A task-oriented container wharf intelligent transfer vehicle lane-level route planning method and system

Abstract

The application relates to a task-oriented intelligent transfer vehicle lane-level route planning method and system in a container terminal, which comprises the following steps: S1, obtaining a directed grid map by modeling a container terminal road network, and obtaining task state information of each intelligent transfer vehicle at each time step; S2, generating a candidate action set for each intelligent transfer vehicle at each time step, and mapping the candidate action set into a next time step occupation request set; S3, collecting all next time step occupation request sets, performing conflict detection and priority arbitration, obtaining an occupation granting result of each intelligent transfer vehicle, and performing according to the occupation granting result; and S4, returning to step S2 for time step advancement and circulation on the basis of the occupation granting result execution, until each intelligent transfer vehicle completes the task, and a real-time evolving lane-level track is obtained. Compared with the prior art, the application has the advantages of improving real-time response and conflict control capability.

Description

Technical Field

[0001] This invention relates to the field of container terminal operation optimization, and in particular to a task-oriented intelligent transfer vehicle lane-level route planning method and system within a container terminal. Background Technology

[0002] With the automation and digital transformation of container terminals, intelligent transfer vehicles have been widely used in terminal operations due to their flexible deployment and low infrastructure dependence. However, container terminals have dense road networks with numerous intersections and complex merging zones, and their operational tasks are highly variable. When multiple intelligent transfer vehicles operate in parallel within the same spatial and temporal area, frequent vehicle interactions can lead to path conflicts and resource competition, especially in micro-conflict scenarios such as intersections and merging zones, where traffic competition is particularly prominent. Traditional route planning methods are insufficient to effectively meet the demands for efficient and safe operations, necessitating new solutions.

[0003] In existing technologies, the vehicle routing problem at container terminals is typically addressed by simplifying the road network structure to reduce computational complexity. For example, multi-lane roads are abstracted into single logical links, or coarse-grained node-edge connectivity is used to describe feasible vehicle paths [Cao Y, Yang A, Liu Y, et al. AGV dispatching and bidirectional conflict-free routing problem in automated container terminal. Computers & Industrial Engineering, 2023, 184: 109611]. While these methods can improve planning efficiency to some extent, their limited ability to characterize lane-level spatial structures and constraints makes it difficult to accurately describe micro-level conflicts between vehicles when dealing with complex local scenarios such as intersections and merging zones. Consequently, the planning results often require frequent adjustments during actual operation, potentially impacting traffic efficiency.

[0004] Furthermore, existing multi-vehicle routing and conflict resolution methods are mostly based on preset rules or general decision criteria for coordination [Li S, Fan L, Jia S. A hierarchical solution framework for dynamic and conflict-free AGV scheduling in an automated container terminal. Transportation Research Part C: Emerging Technologies, 2024, 165: 104724]. The decision-making process typically does not fully consider the differences in timeliness requirements and vehicle load status among different transportation tasks. When multiple vehicles simultaneously use limited road or lane resources, the relevant coordination strategies still have limitations in balancing the operational needs of different vehicles. Under operating conditions of continuously arriving tasks and constantly changing traffic conditions, existing methods also need further improvement in the timeliness and flexibility of route adjustments to maintain the stable and efficient operation of the system.

[0005] In summary, how to achieve precise modeling and effective constraints of lane-level road resources in the complex road network environment of container terminals, and how to rationally coordinate the traffic relationships between vehicles during multi-vehicle parallel operation to improve the system's operational efficiency and safety under dynamic operating conditions, remains a technical problem that needs further research and solutions in this field. Summary of the Invention

[0006] The purpose of this invention is to provide a task-oriented method and system for lane-level route planning of intelligent transfer vehicles in container terminals, which improves real-time safety and operational efficiency.

