateral offset, the docking convergence start point parameter, and the end docking length parameter are determined. The composite candidate solution refers to the parameterized description result of a set of candidate passage schemes used to characterize the unmanned vehicle from the start point to the end point. It includes both discrete topological information of which corridor edges the unmanned vehicle passes through and continuous control information of how the unmanned vehicle selects its passage position within each corridor edge and how it converges and stops at the end.

[0035] Specifically, in step 21, a composite candidate solution is defined based on the directed corridor map of the park and the start-end state parameters of the unmanned vehicle. The composite candidate solution includes a corridor sequence, a longitudinal anchor point ratio, a lateral offset, and an end-berthing length parameter. The corridor sequence refers to a sequence of edges formed by one or more interconnected corridor edges in the directed corridor map, arranged in the order of passage. This sequence represents the candidate passage skeleton of the unmanned vehicle from the starting area to the ending area. The longitudinal anchor point ratio refers to the normalized position ratio of a certain anchor point along the length of the corresponding corridor edge. Its value is used to determine whether the anchor point is located at the front, middle, or rear of the corridor edge. The lateral offset refers to the offset distance of the anchor point relative to the centerline of the corridor edge in the normal direction. This offset represents whether the unmanned vehicle travels to the left, center, or right within the corridor edge. The end-berthing length parameter refers to the length of the final path of the unmanned vehicle as it gradually adjusts its attitude and decelerates towards the target heading at the end point. In practice, the starting and ending connection areas can be determined first in the directed corridor map of the park based on the starting and ending positions. Then, a set of candidate corridor edges connecting the starting and ending points can be selected as a corridor sequence. Subsequently, a corresponding longitudinal anchor point ratio and lateral offset can be configured for each corridor edge in the corridor sequence, and an end-docking length parameter can be configured near the end point, thus forming a complete set of composite candidate solutions. For example, if an unmanned vehicle needs to drive from the park entrance to the loading and unloading station, and the candidate path passes through the main channel, the turning connection road, and the station pre-guide road in sequence, then these three corridor edges can jointly form a corridor sequence. Furthermore, corresponding anchor point ratios and lateral offsets can be set on these three corridor edges respectively, and an end-docking length parameter can be set at the end of the last guide road to characterize how the unmanned vehicle transitions from the main channel to the precise stopping process in front of the station.

[0036] Step 22: For each corridor edge in the corridor sequence, multiply the corresponding longitudinal anchor point ratio by the length of the corridor edge to obtain the longitudinal positioning length. Determine the longitudinal positioning point based on the centerline of the corridor edge, and then determine the plane position of the anchor point along the normal direction of the longitudinal positioning point with the corresponding lateral offset. Here, the longitudinal positioning length refers to the length accumulated from the starting point of the corridor edge along the centerline to the projected position of the anchor point; the longitudinal positioning point refers to the reference point located on the centerline of the corridor edge; the normal direction refers to the lateral direction perpendicular to the tangential direction of the centerline of the corridor edge at the longitudinal positioning point; and the plane position of the anchor point refers to the two-dimensional plane coordinate position obtained by offsetting the longitudinal positioning point along the normal direction. In specific implementation, the centerline of the corridor edge can be parameterized by arc length, and the longitudinal reference position of the anchor point on the centerline can be determined by multiplying the longitudinal anchor point ratio by the length of the corridor edge. Then, the normal direction of the centerline is obtained at this reference position, and the reference point is moved to the left or right by a corresponding distance along the normal direction according to the sign and magnitude of the lateral offset to obtain the corresponding plane position of the anchor point. For example, when the length of a corridor side is 20 meters and the longitudinal anchor point ratio is 0.5, the corresponding longitudinal positioning length is 10 meters, indicating that the anchor point is located near the center of the corridor side's center line; if the lateral offset is +0.4 meters, it means that the anchor point is further offset to one side by 0.4 meters from this position.

[0037] Step 23: For each corridor edge in the corridor sequence, read the corresponding road width and road boundary safety margin. Subtract the road boundary safety margin from half of the road width to obtain the allowable lateral offset distance on one side. Determine the feasible lateral offset range for the corridor edge based on the positive and negative ranges of the allowable lateral offset distance on one side. The allowable lateral offset distance on one side refers to the maximum safe distance the anchor point is allowed to deviate along the normal direction on one side, with the corridor edge centerline as a reference. The feasible lateral offset range refers to the range of allowable lateral offset values ​​without exceeding the road boundary and meeting the safety margin requirements. In practice, first read the road width corresponding to the corridor edge and use half of it as the theoretical distance from the centerline to the boundary; then subtract the road boundary safety margin from this theoretical distance to remove the risk area where the vehicle may intrude into the boundary due to control errors, positioning errors, or boundary disturbances; finally, determine the feasible lateral offset range for the corridor edge by combining the positive and negative ranges of the obtained allowable lateral offset distance on one side. For example, if the width of a road along a corridor is 3 meters and the safety margin at the road boundary is 0.3 meters, then the allowable lateral offset distance on one side is 1.2 meters, and the corresponding feasible range of lateral offset can be determined to be from -1.2 meters to 1.2 meters. By pre-determining the feasible range of lateral offset for each corridor side, it can be ensured that the plane position of the subsequently generated anchor points is always within the effective passable area of ​​the road, thereby avoiding problems such as trajectory overtravel, boundary crossing, or boundary collision caused by excessive lateral offset.

[0038] Step 24: Read the actual heading angle, corresponding lateral offset, target heading angle, and basic berthing length configuration data of the last corridor in the corridor sequence. Normalize the difference between the actual heading angle of the last corridor and the target heading angle to obtain the heading error. Multiply the heading error by the product of the heading and attitude error adjustment coefficient and the product of the absolute value of the lateral offset and the lateral offset adjustment coefficient, respectively, and then add them to the berthing length reference value. Limit the superposition result to the minimum and maximum berthing length values ​​to obtain the final berthing length parameters. Here, the actual heading angle of the last corridor refers to the angle of the autonomous vehicle on the corresponding driving direction on the last passage corridor before entering the destination; the heading error is the result of normalizing the angle difference between the actual heading angle and the target heading angle within a unified angle range; the basic berthing length configuration data includes the berthing length reference value, minimum berthing length, maximum berthing length, heading and attitude error adjustment coefficient, and lateral offset adjustment coefficient. In practice, the actual approach direction of the unmanned vehicle (UAV) as it approaches the destination can be determined first based on the direction of the last corridor. This direction is then compared with the target parking direction required at the destination to obtain the heading error. Simultaneously, the lateral offset corresponding to the last corridor is read to reflect whether the UAV is in a relatively ideal lateral position before approaching the destination. If the heading error is large, or the absolute value of the lateral offset is large, it indicates that the UAV still needs a longer distance to complete attitude adjustment and lateral convergence before approaching the destination, and the end-of-line parking length parameter should be increased accordingly. Conversely, if the heading error is small and the lateral offset is close to zero, it indicates that the UAV is close to the ideal parking posture, and the end-of-line parking length parameter can be reduced accordingly. For example, when the actual heading of the last corridor is basically consistent with the target heading of the destination, and the lateral offset is small, a shorter parking length can be used to complete the end-of-line alignment. However, when the angle between the two is large, or the vehicle is currently significantly deviated to one side within the lane, a longer parking segment is required to ensure that the UAV can smoothly complete attitude adjustment and achieve a zero-speed stop at the destination.

[0039] Step 25: Determine the berthing convergence start point parameters based on the endpoint position, the endpoint target heading angle, and the final berthing length parameters. Specifically, the endpoint target heading angle is used to determine the endpoint target heading direction. Starting from the endpoint position, the position is offset along the opposite direction of the endpoint target heading direction according to the final berthing length parameters. The offset position is then determined as the berthing convergence start point parameter. The berthing convergence start point parameter refers to the position parameter at which the unmanned vehicle begins to switch from a normal travel state to a precise berthing convergence state, representing the starting position of the final berthing segment. In practice, starting from the endpoint position, the vehicle can retreat a distance corresponding to the final berthing length parameters along the opposite direction of the endpoint target heading direction to determine a berthing convergence start point located ahead of the endpoint. During subsequent trajectory generation, the unmanned vehicle can primarily follow the normal travel path before this point, and then gradually converge towards the endpoint target heading and endpoint position after this point. By setting the berthing convergence start point parameter, the normal path tracking stage and the precise berthing stage can be spatially segmented, allowing the subsequent trajectory generation process to consider both path traversability and endpoint berthing accuracy.

[0040] This embodiment represents candidate travel paths in the directed corridor graph of the park as composite candidate solutions that simultaneously include corridor sequence, longitudinal anchor point ratio, lateral offset, and end-of-pipe docking length parameters. Furthermore, it determines the feasible range of lateral offset, end-of-pipe docking length parameter, and docking convergence start point parameter for each corridor edge. This allows the candidate paths to not only characterize which road segments the autonomous vehicle traverses but also its specific travel position within each road segment and its attitude convergence method near the endpoint. Therefore, it provides a foundation for subsequent search and optimization by representing candidate solutions that combine topological information and continuous geometric control information, thereby improving the adaptability of autonomous vehicle path planning in closed parks to boundary safety, local adjustability, and precise end-of-pipe docking requirements.

[0041] In one embodiment of the present invention, a candidate corridor skeleton, an initial population, and a candidate detour sub-path library are determined based on the directed corridor graph of the park, the feasible range of lateral offset, and composite candidate solutions. The candidate corridor skeleton refers to a sequence of candidate corridors with connectivity from the starting point to the ending point and satisfying basic traffic conditions, used to characterize the initial candidate traffic skeleton of the unmanned vehicle in the park's road network. The initial population refers to a set of initial composite candidate solutions constructed based on the candidate corridor skeleton and combined with parameters such as the longitudinal anchor point ratio, the lateral fixed level position, and the end docking length, used as the initial search objects for subsequent improvements to the squirrel search algorithm. The candidate detour sub-path library refers to a set of pre-generated and categorized connected detour sub-paths for each road connection node pair in the directed corridor graph of the park, used to replace locally high-cost segments during subsequent path optimization.