[0007] The objective of this invention can be achieved through the following technical solutions: A task-oriented method for lane-level route planning of intelligent transfer vehicles in container terminals includes the following steps: S1. Obtain the directed mesh graph obtained by discretizing the lane-level model of the container terminal vehicle road network. It also obtains the task status information of each intelligent transfer vehicle at each time step, wherein each node in the directed mesh graph represents a passable mesh cell. For nodes, It is a directed edge; S2. At each time step, based on the directed mesh graph Based on task status information, a set of candidate actions is generated for each intelligent transfer vehicle, and based on directed edges... Map the candidate action set to the next time step occupancy request set; S3. Collect the next time step occupancy request set of all intelligent transfer vehicles, perform conflict detection and priority adjudication, obtain the occupancy grant result of each intelligent transfer vehicle for the next time step, and execute according to the occupancy grant result of the next time step; S4. Based on the execution of the occupancy grant result in the next time step, return to step S2 to advance and loop through the time steps until each intelligent transfer vehicle completes its task and obtains the real-time evolving lane-level trajectory.

[0008] Furthermore, the task status information includes the task starting point. Current destination Remaining time for the task Task Type Load condition Current occupied set subscript Indicates the first Intelligent transfer vehicle, underlined Indicates intelligent transfer vehicle The number of grid cells occupied at each time step is expressed as: , In the formula, For intelligent transfer vehicles The length of the car body, For driving distance, For vehicle width, This refers to the lane width.

[0009] Furthermore, the step of generating the candidate action set includes: Based on the task status information, as well as road network connectivity constraints and vehicle kinematic constraints, each intelligent transfer vehicle... Generate a set of candidate actions , Indicates intelligent transfer vehicle The driving action, Indicates intelligent transfer vehicle At this moment of decision-making The selected current destination, where the road network connectivity constraint represents the intelligent transfer vehicle. The vehicle can only move along the directions permitted in the directed mesh graph, and the vehicle kinematic constraints include at least the intelligent transport vehicle. Move between adjacent reachable grid cells according to the current orientation and the selected candidate action.

[0010] Furthermore, the mapping step for the next time step occupancy request set includes: Obtain a motion envelope template, which is indexed by {action type, orientation, step number}; According to the directed edge And the candidate actions selected from the candidate action set, to calculate the intelligent transfer vehicle The position and orientation of the reference point at the next time step are calculated using the following expression: , In the formula, For intelligent transfer vehicles At any moment Location and orientation It is determined by the action type, vehicle orientation, and a preset turning envelope template. For intelligent transfer vehicles At any moment Position and orientation For intelligent transfer vehicles The driving action; Query the intelligent transfer vehicle The motion envelope template corresponding to the reference point, the currently selected candidate action, orientation, and number of steps executed is queried to obtain the offset set relative to the reference point. ,in, This refers to the stagger sequence number of the current candidate action during continuous execution; The offset set By translating the position and orientation to the next time step, we obtain the intelligent transfer vehicle. The next time step occupies the request set , For intelligent transfer vehicles The next driving action in time step offset set The single offset relative to the reference point in the middle.

[0011] Furthermore, the types of conflicts include: Conflict of Occupation at the Same Point: There are requests to occupy the same grid cell in the next time step occupancy request sets of different intelligent transfer vehicles; Opposing conflict: There are adjacent grid cell regions in front of each other that are requesting to enter in opposite directions in the next time step occupancy request set of two intelligent transfer vehicles. Cross-over conflict: The next time step occupancy request sets of different intelligent transfer vehicles overlap in the cross-over area, or they request to enter the same cross-over area from different directions.