[0042] Specifically, in step 31, based on the starting and ending positions, the corridor edges corresponding to the starting and ending points are determined in the directed corridor graph of the park, and all connected corridor sequences between these two points are extracted. The starting-corresponding corridor edge is the corridor edge that matches or is closest to the starting position of the autonomous vehicle, and the ending-corresponding corridor edge is the corridor edge that matches or is closest to the ending position. A connected corridor sequence is a candidate edge sequence consisting of multiple corridor edges connected end-to-end with the same direction of travel, allowing continuous access from the starting-corresponding corridor edge to the ending-corresponding corridor edge. In practice, the starting and ending positions can be projected onto the corresponding corridor edges in the directed corridor graph of the park. Then, starting from the starting-corresponding corridor edge, the paths in the graph that satisfy the connectivity condition are traversed to extract all connected corridor sequences from the starting-corresponding corridor edge to the ending-corresponding corridor edge. For example, when an autonomous vehicle needs to travel from the park entrance to a loading / unloading station, the corridor edges where the entrance and loading / unloading station are located can be determined first, and then all walkable corridor edge combinations between them can be extracted to provide a basic path set for subsequent candidate skeleton selection. By extracting all connected corridor sequences, we can avoid relying on a single path selection for the initial candidate paths, thus reserving sufficient topology search space for subsequent optimization.

[0043] Step 32: For each corridor sequence, calculate the weighted result of corridor length and length weight, road risk coefficient and road risk weight, and average curvature and average curvature weight for each corridor edge, and then add the weighted result of number of corridor edges and number of corridors to obtain the skeleton cost. Based on the skeleton cost, determine a preset number of candidate corridor skeletons from low to high. The skeleton cost refers to the cost used to comprehensively evaluate the quality of a corridor sequence, reflecting its overall level in terms of travel distance, road risk, geometric curvature, and corridor switching complexity. In practice, for each connected corridor sequence, the total length, road risk coefficient corresponding to each corridor edge, overall path average curvature, and number of corridor edges can be calculated separately, and then weighted and summed according to preset weights to obtain the corresponding skeleton cost. Subsequently, all corridor sequences are sorted from low to high according to skeleton cost, and the top few are selected as candidate corridor skeletons. For example, given two paths that both lead to the destination, if one is slightly shorter but involves high-risk roads and numerous turns, while the other is slightly longer but has lower risk and a gentler curvature, the path more suitable as the initial skeleton can be prioritized after a comprehensive comparison using skeleton cost. By introducing skeleton cost, distance, risk, and traceability can be considered simultaneously in the early screening stage of candidate paths, avoiding the problem of poor foundation for subsequent optimization caused by using only the shortest path as the criterion.

[0044] Step 33: For each corridor edge in each candidate corridor skeleton, read the corresponding feasible lateral offset range. Divide the total lateral span between the negative and positive allowable lateral offset distances within this range into equal intervals according to the adjacent level intervals corresponding to the number of lateral levels, obtaining the level spacing. Using the position corresponding to the negative allowable lateral offset distance as the starting level position, progressively add the level spacing according to the lateral level sequence to obtain the fixed level position corresponding to each lateral level. Set the longitudinal anchor point ratio of each corridor edge to the midpoint ratio from the corridor entrance to the corridor exit. Here, a lateral level refers to several fixed lateral positions formed after discretizing the lateral passable range of the corridor edge; a fixed level position refers to the specific offset position of each lateral level in the normal direction of the corridor edge. In practice, the feasible lateral offset range of each corridor edge can be divided into several equally spaced hierarchical positions from left to right or from negative to positive, and these hierarchical positions can be used as discrete values ​​for the lateral offset in subsequent initial candidate solutions. Simultaneously, to simplify the initial construction process, the proportion of the longitudinal anchor points of each corridor edge can be uniformly set to the midpoint proportion, meaning the initial anchor points are located in the central region of each corridor edge. For example, if the feasible lateral offset range of a corridor edge is from -1.2 meters to +1.2 meters, setting 5 lateral levels can form 5 fixed hierarchical positions within this range, representing different initial passage positions such as left-leaning, left-center, center, right-center, and right-leaning. By discretizing the feasible lateral offset range hierarchically and initially setting the longitudinal anchor point proportion to the midpoint proportion, while ensuring the feasibility of the initial candidate solutions, candidate position combinations with certain lateral differences can be quickly constructed, providing a reasonable starting point for subsequent continuous variable optimization.

[0045] Step 34: Combine each candidate corridor skeleton, the longitudinal anchor point ratio of each corridor side, and the fixed level position to form an initial composite candidate solution. Then, determine the end-berthing length parameter corresponding to the initial composite candidate solution according to the calculation rules of Step 24. Add the end-berthing length parameter to the initial composite candidate solution to obtain the initial population. The initial population refers to a set of candidate solutions composed of multiple initial composite candidate solutions, used as the initial input for subsequent improvements to the squirrel search algorithm. In specific implementation, for each candidate corridor skeleton, combine it with the midpoint longitudinal anchor point ratio of each corridor side and different lateral fixed level positions to form several sets of initial composite candidate solutions. Then, based on the actual heading angle of the corridor side where the end of each set of initial composite candidate solutions is located, the corresponding lateral offset, and the target heading angle at the endpoint, determine the corresponding end-berthing length parameter according to the calculation rules of Step 24, and add this parameter to the corresponding candidate solution to obtain a complete set of initial composite candidate solutions. For example, for the same candidate corridor skeleton, selecting two fixed-level positions—central and slightly to the right—on each corridor side can generate different initial composite candidate solutions. Although these candidate solutions have the same topological skeleton, they may exhibit different safety and docking effects due to differences in lateral passage positions and end-berthing lengths. By constructing an initial population, diverse and fundamentally feasible search starting points can be provided for subsequent search algorithms, thereby improving the convergence effect of global search and local optimization.

[0046] Step 35: For each road connection node pair in the directed corridor graph of the park, generate corresponding connected detour sub-paths; calculate the skeleton cost of each connected detour sub-path according to the skeleton cost calculation method in Step 32, sort them from low to high skeleton cost, and then classify and integrate them according to road connection node pairs to obtain a candidate detour sub-path library. Here, a road connection node pair refers to a pair of nodes in the directed corridor graph of the park that can be used as a local replacement start and end point; a connected detour sub-path refers to several sub-paths that connect the node pair and can be used as a local replacement segment of the original path; the candidate detour sub-path library refers to the set of detour sub-paths stored according to node pairs. In specific implementation, several locally connected sub-paths can be generated for each node pair that satisfies the connectivity relationship in the directed corridor graph of the park, and these sub-paths can be evaluated and sorted according to the skeleton cost rules in Step 32; subsequently, the sub-paths corresponding to the same node pair are centrally classified and stored to form a candidate detour sub-path library oriented towards different local connectivity relationships. For example, if a local path segment connects node A and node B, multiple alternative detour sub-paths can be pre-generated for the route from node A to node B. If, during subsequent optimization, a high-risk or high-cost segment is found between A and B in the original path, an alternative sub-path can be quickly selected from the candidate detour sub-path library corresponding to that node pair. By pre-establishing a candidate detour sub-path library, it is possible to avoid re-performing local path generation every time a local high-cost segment appears during subsequent searches, thereby improving local replacement efficiency and enhancing the flexibility of the path optimization process.

[0047] This embodiment extracts all connected corridor sequences from the starting point to the ending point from the directed corridor graph of the park, and obtains candidate corridor skeletons based on skeleton cost screening. Then, it performs lateral hierarchical discretization and longitudinal midpoint initialization on the candidate corridor skeletons, combining the lateral offset feasible range, to construct an initial population. Simultaneously, a candidate detour sub-path library is pre-established for each road connection node pair, ensuring that the subsequent optimization process has both a good initial search foundation and the ability to quickly replace locally high-cost sub-segments. Therefore, it can improve the initialization quality, local detour flexibility, and overall search efficiency of unmanned vehicle path planning in closed parks.

[0048] In one embodiment of the present invention, a unified overall fitness function is constructed based on the dynamic restricted access time window, the initial population, and the basic parameters of the autonomous vehicle, and the fitness evaluation value of each candidate solution in the initial population is determined. The unified overall fitness function is a unified cost function that comprehensively evaluates candidate solutions in multiple aspects, including path length, boundary safety, curvature constraints, dynamic obstacle timing conflicts, end-point stopping error, vehicle trackability, and corridor switching complexity. The fitness evaluation value is the comprehensive evaluation result obtained after substituting each candidate solution into the unified overall fitness function, used to characterize the quality of the candidate solution. By constructing a unified overall fitness function, the requirements for geometric feasibility, motion feasibility, dynamic obstacle avoidance capability, and end-point stopping accuracy involved in the path planning of autonomous vehicles in closed parks can be uniformly incorporated into the same evaluation framework, providing a consistent evaluation basis for subsequent improvements to the squirrel search algorithm's search updates and candidate solution selection.

[0049] Specifically, in step 41, each candidate solution in the initial population is read, and the distance between adjacent anchor points in the driving order is accumulated to obtain the path length evaluation term. The actual distance from each anchor point to the nearest road boundary is calculated, and the portion greater than zero of the difference between the safety gap threshold and the actual distance is squared point by point and accumulated to obtain the safety gap evaluation term. The path length evaluation term characterizes the total driving distance of the path corresponding to the candidate solution, while the safety gap evaluation term characterizes the degree of safety margin deficiency of each anchor point relative to the road boundary in the candidate solution. In practice, the distance between adjacent anchor points can be calculated segment by segment according to the sequential order of the anchor points in the candidate solution and accumulated as the path length evaluation term for that candidate solution. Then, the actual distance from each anchor point to the nearest road boundary is calculated and compared with the preset safety gap threshold. When the actual distance is less than the safety gap threshold, the insufficient portion is squared and accumulated to obtain the safety gap evaluation term. For example, if a candidate solution has a relatively short overall path, but multiple anchor points are close to the road boundary, this candidate solution, while advantageous in length, will have an excessively large safety clearance evaluation term, thus being suppressed in the overall evaluation. By simultaneously introducing path length and safety clearance evaluation terms, we can avoid pursuing shorter distances at the expense of boundary safety in path search, making the initial candidate solution evaluation results more consistent with the actual operational requirements of unmanned vehicles in closed parks.