[0012] Furthermore, the steps of conflict detection and priority determination include: Initialize the semaphore state of each grid cell , Indicates the current time , For grid cells The maximum capacity per grid cell Satisfy capacity constraints; Define a global occupancy table, which is used to record the current time step. The occupancy status of each grid cell and the planned occupancy status for the next time step; Determine whether the next time step occupancy request sets of all intelligent transfer vehicles will overlap on any grid cell. If so, mark the overlapping grid cells as contentious resources and form a contentious vehicle set. If not, directly grant occupancy rights to the next time step occupancy request sets of intelligent transfer vehicles that will not overlap, wherein occupancy mutual exclusion constraints are used for the determination. Based on the remaining time of the task Task Type Load condition Task Start Point Current destination Current occupied set In addition to historical operational information, the system calculates the number of intelligent transfer vehicles in the disputed vehicle set. Dynamic priority scalar ; Based on the current state of each competing resource semaphore states Combined with the next time step The reserved occupancy status, when the intelligent transfer vehicle in the set of competing vehicles... The next time step's occupancy request set satisfies the capacity constraint of the corresponding contested resource. This indicates that the intelligent transfer vehicle It has the qualification to grant occupancy, and it has authority over all intelligent transfer vehicles that have the qualification to grant occupancy. According to the dynamic priority scalar Granting a contested resource, and updating the semaphore state of the corresponding resource after each grant. ;when When the time step is reached, it indicates that the resource being contested has no remaining capacity in the next time step, and new requests for occupancy will no longer be granted. For intelligent transfer vehicles that have not been granted occupancy authorization The system either remains in its current position and waits, or removes conflicting candidate actions from the candidate action set and regenerates a new set of occupancy requests for the next time step, then re-enters the conflict detection and priority adjudication process within the same time step, continuing this process until the intelligent transfer vehicle... Obtain the right to occupy, or reach the preset limit for the number of reselections; For intelligent transfer vehicles that have been granted occupancy rights The intelligent transfer vehicle will then execute the corresponding action for the next time step. The occupancy set is updated in the next time step. Then release the intelligent transfer vehicle. The system handles contention for resources at the current time step and updates the global resource allocation table synchronously to complete the switching of resource allocation status from the current time step to the next time step.

[0013] Furthermore, the capacity constraint is: , In the formula, A collection of intelligent transfer vehicles, This is an indicator function; it takes the value 1 when the condition inside the parentheses is true, and 0 otherwise. The occupancy mutual exclusion constraint is as follows: , In the formula, For intelligent transfer vehicles The next time step to occupy the request set, For intelligent transfer vehicles The next time step occupies the request set. It is an empty set.

[0014] Furthermore, in calculating the dynamic priority scalar At the same time, the road network structure perception coding results are also introduced, and the road network structure perception coding results are compared with those of the intelligent transfer vehicle. The local states are concatenated to form the extended state input. To utilize the extended state input Calculate the dynamic priority scalar ,in, The steps for obtaining the road network structure-aware coding results, which are feasible reference paths generated by a heuristic search algorithm, include: The directed mesh graph Each node in the construct features vector , where subscript For the first The feature vector of the node At a minimum, it should include information on occupancy status, obstacle identification, and area type; The directed mesh graph In the input graph coding model, the feature vectors of each node are used as the basis for input. Perform neighborhood-based weighted aggregation to obtain the updated node embeddings. , represented as: , In the formula, For activation function, For attention weights, For learnable weight matrix, For nodes The set of adjacent nodes, For the first The feature vector of each node; Embed the updated nodes corresponding to all nodes Perform aggregation processing to form a global feature vector representing the current road network structure. This serves as the result of the road network structure perception coding.

[0015] Furthermore, the step of obtaining the feasible reference path includes: In the directed mesh graph Using the task start point and candidate destination as endpoints, and with obstacle nodes, prohibited grid cells, unreachable direction edges, and static capacity limits as constraints, a heuristic search algorithm is used to find the shortest feasible path from the current start point to the target location, which serves as the feasible reference path. ; The feasible reference path mentioned above As prior knowledge, it is embedded into the task state information so that after generating a set of candidate actions at each time step, the current candidate action is compared with the feasible reference path. Perform alignment comparison: If a candidate action causes the intelligent transfer vehicle to... If the candidate action is kept on the feasible reference path or moves toward the next key node on the feasible reference path, the sampling probability of the candidate action is increased. If a candidate action deviates from the feasible reference path by more than a set threshold and does not bring conflict avoidance benefits, the selection tendency of the candidate action is reduced. When a conflict is detected, the intelligent transfer vehicle is allowed. The path may temporarily deviate from the feasible reference path, and once the conflict is resolved, it may revert to the feasible reference path to continue moving forward, or it may be based on an intelligent transfer vehicle. The heuristic search algorithm is re-invoked to generate an updated feasible reference path based on the current starting point and the current target position.