[0050] Step 42: For each candidate solution, calculate the discrete curvature corresponding to each anchor point based on the geometric relationship between the planar positions of the three adjacent anchor points, and determine the difference between adjacent discrete curvatures as the curvature change; square the portion of the discrete curvature exceeding the maximum allowable curvature of the path point by point and sum them up, then sum the curvature change and the corresponding curvature change penalty coefficient to obtain the curvature and curvature change evaluation terms; the curvature and curvature change evaluation terms can be expressed as: ;

[0051] in, This represents the evaluation terms for curvature and curvature variation, where N represents the number of anchor points. Let represent the discrete curvature corresponding to the i-th anchor point. Discrete curvature is used to characterize the degree of curvature of the path near the anchor point. This represents the discrete curvature corresponding to the (i+1)th anchor point. Indicates the maximum allowed curvature of the path. The curvature change penalty coefficient is represented by the first term in the above formula, which is used to penalize the part of the discrete curvature that exceeds the maximum curvature allowed by the path, and the second term is used to penalize the case where the curvature of the path changes too quickly, so that the path corresponding to the candidate solution can maintain good smoothness while satisfying the turning constraints.

[0052] Further, the distance between adjacent anchor points is divided by the preset reference speed of the corresponding path segment and accumulated in the order of travel to obtain the time for each anchor point to reach the corresponding corridor unit. Then, the arrival time of each anchor point in the corresponding corridor unit is compared with the dynamic no-entry time window one by one. A preset dynamic conflict penalty coefficient is assigned to each anchor point falling into the dynamic no-entry time window, and the preset dynamic conflict penalty coefficients corresponding to all anchor points are accumulated to obtain the dynamic obstacle temporal conflict evaluation term. Specifically, the dynamic obstacle temporal conflict evaluation term is used to characterize the degree of conflict between the candidate solution and the time period occupied by the dynamic obstacle in the time dimension, and can be expressed as: ,in, This indicates a dynamic obstacle timing conflict evaluation item. This indicates the indicator function, representing the estimated arrival time of the j-th anchor point. Dynamic prohibition time window for falling into the corresponding corridor unit The value is 1 if the condition is met, and 0 otherwise. This represents the preset dynamic conflict penalty coefficient. In this way, the case where the candidate solution overlaps with the time period occupied by the dynamic obstacle can be quantitatively penalized, so that the path evaluation not only considers spatial accessibility, but also the temporal obstacle avoidance capability in the dynamic environment.

[0053] For example, if a candidate solution has a relatively smooth geometric path, but its expected arrival time at a certain intersection corridor cell coincides with another vehicle occupying that cell, the dynamic obstacle timing conflict evaluation term for that candidate solution will increase. By simultaneously introducing curvature and curvature change evaluation terms as well as dynamic obstacle timing conflict evaluation terms, the evaluation of candidate solutions can focus not only on whether the geometric path is smooth, but also on its temporal drivability in a dynamic environment.

[0054] Step 43: For each candidate solution, calculate the distance error between the endpoint of the final trajectory and the endpoint position, and calculate the heading error between the heading angle of the endpoint of the final trajectory and the heading angle of the target endpoint. Sum the squared distance error with the heading error, combined with the final heading error penalty coefficient, to obtain the final docking error evaluation term. Sum the portion of discrete curvature exceeding the physical limit curvature of the autonomous vehicle point by point to obtain the vehicle trackability evaluation term. Count the number of corridor sequence switches to obtain the corridor switching frequency evaluation term. The final docking error evaluation term characterizes the degree of deviation of the candidate solution in terms of endpoint position alignment accuracy and endpoint attitude alignment accuracy. The vehicle trackability evaluation term characterizes whether the path corresponding to the candidate solution exceeds the physical steering capability of the autonomous vehicle. The corridor switching frequency evaluation term characterizes the switching complexity of the candidate path between different corridor edges. For example, if a candidate solution has a short total distance but its final orientation deviates significantly from the required direction of the station, or if there are multiple sharp bends and frequent corridor switches along the way, its final docking error evaluation term, vehicle trackability evaluation term, or corridor switching frequency evaluation term will increase accordingly. By introducing these three types of evaluation items, the evaluation of candidate solutions can simultaneously take into account the quality of the endpoint stop, the vehicle motion limit constraints, and the path structure complexity, thereby improving the engineering feasibility of subsequent search results.

[0055] Step 44: Multiply the path length evaluation item, safety gap evaluation item, curvature and curvature change evaluation item, dynamic obstacle timing conflict evaluation item, end-point docking error evaluation item, vehicle trackability evaluation item, and corridor switching frequency evaluation item by their corresponding weight configuration coefficients, and then sum them to obtain a unified overall fitness function. For each candidate solution in the initial population, extract the corresponding evaluation items and substitute them into the unified overall fitness function to obtain the fitness evaluation value of each candidate solution. The weight configuration coefficients are used to adjust the influence of different evaluation items in the comprehensive evaluation and can be preset according to the park application scenario, vehicle operation objectives, and safety requirements. In practical implementation, the path length evaluation item, safety gap evaluation item, curvature and curvature change evaluation item, dynamic obstacle timing conflict evaluation item, end-stop error evaluation item, vehicle trackability evaluation item, and corridor switching number evaluation item corresponding to each candidate solution can be weighted and summed according to a unified weighting rule to form a unified overall fitness function value. Subsequently, the fitness evaluation value of each candidate solution in the initial population is calculated one by one, and this fitness evaluation value is used as the basis for subsequent improvement of the squirrel search algorithm for optimization, updating, and mode switching. For example, in the station connection scenario with high requirements for stopping accuracy, the weight of the end-stop error evaluation item can be appropriately increased; while in the mixed-traffic scenario with high requirements for dynamic obstacle avoidance, the weights of the dynamic obstacle timing conflict evaluation item and the safety gap evaluation item can be increased. By constructing a unified overall fitness function and calculating the fitness evaluation value of each candidate solution, multiple constraints and performance objectives can be uniformly mapped to a comparable single evaluation result, thereby providing a clear, unified, and adjustable evaluation standard for the subsequent search and optimization process.

[0056] This embodiment calculates path length evaluation, safety gap evaluation, curvature and curvature change evaluation, dynamic obstacle temporal conflict evaluation, end-point docking error evaluation, vehicle trackability evaluation, and corridor switching frequency evaluation for each candidate solution in the initial population. A unified overall fitness function is then used for comprehensive weighted evaluation, achieving a unified quantitative representation of the geometric feasibility, dynamic drivability, end-point docking accuracy, and vehicle motion constraints of the candidate solutions. This provides a consistent evaluation basis for subsequent candidate solution selection and search updates, thereby improving the overall quality and engineering applicability of unmanned vehicle path planning results in closed parks.

[0057] In one embodiment of the present invention, the algorithm convergence slope and corridor entropy are determined based on the fitness evaluation values ​​in the most recent preset number of iterations, and the squirrel search mode switching result is determined based on the algorithm convergence slope and corridor entropy. The algorithm convergence slope characterizes the improvement trend of the optimal fitness evaluation value in the most recent iterations; the corridor entropy characterizes the dispersion of current high-quality candidate solutions in corridor selection; and the squirrel search mode switching result indicates whether the subsequent search process should adopt a jump search or an incremental search. By simultaneously examining the fitness convergence trend and the corridor diversity of candidate paths, the search process can adaptively switch between continuing local refinement and re-expanding the search scope, thereby improving the search efficiency and result quality of unmanned vehicle path planning in closed parks.

[0058] Specifically, in step 51, the optimal fitness evaluation value of each generation in the most recent preset number of iterations is extracted. A linear regression is performed with the iteration generation as the independent variable and the optimal fitness evaluation value as the dependent variable. The slope of the linear regression is then determined as the algorithm's convergence slope. Here, the most recent preset number of iterations refers to a fixed number of consecutive iterations backtracking from the current iteration generation; the optimal fitness evaluation value refers to the function value with the best fitness evaluation value among the candidate solutions in each generation; and the algorithm's convergence slope reflects the changing trend of the optimal fitness evaluation value as the iteration progresses. In practice, the sequence of optimal fitness evaluation values ​​from the most recent several generations can be extracted, and a linear regression can be performed with the iteration generation as the horizontal axis and the optimal fitness evaluation value as the vertical axis to obtain the corresponding regression slope. The algorithm's convergence slope can be expressed as: ,in, This represents the convergence slope of the algorithm. This represents the algebra of the most recent preset number of iterations. This represents the iteration number corresponding to the i-th sampling point. This represents the optimal fitness evaluation value corresponding to the i-th sampling point. This represents the average of the most recent preset number of iterations. This represents the average of the optimal fitness evaluation values ​​over the most recent preset number of iterations. If the slope is significantly negative, it indicates that the optimal fitness evaluation value continues to decrease as iterations progress, and the search process still shows a significant optimization trend. If the slope is close to zero, it indicates that the recent changes in the optimal fitness evaluation value are small, and the search may have entered a convergence stagnation state. For example, if the changes in the optimal fitness evaluation value over the most recent few generations are only small fluctuations, the regression slope is usually close to zero, indicating that the benefits of continuing to optimize in the current direction are limited. By introducing the algorithm's convergence slope, we can avoid judging the search state solely based on the optimal value of a single generation, thereby improving the stability of mode switching judgments.

[0059] Step 52: Select a predetermined number of candidate solutions as elite solutions based on their fitness evaluation values, from best to worst. Count the frequency of each corridor edge in the corridor sequences of the elite solutions, and determine the occurrence ratio of each corridor edge by the proportion of its occurrences to the total number of occurrences of all corridor edges. Here, elite solutions refer to candidate solutions with the best fitness evaluation values ​​in the current iteration population; the frequency of a corridor edge occurrence refers to the cumulative number of times a particular corridor edge is included in the corridor sequences of all elite solutions; and the occurrence ratio refers to the ratio of the frequency of a particular corridor edge occurrence to the total number of occurrences of all corridor edges. In practice, several candidate solutions with the best fitness evaluation values ​​can be selected as elite solutions from the current iteration population first. Then, the corridor sequences corresponding to these elite solutions are traversed, and the frequency of each corridor edge is counted. The occurrence ratio is obtained by dividing the frequency of each corridor edge occurrence by the total number of occurrences of all corridor edges. For example, if the same set of corridor edges repeatedly appears in most elite solutions, it indicates that the current search results are becoming concentrated in corridor selection; conversely, if the corridor sequences corresponding to different elite solutions differ significantly, it indicates that the current search still retains good path diversity. By statistically analyzing the proportion of corridor edges appearing in elite solutions, we can provide a basis for calculating corridor entropy, thereby quantifying the diversity level of the current candidate paths.