[0016] This invention also provides a task-oriented intelligent transfer vehicle lane-level route planning system within a container terminal, comprising: Input Acquisition Unit: Used for lane-level discretization modeling of the container terminal vehicle road network to obtain a directed mesh graph. It also obtains the task status information of each intelligent transfer vehicle at each time step, wherein each node in the directed mesh graph represents a passable mesh cell. For nodes, It is a directed edge; Occupancy request generation unit: used to generate requests based on the directed mesh graph at each time step. Based on task status information, a set of candidate actions is generated for each intelligent transfer vehicle, and based on directed edges... Map the candidate action set to the next time step occupancy request set; Conflict Detection and Priority Resolution Unit: Collects the next time step occupancy requests from all intelligent transfer vehicles, performs conflict detection and priority resolution, obtains the occupancy grant result for each intelligent transfer vehicle for the next time step, and executes according to the occupancy grant result for the next time step; Loop Execution Unit: Based on the execution of the occupancy grant result in the next time step, it returns to the occupancy request generation unit to advance and loop through the time steps until each intelligent transfer vehicle completes its task, thus obtaining the real-time evolving lane-level trajectory.

[0017] Compared with the prior art, the present invention has the following beneficial effects: (1) Based on the fine modeling of lane-level road resources, this invention generates a set of occupancy requests for the next time step for each intelligent transport vehicle at each time step. This set is used to perform real-time conflict detection and priority adjudication for multiple vehicles. During the parallel operation of multiple vehicles, the trajectory of the vehicles naturally evolves through a series of occupancy requests granted in real time, achieving complete synchronization between planning and execution. Furthermore, by submitting the set of occupancy requests for the next time step, all vehicles can accurately know the precise location and occupancy status of all vehicles at any given time. This can effectively achieve collaborative work under complex and dynamic operating conditions, improving real-time safety and operational efficiency.

[0018] (2) This invention utilizes a lane-level spatiotemporal occupancy modeling and representation method to discretize the container terminal vehicle road network into a directed grid graph composed of grid cells. The grid cells are used as the minimum passage resources for modeling, which not only enables precise resource management, but also allows for quick and accurate determination of whether there are overlapping grid cells in the future paths of any two vehicles. This helps to achieve the accuracy and efficiency of conflict detection. Furthermore, it makes path planning no longer a simple spatial curve search, but a process of generating and arbitrating "resource requests", thus ensuring that the executed path is conflict-free and meets capacity constraints. A current occupancy set is defined for each intelligent transfer vehicle to accurately represent the lane-level spatial range it occupies at each time step. In particular, the geometric envelope of vehicle turning and other maneuvering actions is considered, allowing a single vehicle to legally occupy multiple adjacent grid cells in a single time step, thus achieving a high-fidelity characterization of the micro-conflict risk in complex areas such as intersections.

[0019] (3) This invention proposes a conflict detection and resolution method based on mutual exclusion of occupancy and task-oriented arbitration. This method is based on the current occupancy set and performs conflict detection by comparing the "occupancy request" sets of different vehicles at the same time step. This improves the ability to identify and avoid micro-interaction conflicts such as occupancy at the same point, opposing conflicts and cross conflicts, and reduces the risk of conflict and deadlock. A closed-loop control mechanism of "occupancy request → grant → update" is adopted to implement strict mutual exclusion occupancy management for contested resources to ensure physical feasibility. Furthermore, a task-oriented right-of-way arbitration mechanism is introduced. Based on multi-dimensional vehicle status information such as task urgency, load status, task type and historical operation information, it dynamically calculates or decides the passage priority of each vehicle in contested resources, and intelligently resolves and schedules conflicts according to the priority, so as to realize task value-driven collaboration, reduce delays and improve the ability to complete tasks on time.

[0020] (4) This invention uses the generated feasible reference path as prior knowledge, embeds it into the vehicle state, and participates in subsequent online decision-making. This allows the vehicle to explore a limited range around the path, avoiding blind searching in a vast state space, significantly reducing computational burden and decision delay. At the same time, it ensures that the vehicle's online decision-making is still moving towards the goal, avoiding getting lost or trapped in a loop due to local optima. When encountering sudden obstacles or competing for resources with other vehicles, the vehicle can temporarily deviate from the reference path. After the conflict is resolved, the prior knowledge of the system state (position) and the reference path can be used together as input to quickly generate a new, feasible sequence of subsequent actions, achieving smooth recovery.