[0060] Step 53: For each corridor edge with a occurrence ratio greater than zero, perform a logarithmic operation on the occurrence ratio, multiply it by the corresponding occurrence ratio, sum the results, and then normalize the result by combining it with the total number of corridor edges in the directed corridor graph of the park to determine the corridor entropy. Corridor entropy characterizes the dispersion of elite solutions in corridor edge selection. A larger corridor entropy indicates a more dispersed distribution of elite solutions across different corridor edges, resulting in higher search diversity; a smaller corridor entropy indicates a high concentration of elite solutions on a few corridor edges, resulting in lower search diversity. This can be expressed as: , Let E represent the corridor entropy, and let E represent the set of corridor edges in the directed corridor graph of the park. This indicates the total number of corridor edges in the corridor diagram of the park. This represents the proportion of corridor edge e in the elite solution corridor sequence. For example, if elite solutions are almost all concentrated near the same main path framework, only a few corridor edges will have a high proportion, resulting in a relatively small corridor entropy. If elite solutions are distributed across multiple candidate paths, more corridor edges will appear in a relatively dispersed proportion, leading to a larger corridor entropy. By introducing corridor entropy, we can determine from the perspective of topological diversity whether the current search has become overly concentrated in local path regions.

[0061] Step 54 defines a convergence stagnation state as an algorithm with a convergence slope greater than or equal to a negative convergence threshold, and a corridor entropy less than the minimum corridor entropy as a state of insufficient corridor diversity. The convergence threshold measures whether the optimal fitness evaluation value is still effectively improving in the most recent iteration; the minimum corridor entropy measures whether the current elite solution still has sufficient diversity in corridor edge selection. In practice, the algorithm convergence slope obtained in step 51 is first compared with the preset negative convergence threshold. If the convergence slope is greater than or equal to this negative threshold, it indicates that the optimal fitness evaluation value is not decreasing significantly, and it can be determined as a convergence stagnation state. Simultaneously, the corridor entropy obtained in step 53 is compared with the preset minimum corridor entropy. If the corridor entropy is lower than this minimum value, it indicates that the current elite solution is too concentrated in corridor selection, and it can be determined as a state of insufficient corridor diversity. By setting judgment conditions from both the optimization trend and path diversity aspects, it is possible to more accurately identify whether the current search needs to change the search mode.

[0062] Step 55: When the convergence stagnation state or the corridor diversity insufficiency state is established, the squirrel search mode switching result is determined to be a jump search; when neither the convergence stagnation state nor the corridor diversity insufficiency state is established, the squirrel search mode switching result is determined to be an incremental search. Jump search refers to a search method that expands the search range and escapes local convergence by adjusting the candidate solution structure or replacing local corridor segments over a larger range; incremental search refers to a search method that refines continuous parameters based on the current better candidate solution to further improve the quality of the candidate solution. In specific implementation, as soon as the convergence stagnation state or the corridor diversity insufficiency state is detected, it can be determined that the benefit of continuing to refine the current search along the original direction is limited, and at this time, a jump search is switched to enhance the global escape capability of the search; only when neither the convergence stagnation state nor the corridor diversity insufficiency state is established will incremental search continue to perform local fine-tuning of the current candidate solution. For example, if the optimal fitness evaluation value has almost stopped decreasing in recent generations, or if most elite solutions are concentrated on the same corridor skeleton, then the subsequent search mode can be switched to jump search; conversely, if the optimal fitness continues to improve and the distribution of elite solutions is still relatively dispersed, it indicates that there is still room for further refinement in the current search, and incremental search can be maintained. By setting the above mode switching rules, the improved squirrel search algorithm can achieve an adaptive balance between global exploration and local development.

[0063] This embodiment determines the algorithm's convergence slope based on the optimal fitness evaluation value in the most recent preset number of iterations, and determines the corridor entropy based on the proportion of corridor edges appearing in elite solutions. It then adaptively switches the squirrel search mode by combining convergence stagnation states and insufficient corridor diversity states, achieving joint discrimination of the optimization trend and path diversity in the search process. Therefore, it can promptly switch to jump search when the search gets stuck in a local stagnation or candidate paths are overly concentrated, and maintain incremental search when the search still has optimization potential and good diversity, thereby improving the search efficiency, ability to escape local optima, and overall optimization effect of unmanned vehicle path planning in closed parks.

[0064] In one embodiment of the present invention, when the squirrel search mode switching result is a jump search, high-cost corridor segments in the candidate solutions are replaced according to the unified overall fitness function and the candidate detour sub-path library to obtain updated candidate solutions, and the fitness evaluation value of the updated candidate solutions is determined. Here, jump search refers to a search method that expands the search scope again by replacing local high-cost corridor segments when the optimization of the current candidate solution has stagnated or path diversity is insufficient; a high-cost corridor segment refers to a local continuous corridor edge segment in the corridor sequence corresponding to the candidate solution that has a high overall cost and a significant impact on the overall path quality; an updated candidate solution refers to a new candidate solution formed by replacing high-cost corridor segments while retaining the remaining effective structural parameters of the original candidate solution.

[0065] Specifically, in step 61, when the squirrel search mode switching result is a jump search, for each corridor sub-segment in each candidate solution, the weighted result of the sub-segment length and length weight, the weighted result of the square of the greater than zero part of the difference between the safety gap threshold and the actual distance and the safety obstacle avoidance weight, the weighted result of the discrete curvature and curvature weight, and the weighted result of the dynamic no-entry time window hit result and the dynamic obstacle avoidance weight are summed to obtain the comprehensive cost.

[0066] Here, a corridor segment refers to a locally continuous path segment in the candidate solution corridor sequence, consisting of one or more adjacent corridor edges; the comprehensive cost refers to the cost used to evaluate the local quality of the corridor segment. In practice, each candidate solution can be divided according to locally continuous corridor edges, and the length cost, insufficient safety clearance cost, curvature cost, and dynamic no-entry time window hit cost of each corridor segment can be calculated separately. Then, a weighted sum is performed according to preset weights to obtain the comprehensive cost corresponding to the corridor segment. The insufficient safety clearance cost can be expressed as: , This indicates the cost of insufficient safety clearance. This represents the set of anchor points involved in the r-th corridor sub-segment. Indicates the safety gap threshold. This represents the actual distance from the i-th anchor point to the nearest road boundary. For example, if a candidate solution has a short middle section, but multiple anchor points are close to the boundary, some turns are too sharp, and the expected passage time overlaps with the time period occupied by dynamic obstacles, then the overall cost of this corridor segment will be high. By calculating the overall cost of each corridor segment within the candidate solution, we can more accurately locate the local weak areas affecting the overall path quality, providing a basis for subsequent replacements.

[0067] Step 62: Compare the comprehensive costs of each corridor segment in each candidate solution, determine the high-cost corridor segment with the highest comprehensive cost, and designate the starting connection node of the high-cost corridor segment in the original corridor sequence as the front connection node and the ending connection node as the back connection node. The front and back connection nodes represent the entry and exit connection positions of the corridor segment to be replaced in the original corridor sequence, respectively. In practice, the comprehensive costs of each corridor segment within the same candidate solution can be compared, and the corridor segment with the highest comprehensive cost can be selected as the high-cost corridor segment. Then, the start and end positions of this high-cost corridor segment in the original corridor sequence are read, and the node connected to its first end is designated as the front connection node, and the node connected to its last end is designated as the back connection node. For example, if a candidate solution goes from node A through nodes B and C to node D, and the local path segment between nodes B and C has the highest comprehensive cost, then node B can be designated as the front connection node, and node C as the back connection node. By further locating high-cost corridor segments to a pair of clearly defined connecting nodes, an indexing basis can be established for the subsequent extraction of alternative sub-paths from the candidate detour sub-path library.

[0068] Step 63: Based on the preceding and following connected nodes, extract the corresponding candidate detour sub-paths from the candidate detour sub-path library; determine the alternative sub-path by using the sum of the current iteration number and the current candidate solution number, combined with the number of sub-paths in the candidate detour sub-path library, through cyclic indexing. The candidate detour sub-path library refers to the set of local alternative paths that match the current preceding and following connected nodes; cyclic indexing refers to the selection method of cyclically looping back according to the number of sub-paths when the index value exceeds the number of sub-paths. In specific implementation, first extract all candidate detour sub-paths for the corresponding node pair from the candidate detour sub-path library based on the preceding and following connected nodes, then add the current iteration number and the current candidate solution number, and perform cyclic indexing based on the number of candidate detour sub-paths for that node pair to determine the alternative sub-path used in this replacement. The purpose of this setup is to ensure that different candidate bypass sub-paths are selected more evenly across different iterations and candidate solutions during the jump search, preventing all candidate solutions from being replaced by the same path at the same time, thus maintaining the diversity of the jump search. The alternative sub-path index can be represented as: ,in, This indicates the index number of the alternative sub-path in the candidate detour sub-path sub-database. Indicates the current iteration generation. Indicates the current candidate solution number. This represents the number of sub-paths in the candidate bypass sub-path sub-path library corresponding to the preceding connected node u and the following connected node v. mod represents the modulo operation, which is used to implement cyclic indexing.

[0069] Step 64: Replace the high-cost corridor segments in the original corridor sequence with alternative sub-paths to obtain a new corridor sequence; inherit the original longitudinal anchor point ratio and the original lateral offset parameters for the corridor edges in the new corridor sequence that correspond to the original corridor sequence; map the original longitudinal anchor point ratio and the original lateral offset parameters for the newly added corridor edges in relative order, and ensure that the original lateral offset parameters are within the lateral offset feasible range of the corresponding corridor edges; retain the end berthing length parameter to obtain an updated candidate solution. In practice, the high-cost corridor segments located between the preceding and following connecting nodes in the original corridor sequence can be entirely replaced with the alternative sub-paths determined in step 63, forming a new corridor edge sequence. For corridor edges in the new corridor sequence that are still consistent with the original sequence, the longitudinal anchor point ratio and lateral offset parameters corresponding to the original candidate solution are directly inherited. For newly added corridor edges after replacement, the longitudinal anchor point ratio and lateral offset parameters in the original high-cost corridor segments are mapped according to their relative order in the new sub-path, and the lateral offset is truncated or adjusted to the feasible range of the corresponding corridor edge when necessary. At the same time, the end berthing length parameter in the original candidate solution is retained, thus forming a complete updated candidate solution. For example, if the original high-cost corridor segment contains two corridor edges, and the alternative sub-path contains three new corridor edges, the original parameters can be mapped to the three new edges in relative order, and the offsets that exceed the feasible range of the lateral offset are corrected. By inheriting the continuous parameters of the original candidate solution as much as possible after replacing the corridor segment, the updated candidate solution can maintain the continuity of the overall parameter structure while achieving local transitions, thus avoiding re-initialization after each replacement.