[0021] (5) This invention also uses the generated feasible reference path as a key input for calculating the dynamic priority scalar, which can prioritize vehicles that are closer to the target (with shorter remaining reference path length) or have more urgent tasks. This can accelerate the passage of these vehicles through bottleneck areas, reduce their queuing time before competing for resources, and thus reduce the average passage time of all vehicles. Moreover, the dynamic priority can actively guide the formation of "traffic waves," allowing key vehicles to pass first, thereby clearing nodes and preventing systemic congestion and deadlocks. In addition, the priority is no longer fixed. A vehicle may have a lower priority when it is far from the target, but as it gets closer to the target, its priority will be dynamically increased, thereby obtaining "giving way" compensation. This ensures both overall efficiency and individual fairness, and avoids any vehicle being delayed indefinitely.

[0022] (6) This invention is a real-time path planning process that can dynamically update vehicle occupancy requests and global scheduling schemes at each time step based on the latest vehicle status, task updates and changes in road network occupancy, thereby realizing continuous online evolution and replanning of paths, and thus ensuring the real-time performance, safety and overall operating efficiency of the system in dynamic and high-density operating environments.

[0023] (7) By combining road network structure coding, heuristic feasible priors and learning optimization, the efficiency of online solution and the robustness of strategy are improved, and the system’s ability to adapt to dynamic disturbances and task changes is enhanced. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a schematic diagram of the intelligent transport vehicle's turning movement and grid cell occupation according to the present invention; Figure 3 This is a schematic diagram of the conflict types of the intelligent transfer vehicle of the present invention, wherein (a) represents cross conflict, (b) represents opposite conflict, and (c) represents same-point occupation conflict; Figure 4 This is a heat map of the trajectory of the intelligent transfer vehicles of the present invention when the total number is 20; Figure 5 This is a schematic diagram showing the comparison results of the four types of methods of the present invention in terms of key performance. Detailed Implementation

[0025] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0026] This embodiment provides a task-oriented method for lane-level route planning of intelligent transfer vehicles in container terminals. This method achieves real-time route planning through a closed-loop process: "lane-level spatiotemporal occupancy representation—multi-vehicle decision modeling—strategy generation of occupancy requests—mutual exclusion feasibility verification and task-oriented arbitration—occupancy update and trajectory output—training and online update." Through this process, intelligent transfer vehicles (hereinafter referred to as vehicles) can form executable lane-level operation plans in dense, time-varying traffic environments, reducing conflicts and task delays, and meeting the needs of online real-time decision-making. This method is implemented based on a corresponding system, which includes an input acquisition unit for obtaining directed grids and task status information of each intelligent transfer vehicle at each time step; an occupancy request generation unit for generating a set of occupancy requests for the next time step for each intelligent transfer vehicle at each time step; a conflict detection and priority adjudication unit for conflict detection and priority adjudication; and a loop execution unit that advances system operation by time steps. Specifically, combining the functions of each unit of the system, such as... Figure 1 As shown, the method includes the following steps: Step 1: Establish a directed grid diagram of the wharf road network The system operates in time steps, and the task status information of each intelligent transfer vehicle is input (including task start point, current destination, remaining task time, task type, load status, current occupied set, etc.).

[0027] To achieve lane-level fine-grained control, the actual roads and lanes at the dock are first discretized and modeled, abstracting them into a directed mesh graph composed of regular mesh cells. ,in, Each node represents a walkable grid cell. Directed edges. Feasible movement relationships between adjacent grids are defined strictly based on the physical connections of lanes, traffic direction (one-way and two-way), and restricted areas. Here, "lane-level" control refers to explicitly encoding the lane topology, lane-changing restrictions, and intersection rules at the grid level, thereby supporting the generation of executable lane sequence plans. The grid size can be selected based on a trade-off between computational power and planning accuracy requirements; preferably, the grid size is discretized according to the vehicle's travel distance and lane width within a time step. This implementation uses a 4-meter grid. The grid is 4 meters, where the vertical scale corresponds to the typical travel distance of a vehicle within one time step, and the horizontal scale corresponds to the lane width or its discretized approximation. The system operates in discrete time steps. Advancement, optimal time step selection It takes 1 second.