[0070] Step 65: Substitute the updated candidate solutions into the unified overall fitness function to recalculate the path length evaluation item, safety gap evaluation item, curvature and curvature change evaluation item, dynamic obstacle temporal conflict evaluation item, end-stop error evaluation item, vehicle trackability evaluation item, and corridor switching number evaluation item, and determine the fitness evaluation value of the updated candidate solutions. In specific implementation, the evaluation indicators of the updated candidate solutions can be re-extracted according to the aforementioned calculation rules of the unified overall fitness function, and their new fitness evaluation values ​​can be calculated for subsequent candidate solution ranking, retention, and further optimization. For example, if an updated candidate solution significantly reduces the dynamic obstacle temporal conflict evaluation item and curvature evaluation item by replacing the original high-cost corridor segment, even if its path length increases slightly, its overall fitness evaluation value may still be better than the original candidate solution before the replacement. By recalculating the fitness evaluation value of the updated candidate solutions, it can be ensured that the new candidate solutions generated by the jump search are still comparable to the original candidate solutions under the unified evaluation criteria, thereby ensuring the consistency of the entire search optimization process.

[0071] This embodiment calculates the comprehensive cost of each corridor segment within the candidate solution when the squirrel search mode switches to a jump search, locates the high-cost corridor segment with the highest comprehensive cost, and then replaces this high-cost corridor segment with the candidate detour sub-path library. Simultaneously, it inherits and maps continuous parameters from the original candidate solution to form an updated candidate solution, and further recalculates its fitness evaluation value, thus achieving targeted correction of locally poor path segments. Therefore, while maintaining the overall feasible structure of the original candidate solution, it can effectively escape locally poor path regions, improving the global exploration capability, local detour flexibility, and search optimization efficiency of unmanned vehicle path planning in closed parks.

[0072] In one embodiment of the present invention, based on the initial population or updated candidate solutions, the continuous variables in each candidate solution are updated according to a unified overall fitness function to obtain fine-tuned composite candidate solutions and their corresponding fitness evaluation values. The optimal corridor sequence is then determined according to preset corridor locking conditions. Continuous variables refer to parameters in the composite candidate solutions that can be continuously adjusted under a given topology, mainly including the longitudinal anchor point ratio, lateral offset, and end-berthing length. Fine-tuned composite candidate solutions refer to candidate solutions obtained by refining and updating the aforementioned continuous variables while keeping the corridor sequence unchanged or essentially unchanged. The optimal corridor sequence refers to the corridor sequence that stably dominates in multiple iterations and satisfies the corridor locking conditions. By iteratively optimizing at the continuous variable level of candidate solutions, the path geometry, boundary safety, and end-berthing quality can be further improved while retaining a superior topological framework. Furthermore, when the corridor sequence has stabilized, the search focus shifts to fine-tuning continuous parameters, thereby improving search efficiency and result stability.

[0073] Specifically, in step 71, if an updated candidate solution obtained from the jump search exists, the updated candidate solution and its fitness evaluation value are read; otherwise, the initial population and its fitness evaluation value are read; and the longitudinal anchor point ratio, lateral offset, and end-of-line berthing length in each candidate solution are integrated into a continuous variable vector. Here, the continuous variable vector refers to the vectorized parameter expression formed by arranging all continuously adjustable parameters in the same candidate solution in a preset order. In specific implementation, when a jump search was performed in the previous stage and an updated candidate solution was generated, the updated candidate solution can be used as the input object for the current continuous variable optimization; if no jump search was performed, the initial population is directly used as the input object. Subsequently, the longitudinal anchor point ratio, lateral offset, and end-of-line berthing length parameters corresponding to each corridor edge in each candidate solution are concatenated in a unified order to form a continuous variable vector. For example, for a candidate solution containing three corridor edges, three sets of longitudinal anchor point ratios, three sets of lateral offsets, and one set of end-of-line berthing length parameters can be sequentially integrated into a continuous variable vector. By vectorizing the continuous adjustment parameters in the candidate solutions, a standardized input form can be provided for subsequent dimension-by-dimensional perturbation judgment and parameter update.

[0074] Step 72: For each variable dimension of each continuous variable vector, calculate the unified total fitness function value corresponding to the positive perturbation and the unified total fitness function value corresponding to the negative perturbation. When the unified total fitness function value corresponding to the positive perturbation is less than the unified total fitness function value corresponding to the negative perturbation, determine that the variable dimension is updated in the positive direction; otherwise, determine that the variable dimension is updated in the negative direction. Here, positive perturbation and negative perturbation refer to applying a preset small positive change and a negative change to a variable dimension based on its current value, respectively. The optimization adjustment direction refers to the update direction of the variable dimension that is more conducive to reducing the unified total fitness function value in the current iteration. In specific implementation, positive and negative perturbations can be applied to each variable dimension in each continuous variable vector, and the perturbed candidate solutions can be substituted into the unified total fitness function for evaluation, comparing the function values ​​corresponding to the two perturbations. If the function value after positive perturbation is smaller, it indicates that adjusting the variable in the positive direction is more conducive to improving the quality of the candidate solution; in this case, the optimization adjustment direction of the variable dimension is determined to be positive, otherwise it is determined to be negative. This allows for the acquisition of favorable update directions for continuous variables through local bidirectional trial and error without directly constructing complex analytical gradients, thereby enhancing the adaptability of continuous variable optimization.

[0075] Step 73: Determine the globally optimal continuous variable vector and the elite neighborhood optimal continuous variable vector based on the fitness evaluation value; determine the global learning coefficient, elite neighborhood learning coefficient, and gradient descent coefficient corresponding to the current iteration based on the total number of iterations and the initial value of the time-varying coefficients; update each continuous variable vector based on the globally optimal continuous variable vector, the elite neighborhood optimal continuous variable vector, and the optimization adjustment direction, and restrict them to the feasible ranges of the vertical anchor point ratio, the lateral offset, and the end-berthing length, respectively. The globally optimal continuous variable vector refers to the continuous variable vector corresponding to the candidate solution with the best fitness evaluation value in the current population; the elite neighborhood optimal continuous variable vector refers to the continuous variable vector with the best performance within the elite neighborhood of the current candidate solution; the global learning coefficient, the elite neighborhood learning coefficient, and the gradient descent coefficient are used to adjust the degree to which the current candidate solution moves closer to the globally optimal solution, the neighborhood optimal solution, and the current favorable update direction. In practice, the following steps can be taken: First, determine the global optimal candidate solution based on the fitness evaluation value, and then determine the corresponding elite neighborhood optimal vector for each candidate solution within the preset elite solution range. Next, determine the corresponding time-varying learning coefficient and descent coefficient based on the current iteration stage to control the influence intensity of different guiding factors. Finally, jointly update each continuous variable vector based on the global optimal continuous variable vector, the elite neighborhood optimal continuous variable vector, and the optimization adjustment direction determined in step 72. After the update, the vertical anchor point ratio should be limited to between zero and one, the lateral offset should be limited to the feasible lateral offset range of the corresponding corridor edge, and the end berthing length should be limited to between the minimum and maximum berthing length values. For example, in the early stages of iteration, the role of global learning and neighborhood learning can be appropriately enhanced to increase the search range; in the later stages of iteration, the role of small-step adjustments along favorable directions can be appropriately enhanced to improve the precision of parameter fine-tuning. By introducing a mechanism combining global optimal solution guidance, neighborhood optimal solution guidance, and direction trial update, continuous variable optimization can simultaneously consider search efficiency and local refinement capabilities.

[0076] Step 74: Write the updated continuous variable vectors back into the corresponding candidate solutions for the longitudinal anchor point ratio, lateral offset, and end-of-line berthing length, respectively, to obtain the fine-tuned composite candidate solutions. Substitute each fine-tuned composite candidate solution into the unified overall fitness function to obtain the corresponding fitness evaluation value. In specific implementation, the updated variable values ​​can be filled back into the positions of the longitudinal anchor point ratio, lateral offset, and end-of-line berthing length parameters of each corridor edge in the corresponding candidate solutions according to the original arrangement order of the parameters in the continuous variable vectors, forming new composite candidate solutions. Subsequently, the fine-tuned composite candidate solutions are substituted into the unified overall fitness function again to recalculate their fitness evaluation values. For example, if a candidate solution has a lateral offset closer to the middle of the road and an end-of-line berthing length that better matches the endpoint attitude requirements after the continuous variables are updated, its safety clearance evaluation term or end-of-line berthing error evaluation term may be improved, thereby reducing the overall fitness evaluation value. By re-evaluating the candidate solutions after the continuous variables are updated in a unified manner, it can be ensured that the results of the continuous variable fine-tuning are still consistently judged under the unified evaluation system.

[0077] Step 75: Select a preset number of elite solutions based on their fitness evaluation values, from best to worst. Use the total number of elite solutions as the denominator to determine the proportion of each corridor sequence appearing in the elite solutions. When the proportion of the same corridor sequence reaches a corridor locking ratio threshold for a consecutive preset number of generations, that corridor sequence is designated as the optimal corridor sequence. All corridor sequences from candidate solutions are then locked as the optimal corridor sequence, and subsequent updates only involve continuous variable vectors. The corridor locking ratio threshold is the threshold used to determine whether a corridor sequence has achieved sufficient stable dominance in the elite solutions. The consecutive preset number of generations refers to the number of consecutive iterations required for the proportion of the corridor sequence to continuously meet the threshold condition. In practice, after each round of continuous variable updates, several candidate solutions with the best fitness evaluation values ​​can be selected from the current population as elite solutions, and the proportion of each corridor sequence appearing in the elite solutions can be calculated. If a corridor sequence reaches or exceeds the corridor locking ratio threshold for several consecutive generations, it can be considered a stable and dominant topological path in the current search phase. In this case, it is designated as the optimal corridor sequence, and all corridor sequences in the candidate solutions are replaced, with only the differences in continuous variables retained for subsequent fine-tuning. For example, if a corridor sequence consistently accounts for the vast majority of elite solutions across multiple generations, it indicates that the current search has essentially confirmed that this topological framework is superior to other candidate paths. In this case, continuing to frequently change the corridor sequence is less meaningful, and it is more appropriate to shift the search focus to refining and optimizing continuous parameters such as anchor point ratio, lateral offset, and berthing length. By setting corridor locking conditions, the search range can be narrowed in a timely manner when the topology has stabilized, thereby improving the convergence speed and result stability of subsequent searches.