[0028] The input acquisition unit obtains the directed mesh graph constructed above. The task status information for each intelligent transfer vehicle at each time step includes at least the task start point. Current destination Remaining time for the task Task Type Load condition Current occupied set To accurately describe the space occupied by the vehicles, a current occupancy set is defined for each intelligent transfer vehicle at each time step. This current occupancy set represents the vehicle's position at time step [time value missing]. The range of grid cells covered subscript Indicates vehicle The number of grid cells occupied at any given moment is determined by the vehicle length. The distance traveled by the vehicle within a time step Vehicle width and lane width Confirmed, satisfied .

[0029] Step 2: Within each time step, based on the directed mesh graph Based on task status information, a set of candidate actions is generated for each intelligent transfer vehicle, and based on directed edges... Map the set of candidate actions to the set of occupancy requests for the next time step.

[0030] This step utilizes the occupancy request generation unit to first generate a set of candidate actions for each intelligent transfer vehicle, and then uses directed edges... Map this to the set of requests to be occupied in the next time step. Specifically: The vehicle generates a set of candidate actions based on its current state (current destination, task attributes, currently occupied set, etc.), including forward, backward, left turn, right turn, and waiting. A single vehicle's actions can be represented as... ,in, Indicates the movement of the vehicle. Indicates vehicle At this moment of decision-making The selected current destination. Vehicle movement must simultaneously satisfy both road network connectivity constraints and vehicle kinematic constraints. Road network connectivity constraints limit vehicle movement to permitted directions within the directed grid; vehicle kinematic constraints include at least the following: the vehicle moves between adjacent reachable grids according to its current orientation and selected action.

[0031] To accurately characterize the geometric envelope of vehicles during maneuvers such as turning, especially the micro-conflict risks in areas like intersections, the system pre-calculates and stores a "motion envelope template." The template is designed for different actions such as going straight, turning left, and turning right, as well as discretized vehicle orientations. Figure 2 This demonstrates the vehicle's turning motion and the occupancy of grid cells. The template, indexed by {action type, orientation, step number}, stores the offset set of the grid cells covered by the vehicle contour relative to a reference point in each execution step. Preferably, the vehicle's geometric center is used as the reference point; in other embodiments, the rear axle center can also be used. During online decision-making, the corresponding template is queried based on the vehicle's current action, orientation, and number of executed steps. The precise occupancy set for the next time step can be quickly obtained through translation. For actions requiring multiple steps (such as turns lasting several steps), the system generates only the occupancy request for the next step in each time step, completing the entire maneuver through continuous steps. Therefore, based on the above action and occupancy modeling, each candidate action can be mapped to the occupancy request set for the next time step. The specific mapping process is as follows: First, based on the set of directed edges And the selected action, calculate the position state of the vehicle reference point at the next time step. : , in For vehicles at any time Location and orientation The action type, vehicle orientation, and preset turning envelope template are used to determine the action. For actions that are prohibited from reversing, in restricted areas, or that are inconsistent with the current orientation, the system will directly determine them as non-actions and remove them from the candidate actions.

[0032] Then, query the motion envelope template corresponding to the vehicle reference point and the action type, vehicle orientation, and execution step to obtain the offset set relative to the reference point. .

[0033] Finally, the offset set is shifted to the position corresponding to the next time step to obtain the vehicle's occupancy request set for the next time step. The system extracts only the template offset set corresponding to the current time step at each time step to generate the occupancy request set for the next time step, and completes the entire maneuver process through rolling updates across consecutive time steps. Furthermore, for linear maneuvers, This corresponds to the current rectangular set occupied by the vehicle after translation; for actions requiring multiple steps, such as left turns and right turns... Corresponding to the pre-stored multi-step turning envelope template (as attached) Figure 2 (As shown).

[0034] Step 3: Gather the next time step occupancy requests from all smart transfer vehicles, perform conflict detection and priority determination, and return to Step 2 for cyclical updates.