[0078] This embodiment integrates the longitudinal anchor point ratio, lateral offset, and end-of-pipe docking length into a continuous variable vector based on the initial population or updated candidate solutions. It then iteratively optimizes the continuous variables using bidirectional perturbation exploration, global optimal solution guidance, elite neighborhood optimal solution guidance, and favorable direction update mechanisms to obtain fine-tuned composite candidate solutions and their fitness evaluation values. Simultaneously, by statistically analyzing the occurrence ratio of corridor sequences in elite solutions and setting corridor locking conditions, the search focus shifts to fine-tuning the continuous variables after the dominant topology framework stabilizes. Therefore, while maintaining a relatively optimal path topology, it can further improve the boundary safety, trajectory smoothness, and end-of-pipe docking accuracy of unmanned vehicle path planning in closed parks, and enhance the convergence efficiency and optimization stability of the search process.

[0079] In one embodiment of the present invention, a deterministic executable trajectory is generated based on the fine-tuned composite candidate solution, the optimal corridor sequence, the dynamic no-entry time window, the docking convergence start point parameters, and the end docking length parameters. The deterministic executable trajectory refers to the final trajectory output result that simultaneously includes spatial position, heading angle, and reference speed, and satisfies road boundary constraints, vehicle kinematic constraints, end-point docking constraints, and dynamic obstacle timing avoidance constraints. Unlike methods that only provide the geometric path, the trajectory generated in this embodiment not only indicates the positions the autonomous vehicle should traverse but also provides the corresponding heading changes and speed references, thus directly providing input for subsequent control execution. By further transforming the fine-tuned composite candidate solution into a deterministic executable trajectory that satisfies geometric continuity, curvature constraints, speed constraints, and dynamic time window constraints, the path planning results of the present invention can be more applicable to the actual operation scenarios of autonomous vehicles in closed parks.

[0080] Specifically, in step 81, the fine-tuned composite candidate solution, the optimal corridor sequence, the docking convergence start point parameters, and the end docking length parameters are read. The planar positions of each anchor point determined by the fine-tuned composite candidate solution are extracted according to the optimal corridor sequence, and these positions, along with the start point position, the docking convergence start point parameters, and the end point position, are arranged in the driving sequence to form a trajectory point sequence. Here, the trajectory point sequence refers to a set of ordered control points used to construct a continuous geometric trajectory; the docking convergence start point parameters are used to characterize the starting position when the autonomous vehicle switches from a normal driving state to the end docking convergence state. In practice, the planar positions of each anchor point corresponding to the optimal corridor sequence can be extracted first from the fine-tuned composite candidate solution. Then, the start point position is used as the trajectory start control point, the docking convergence start point parameters are used as the end convergence control point, and the end point position is used as the trajectory termination control point, forming a complete trajectory point sequence according to the actual driving sequence of the autonomous vehicle. For example, when an autonomous vehicle needs to enter the pre-station guide lane along the main channel and finally stop, the anchor points on the main channel and guide lane can be connected sequentially, and a docking convergence start point can be inserted before the endpoint. This allows the trajectory point sequence to simultaneously meet the requirements of normal traffic and precise docking at the end. By constructing a trajectory point sequence that includes the start point, anchor points along the way, docking convergence start point, and endpoint, a clear geometric control basis can be provided for the subsequent generation of continuous curves.

[0081] Step 82: Based on the positional and tangential constraints of adjacent control points in the trajectory point sequence, construct cubic polynomial continuous curves for each segment to obtain the initial continuous geometric trajectory; calculate the continuous curvature of each segment, and shrink the tangential scale of continuous curves whose continuous curvature exceeds the physical limit curvature of the autonomous vehicle to obtain a continuous and compliant geometric trajectory for the entire path. Here, a cubic polynomial continuous curve refers to a cubic continuous curve segment jointly determined by the positional and tangential conditions between adjacent control points; continuous curvature refers to the degree of local bending at each position along the continuous curve; tangential scale shrinkage refers to the method of reducing the degree of local bending of the curve by reducing the tangential guide length at the control point. In specific implementation, cubic polynomial curves satisfying positional and tangential continuity can be constructed segment by segment according to the position and adjacent direction relationship of each control point in the trajectory point sequence to form the initial continuous geometric trajectory; subsequently, the continuous curvature of the initial continuous geometric trajectory is calculated segment by segment. If some curve segments have continuous curvature exceeding the physical limit curvature of the autonomous vehicle, the tangential scale at the corresponding control point is shortened to reduce the degree of bending of the curve until the corresponding curve segment meets the curvature constraint that the vehicle can pass through. For example, when a control point turns too sharply, the curve obtained by direct interpolation may locally overshoot. In this case, the path transition can be smoothed by shrinking the tangential scale of that segment. By further generating a continuous and compliant geometric trajectory that satisfies curvature constraints from the trajectory point series, it can be ensured that the subsequent output path is both continuous and smooth, and has actual vehicle trackability.

[0082] Step 83: Configure the full-path reference speed curve based on the continuous compliant geometric trajectory of the entire path; for the ordinary traffic segment, determine the speed limit constraint of the park road network and the speed constraint determined by the maximum lateral acceleration of the vehicle and the trajectory curvature, and take the smaller value of the two as the reference speed; for the berthing segment, start from the berthing segment entrance speed and decrease according to the quadratic curve, and set the reference speed to zero at the endpoint corresponding to the berthing length parameter. Here, the full-path reference speed curve refers to the speed reference sequence distributed along the continuous compliant geometric trajectory of the entire path; the ordinary traffic segment refers to the regular driving path segment before the berthing convergence start point; the berthing segment refers to the final convergence path segment between the berthing convergence start point and the endpoint. In practical implementation, in ordinary traffic sections, both the speed limits of the park roads and the upper limit of safe speed determined by the path curvature and the maximum lateral acceleration of the vehicle can be considered simultaneously. The smaller of the two values ​​is taken as the reference speed at that location, thus balancing road management constraints and vehicle movement safety constraints. In the docking section, the reference speed is gradually reduced from the speed at the entrance of the docking section to a smoothly decreasing quadratic curve, and then set to zero at the endpoint corresponding to the docking length parameter at the end, to ensure that the unmanned vehicle decelerates smoothly and completes a precise stop. For example, the unmanned vehicle can travel at a higher speed in the main channel, while gradually decelerating to zero speed in the docking section near the loading and unloading station. By configuring the reference speed curve along a continuous and compliant geometric trajectory throughout the entire path, the final trajectory can not only meet spatial accessibility requirements but also speed control and terminal zero-speed docking requirements.

[0083] Step 84: Establish a mapping relationship between trajectory arc length and travel time based on the full-path reference speed curve; for each corridor unit, extract the arc length positions corresponding to the trajectory entering and leaving the corridor unit, and determine the entry and exit times through the mapping relationship to form the travel time interval of the corridor unit; perform intersection verification between the travel time interval of each corridor unit and the corresponding dynamic restricted time window one by one. The mapping relationship between trajectory arc length and travel time is used to characterize the time corresponding to the unmanned vehicle traveling along the trajectory to each arc length position; the travel time interval refers to the start and end time range of the unmanned vehicle passing through a certain corridor unit; the intersection verification refers to comparing whether there is overlap between the travel time interval and the dynamic restricted time window. In practice, the travel time at each position on the trajectory can be accumulated based on the full-path reference speed curve to establish a correspondence between arc length and time. Then, for each corridor unit, the arc length values ​​corresponding to the entry and exit positions are determined, and the corresponding entry and exit times are obtained through the arc length-time mapping relationship, thus forming the travel time interval for that corridor unit. Finally, the travel time interval of each corridor unit is compared one by one with its corresponding dynamic no-entry time window to determine whether the trajectory is passable in the dynamic time dimension. For example, if the dynamic no-entry time window of a corridor unit is 10 to 14 seconds, and the autonomous vehicle is expected to enter the corridor unit at 8 seconds and leave at 9 seconds, then there is no overlap. Conversely, if the autonomous vehicle is expected to pass through the corridor unit at 11 to 12 seconds, it indicates that the trajectory conflicts with the time period occupied by dynamic obstacles at that position. By performing unit-by-unit temporal verification after trajectory generation, the final output can simultaneously satisfy geometric constraints and dynamic obstacle avoidance temporal constraints.

[0084] Step 85: When the travel time intervals of each corridor unit do not overlap with the corresponding dynamic restricted time windows, the continuous and compliant geometric trajectory and the reference speed curve of the entire path are integrated into a deterministic executable trajectory including position, heading angle, and reference speed. When the travel time interval of any corridor unit overlaps with the corresponding dynamic restricted time window, the corresponding trajectory is determined as a temporally conflicting trajectory, and candidate solutions are re-determined based on the temporally conflicting trajectory until the travel time intervals of each corridor unit do not overlap with the corresponding dynamic restricted time windows. A temporally conflicting trajectory refers to a trajectory that, although geometrically continuous and feasible, overlaps with the dynamic restricted time window in some corridor units. In practical implementation, when the travel time intervals of all corridor units do not intersect with the corresponding dynamic restricted time windows, the continuous compliant geometric trajectory and the reference speed curve of the entire path can be uniformly encapsulated to form the final deterministic executable trajectory. Position describes the trajectory shape, heading angle describes vehicle attitude changes, and reference speed describes the speed control target. If any corridor unit experiences temporal overlap, it indicates that the current trajectory cannot safely traverse the dynamic environment. In this case, it is determined to be a temporally conflicting trajectory, and candidate solutions are readjusted based on this conflict information until a trajectory satisfying all temporal constraints is regenerated. By introducing a final temporal consistency check and conflict rollback mechanism, it can be ensured that the trajectory output by this invention is not only executable on static road structures but also truly traversable in dynamic obstacle environments.

[0085] This embodiment transforms the fine-tuned composite candidate solution and optimal corridor sequence into a trajectory point sequence containing a starting point, anchor points along the route, a docking convergence start point, and an end point. It further generates a continuous and compliant geometric trajectory that satisfies curvature constraints. Then, it configures a reference speed curve for the entire path based on path curvature, road speed limits, and docking zero-speed requirements. Finally, it verifies the travel time interval of each corridor unit according to a dynamic restricted time window and outputs a deterministic executable trajectory, achieving a complete conversion from candidate paths to directly executable trajectories. Therefore, the path planning results for unmanned vehicles in closed parks can simultaneously meet the requirements of geometric continuity, vehicle trackability, dynamic obstacle avoidance timing constraints, and precise end-point docking, thereby improving the engineering executability and practical application value of the planning output.

[0086] It should be noted that the interval and threshold sizes are set for ease of comparison. The size of the threshold depends on the amount of sample data and the base number set by those skilled in the art for each set of sample data, as long as it does not affect the proportional relationship between the parameter and the quantized value. Furthermore, the above formulas are all dimensionless calculations, and the formulas are derived from software simulations using a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0087] The embodiments of the present invention have been described above, but the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of the present embodiments, all of which are within the protection scope of the present embodiments.

Claims

1. A path planning method for unmanned vehicles in closed parks based on intelligent optimization algorithms, characterized in that, Includes the following steps: Step 1: Obtain basic geographic data of park roads, attribute data of park roads, real-time monitoring data of dynamic obstacles, basic status of the starting point of the unmanned vehicle and basic status of the ending point of the unmanned vehicle, and determine the directed corridor map of the park, dynamic restricted time window and start and end status parameters of the unmanned vehicle. Step 2: Define composite candidate solutions based on the directed corridor graph of the park and the start and end state parameters of the unmanned vehicles, and determine the feasible range of lateral offset, the starting point parameter of docking convergence and the end docking length parameter. Step 3: Based on the directed corridor map of the park, the feasible range of lateral offset, and the composite candidate solution, determine the candidate corridor skeleton, the initial population, and the candidate bypass sub-path library; Step 4: Based on the dynamic restricted access time window, the initial population, and the basic parameters of the unmanned vehicle, construct a unified overall fitness function and determine the fitness evaluation value of each candidate solution in the initial population; Step 5: Determine the algorithm convergence slope and corridor entropy based on the fitness evaluation value in the most recent preset number of iterations, and determine the squirrel search mode switching result based on the algorithm convergence slope and corridor entropy; Step 6: When the squirrel search mode switching result is a jump search, based on the unified overall fitness function and the candidate detour sub-path library, the high-cost corridor sub-segments in the candidate solutions are replaced to obtain updated candidate solutions, and the fitness evaluation value of the updated candidate solutions is determined. Step 7: Based on the initial population or updated candidate solutions, update the continuous variables in each candidate solution according to the unified total fitness function to obtain the fine-tuned composite candidate solutions and their corresponding fitness evaluation values, and determine the optimal corridor sequence according to the preset corridor locking conditions. Step 8: Generate a deterministic executable trajectory based on the fine-tuned composite candidate solution, the optimal corridor sequence, the dynamic no-entry time window, the berthing convergence start point parameters, and the end berthing length parameters.

2. The method for unmanned vehicle path planning in a closed park based on intelligent optimization algorithm according to claim 1, characterized in that, Acquire basic geographic data of park roads, park road attribute data, real-time monitoring data of dynamic obstacles, basic status of the starting point and the ending point of autonomous vehicles, and determine the directed corridor map of the park, dynamic restricted access time windows, and autonomous vehicle start and end status parameters, including: Step 11: Analyze the basic geographic data of the park roads to obtain intersection location data, task station location data, road segment boundary data, and actual spatial location data of road segments; Step 12: Analyze the road attribute data in the park and map the road width, road boundary safety margin, maximum allowable driving curvature, maximum allowable driving speed and road risk coefficient to the road segments represented by the actual spatial location data of each road segment; Step 13: Based on the intersection location data, task station location data, road segment boundary data, actual spatial location data of road segments, and the park road attribute data represented by the actual spatial location data of each road segment, construct a directed corridor map of the park. The directed corridor map of the park includes intersection nodes determined based on the intersection location data, task station nodes determined based on the task station location data, corridor edges determined based on the actual spatial location data of road segments, and the road width, road boundary safety margin, maximum allowable driving curvature, maximum allowable driving speed, and road risk coefficient corresponding to each corridor edge. Step 14: Based on the starting point basic state and the ending point basic state of the unmanned vehicle, determine the starting and ending state parameters of the unmanned vehicle. The starting and ending state parameters of the unmanned vehicle include the starting position, the starting heading angle, the starting speed, the ending position, the ending target heading angle, and the ending speed, and set the ending speed to zero. Step 15: Discretize each corridor edge in the directed corridor map of the park into corridor units according to a preset length interval, determine the occupancy time interval of each dynamic obstacle in each corridor unit based on the real-time monitoring data of dynamic obstacles, and merge the occupancy time intervals corresponding to the same corridor unit to obtain the dynamic prohibition time window corresponding to each corridor unit.

3. The method for unmanned vehicle path planning in a closed park based on intelligent optimization algorithm according to claim 1, characterized in that, Based on the directed corridor map of the park and the start and end state parameters of the autonomous vehicles, a composite candidate solution is defined, and the feasible range of lateral offset, the starting point parameters of docking convergence, and the end docking length parameters are determined, including: Step 21: Define a composite candidate solution based on the directed corridor map of the park and the start and end state parameters of the unmanned vehicle. The composite candidate solution includes the corridor sequence, longitudinal anchor point ratio, lateral offset and end docking length parameters. Step 22: For each corridor edge in the corridor sequence, multiply the corresponding longitudinal anchor point ratio by the length of the corridor edge to obtain the longitudinal positioning length, determine the longitudinal positioning point according to the center line of the corridor edge, and then determine the plane position of the anchor point along the normal direction of the longitudinal positioning point by the corresponding lateral offset. Step 23: For each corridor edge in the corridor sequence, read the road width and road boundary safety margin corresponding to the corridor edge, subtract the road boundary safety margin from half of the road width to obtain the allowable lateral offset distance on one side, and determine the feasible lateral offset range corresponding to the corridor edge based on the positive and negative ranges of the allowable lateral offset distance on one side. Step 24: Read the actual heading angle, corresponding lateral offset, target heading angle, and basic berthing length configuration data of the last corridor in the corridor sequence. Normalize the difference between the actual heading angle of the last corridor and the target heading angle to obtain the heading error. Add the product of the heading error and the heading attitude error adjustment coefficient, and the product of the absolute value of the lateral offset and the lateral offset adjustment coefficient to the berthing length reference value, respectively. Limit the superposition result to the minimum and maximum berthing length values ​​to obtain the end berthing length parameters. Step 25: Determine the berthing convergence start point parameters based on the endpoint position, the endpoint target heading angle, and the end berthing length parameters. Specifically, determine the endpoint target heading direction based on the endpoint target heading angle, and offset the position along the opposite direction of the endpoint target heading direction according to the end berthing length parameters, using the endpoint position as the starting point. The offset position is then determined as the berthing convergence start point parameters.

4. The closed-loop unmanned vehicle path planning method based on intelligent optimization algorithm according to claim 3, characterized in that, Based on the directed corridor map of the park, the feasible range of lateral offset, and the composite candidate solutions, the candidate corridor skeleton, initial population, and candidate bypass sub-path library are determined, including: Step 31: Based on the starting point and ending point, determine the corridor edge corresponding to the starting point and the corridor edge corresponding to the ending point in the directed corridor graph of the park, and extract the corridor sequence that is fully connected between the corridor edge corresponding to the starting point and the corridor edge corresponding to the ending point. Step 32: For each corridor sequence, calculate the weighted result of corridor length and length weight, road risk coefficient and road risk weight, and average curvature and average curvature weight for each corridor edge, and superimpose the weighted result of corridor edge number and corridor number weight to obtain the skeleton cost; determine a preset number of candidate corridor skeletons from low to high according to the skeleton cost. Step 33: For each corridor edge in each candidate corridor skeleton, read the corresponding feasible lateral offset range. Divide the total lateral span between the negative and positive lateral offset distances within the feasible lateral offset range of the corridor edge into equal intervals according to the adjacent level intervals corresponding to the number of lateral levels to obtain the level spacing. Take the position corresponding to the negative lateral offset distance as the starting level position, and stack the level spacing level by level according to the lateral level number to obtain the fixed level position corresponding to each lateral level. Set the longitudinal anchor point ratio of each corridor edge to the midpoint ratio from the corridor entrance to the corridor exit. Step 34: Combine the candidate corridor skeletons, the longitudinal anchor point ratios of each corridor side, and the positions of each fixed level into an initial composite candidate solution. Determine the end berthing length parameter corresponding to the initial composite candidate solution according to the calculation rules of Step 24. Add the end berthing length parameter to the initial composite candidate solution to obtain the initial population. Step 35: Generate corresponding connected detour sub-paths for each road connection node pair in the directed corridor graph of the park; calculate the skeleton cost of each connected detour sub-path according to the skeleton cost calculation method in Step 32, sort them from low to high skeleton cost, and then classify and integrate them according to road connection node pairs to obtain a candidate detour sub-path library.

5. The method for unmanned vehicle path planning in a closed park based on intelligent optimization algorithm according to claim 1, characterized in that, Based on the dynamic no-entry time window, the initial population, and the basic parameters of the autonomous vehicle, a unified overall fitness function is constructed, and the fitness evaluation value of each candidate solution in the initial population is determined, including: Step 41: Read each candidate solution in the initial population, accumulate the distance between adjacent anchor point plane positions for each candidate solution in the driving order to obtain the path length evaluation term; calculate the actual distance from each anchor point plane position to the nearest road boundary, square the part of the difference between the safety gap threshold and the actual distance that is greater than zero point by point and accumulate it to obtain the safety gap evaluation term; Step 42: For each candidate solution, calculate the discrete curvature corresponding to each anchor point based on the geometric relationship between the planar positions of three adjacent anchor points, and determine the difference between adjacent discrete curvatures as the curvature change; squar the portion of discrete curvature that exceeds the maximum allowable curvature of the path point by point and accumulate it, and then accumulate the curvature change and curvature change penalty coefficient to obtain the curvature and curvature change evaluation item; divide the distance between the planar positions of each adjacent anchor point by the preset reference speed of the corresponding path segment and accumulate it in the order of travel to obtain the time for each anchor point to reach the corresponding corridor unit, then compare the time for each anchor point to reach the corresponding corridor unit with the dynamic prohibition time window one by one, assign a preset dynamic conflict penalty coefficient to each anchor point that falls into the dynamic prohibition time window, and accumulate the preset dynamic conflict penalty coefficients corresponding to all anchor points to obtain the dynamic obstacle timing conflict evaluation item; Step 43: For each candidate solution, calculate the distance error between the endpoint of the final trajectory and the endpoint position, and calculate the heading error between the heading angle of the endpoint of the final trajectory and the heading angle of the endpoint target; sum the squared distance error with the heading error and the terminal heading error penalty coefficient to obtain the terminal docking error evaluation item; sum the squared portion of the discrete curvature that exceeds the physical limit curvature of the unmanned vehicle point by point to obtain the vehicle trackability evaluation item; count the number of corridor sequence switching to obtain the corridor switching number evaluation item. Step 44: Multiply the path length evaluation item, safety gap evaluation item, curvature and curvature change evaluation item, dynamic obstacle timing conflict evaluation item, end-stop error evaluation item, vehicle trackability evaluation item, and corridor switching number evaluation item with their corresponding weight configuration coefficients and sum them to obtain a unified overall fitness function; for each candidate solution in the initial population, extract the corresponding evaluation items and substitute them into the unified overall fitness function to obtain the fitness evaluation value of each candidate solution.

6. The method for unmanned vehicle path planning in a closed park based on intelligent optimization algorithm according to claim 1, characterized in that, The algorithm's convergence slope and corridor entropy are determined based on the fitness evaluation values ​​from the most recent preset number of iterations. The squirrel search mode switching results are then determined based on the algorithm's convergence slope and corridor entropy, including: Step 51: Extract the optimal fitness evaluation value of each generation in the most recent preset number of iterations, use the iteration number as the independent variable and the optimal fitness evaluation value as the dependent variable to perform linear regression, and determine the linear regression slope as the algorithm convergence slope. Step 52: Select a preset number of candidate solutions as elite solutions according to the fitness evaluation value from best to worst, count the number of times each corridor edge appears in the corridor sequence of the elite solution, and determine the appearance ratio of each corridor edge by the proportion of the appearance frequency to the total appearance frequency of all corridor edges. Step 53: For each corridor edge with an occurrence ratio greater than zero, perform a logarithmic operation on the occurrence ratio and multiply it by the corresponding occurrence ratio, then sum the results. Combine this with the total number of corridor edges in the directed corridor graph of the park to normalize the data and determine the corridor entropy. Step 54: Define the convergence threshold where the algorithm's convergence slope is greater than or equal to a negative convergence threshold as a convergence stagnation state, and define the corridor entropy as less than the minimum corridor entropy as a state of insufficient corridor diversity. Step 55: When the convergence stagnation state or the corridor diversity insufficiency state is established, determine the squirrel search mode switching result as a jump search; when the convergence stagnation state and the corridor diversity insufficiency state are not established, determine the squirrel search mode switching result as an asymptotic search.

7. The method for unmanned vehicle path planning in a closed park based on intelligent optimization algorithm according to claim 1, characterized in that, When the squirrel search mode switch result is a jump search, based on the unified overall fitness function and the candidate detour sub-path library, high-cost corridor segments in the candidate solutions are replaced to obtain updated candidate solutions, and the fitness evaluation value of the updated candidate solutions is determined, including: Step 61: When the squirrel search mode switching result is jump search, for each corridor sub-segment in each candidate solution, sum the weighted result of the sub-segment length and length weight, the weighted result of the square of the greater than zero part of the difference between the safety gap threshold and the actual distance and the safety obstacle avoidance weight, the weighted result of the discrete curvature and curvature weight, and the weighted result of the dynamic no-entry time window hit result and the dynamic obstacle avoidance weight to obtain the comprehensive cost; Step 62: Compare the comprehensive costs of each corridor segment in each candidate solution, determine the high-cost corridor segment with the largest comprehensive cost, and determine the starting connection node of the high-cost corridor segment in the original corridor sequence as the front connection node and the ending connection node as the back connection node. Step 63: Based on the preceding and following nodes, extract the corresponding candidate detour sub-path library from the candidate detour sub-path library; based on the sum of the current iteration number and the current candidate solution number, perform a cyclic indexing based on the number of sub-paths in the candidate detour sub-path library to determine the alternative sub-path; Step 64: Replace the high-cost corridor segments in the original corridor sequence with alternative sub-paths to obtain a new corridor sequence; inherit the original longitudinal anchor point ratio and the original lateral offset parameters for the corridor edges in the new corridor sequence that correspond to the original corridor sequence; map the original longitudinal anchor point ratio and the original lateral offset parameters for the newly added corridor edges in relative order, and ensure that the original lateral offset parameters are within the lateral offset feasible range of the corresponding corridor edges; retain the end berthing length parameter to obtain the updated candidate solution. Step 65: Substitute the updated candidate solution into the unified overall fitness function, recalculate the path length evaluation item, safety gap evaluation item, curvature and curvature change evaluation item, dynamic obstacle timing conflict evaluation item, end-stop error evaluation item, vehicle trackability evaluation item, and corridor switching number evaluation item, and determine the fitness evaluation value of the updated candidate solution.

8. The method for unmanned vehicle path planning in a closed park based on intelligent optimization algorithm according to claim 1, characterized in that, Based on the initial population or updated candidate solutions, the continuous variables in each candidate solution are updated according to the unified overall fitness function to obtain the fine-tuned composite candidate solutions and their corresponding fitness evaluation values. The optimal corridor sequence is then determined according to preset corridor locking conditions, including: Step 71: When there is an updated candidate solution obtained by the jump search, read the updated candidate solution and its fitness evaluation value; otherwise, read the initial population and its fitness evaluation value; and integrate the vertical anchor point ratio, lateral offset and end berthing length in each candidate solution into a continuous variable vector. Step 72: For each variable dimension of each continuous variable vector, calculate the unified total fitness function value corresponding to the positive perturbation and the unified total fitness function value corresponding to the negative perturbation respectively; when the unified total fitness function value corresponding to the positive perturbation is less than the unified total fitness function value corresponding to the negative perturbation, determine that the variable dimension is updated in the positive direction; otherwise, determine that the variable dimension is updated in the negative direction. Step 73: Determine the global optimal continuous variable vector and the elite neighborhood optimal continuous variable vector based on the fitness evaluation value; determine the global learning coefficient, elite neighborhood learning coefficient, and gradient descent coefficient corresponding to the current iteration based on the total number of iterations and the initial value of the time-varying coefficients; update each continuous variable vector based on the global optimal continuous variable vector, the elite neighborhood optimal continuous variable vector, and the optimization adjustment direction, and restrict them to the feasible range of the vertical anchor point ratio, the lateral offset, and the end berthing length, respectively. Step 74: Write the updated continuous variable vectors back to the longitudinal anchor point ratio, lateral offset and end berthing length in the corresponding candidate solutions to obtain the fine-tuned composite candidate solutions; substitute each fine-tuned composite candidate solution into the unified overall fitness function to obtain the corresponding fitness evaluation value. Step 75: Select a preset number of elite solutions from best to worst fitness evaluation value, and determine the appearance ratio of each corridor sequence in the elite solutions with the total number of elite solutions as the denominator; when the appearance ratio of the same corridor sequence reaches the corridor locking ratio threshold for a preset number of consecutive algebras, the corridor sequence is determined as the optimal corridor sequence, and the corridor sequences of all candidate solutions are locked as the optimal corridor sequence, and only the continuous variable vector is updated thereafter.

9. The method for unmanned vehicle path planning in a closed park based on intelligent optimization algorithm according to claim 1, characterized in that, Based on the fine-tuned composite candidate solution, optimal corridor sequence, dynamic no-entry time window, berthing convergence start point parameters, and end-berthing length parameters, a deterministic executable trajectory is generated, including: Step 81: Read the fine-tuned composite candidate solution, optimal corridor sequence, berthing convergence start point parameters, and end berthing length parameters; extract the plane position of each anchor point determined by the fine-tuned composite candidate solution according to the optimal corridor sequence, and form a trajectory point sequence with the start point position, berthing convergence start point parameters, and end point position in the driving sequence; Step 82: Based on the positional and tangential constraints of adjacent control points in the trajectory point sequence, construct cubic polynomial continuous curves for each segment to obtain the initial continuous geometric trajectory; calculate the continuous curvature of each segment, and shrink the tangential scale of continuous curves whose continuous curvature exceeds the physical limit curvature of the unmanned vehicle to obtain a continuous and compliant geometric trajectory for the entire path. Step 83: Configure the full-path reference speed curve based on the full-path continuous compliant geometric trajectory; for ordinary passage sections, determine the speed limit constraint of the park road network and the speed constraint determined by the maximum lateral acceleration of the vehicle and the trajectory curvature respectively, and take the smaller value of the two as the reference speed; for berthing sections, start from the berthing section entrance speed and decrease according to the quadratic curve, and set the reference speed to zero at the end point corresponding to the berthing length parameter at the end. Step 84: Establish the mapping relationship between trajectory arc length and travel time based on the full-path reference speed curve; for each corridor unit, extract the arc length positions corresponding to the trajectory entering and leaving the corridor unit, and determine the entry time and exit time through the mapping relationship to form the travel time interval of the corridor unit; perform intersection verification between the travel time interval of each corridor unit and the corresponding dynamic restricted time window one by one. Step 85: When the travel time interval of each corridor unit does not overlap with the corresponding dynamic restricted time window, the continuous compliant geometric trajectory and the reference speed curve of the entire path are integrated into a deterministic executable trajectory that includes position, heading angle and reference speed; when the travel time interval of any corridor unit overlaps with the corresponding dynamic restricted time window, the corresponding trajectory is determined as a temporal conflict trajectory, and candidate solutions are re-determined based on the temporal conflict trajectory until the travel time interval of each corridor unit does not overlap with the corresponding dynamic restricted time window.

Images