[0035] This step utilizes a conflict detection and priority adjudication unit to detect conflicts between occupancy requests from different vehicles at the same time step, and to adjudicate the priority of a contested resource. Specifically: After generating the set of occupancy requests for each vehicle in the next time step, the central controller in the system performs conflict detection on the occupancy requests of different vehicles in the same time step. When the occupancy requests of multiple vehicles overlap on the same grid cell, the system identifies it as a "resource contention" scenario and determines that a "conflict has occurred." (See attached...) Figure 3 As shown, this invention mainly identifies three types of spatiotemporal interaction conflicts: the first is co-occupancy conflict, where different vehicles request to occupy the same grid cell; the second is opposing conflict, where two vehicles request to enter adjacent grid areas in front of each other in opposite directions, and their occupancy request sets overlap within the same time step; and the third is cross conflict, where the occupancy request sets of different intelligent transport vehicles overlap in the cross area in the next time step, or they request to enter the same cross area from different directions. Since this embodiment uses the grid occupancy set of a vehicle within a single time step as the basis for conflict determination, and vehicles advance between grids in discrete time steps, opposing conflict, cross conflict, and co-occupancy conflict can all be uniformly characterized as overlapping occupancy request sets.

[0036] Preferably, this embodiment employs a mutual exclusion constraint for priority conflict filtering to avoid unexecutable path planning results and ensure that the occupancy request meets physical feasibility requirements. This mutual exclusion constraint can be expressed as: for any two different vehicles... and any time step The set of requests for occupancy in the next time step should satisfy the following conditions: If the constraint is not met, it indicates that grid cell overlap will occur, and the grid cell will be marked as a "contested resource". If the constraint is met, it indicates that grid cell overlap will not occur, and the right to occupy will be directly granted to the next time step occupancy request set of the intelligent transport vehicle that will not overlap.

[0037] The system also maintains a global data structure called the "global occupancy table," which records the pre-defined occupancy status of each grid cell and subsequently defined intersection area resources in the next time step. At each time step, each vehicle submits its next occupancy request. After the central controller completes coordination, it generates an "grant table" and broadcasts it to all vehicles. After the granted vehicle executes the action, the system releases its occupancy record from the previous time step and writes the newly granted occupancy into this table.

[0038] This step also employs a semaphore-based occupancy control mechanism to schedule contentious resources, forming a closed-loop control cycle of "occupancy request → grant → occupancy update." This semaphore-based occupancy control mechanism can serve as an implementation of a global occupancy table and grant table, used to characterize whether each grid cell or region resource still has grantable capacity at the current time step. Specifically, the system maintains a semaphore state for each grid cell or region resource. Its value range is , Represents grid cells The capacity limit, for any grid cell and time step Capacity constraints must be met. .

[0039] At the start of each global time step, each vehicle generates a set of occupancy requests for the next time step based on its current state and submits it to the central controller; the controller collects relevant requests for each contested resource and makes decisions based on the overall occupancy requests of the vehicles. Based on a certain contested resource at the current moment semaphore states Combined with the next time step Based on the reserved occupancy status, determine the status of each intelligent transfer vehicle in the competing vehicle set. The next time step's occupancy request must satisfy the capacity constraint of the corresponding contested resource. If the intelligent transfer vehicle... The set of occupancy requests in the next time step satisfies the capacity constraint of the corresponding contested resource. This indicates that the intelligent transfer vehicle The vehicle is qualified to grant occupancy. For all intelligent transfer vehicles qualified to grant occupancy, the controller outputs the dynamic priority scalar value according to the task-oriented dynamic priority module. The system prioritizes vehicle requests and allocates passage rights to contended resources in a task-oriented manner to reduce unnecessary waiting and delays and improve overall coordination efficiency. This involves prioritizing vehicle requests and granting rights according to priority. After each grant, the predefined semaphore state of the corresponding contended resource is updated. Preferably, the capacity of the ordinary lane grid unit is taken as... In this implementation, only the vehicle with the highest priority is granted a resource. A contentious resource satisfies... If the resource has no remaining capacity in the next time step, no new occupancy requests will be granted.

[0040] Vehicles that are not granted an occupancy permit remain in their current position and wait, or trigger a reselection mechanism: after removing the action that caused the conflict from the candidate action set, a new occupancy request is generated, and the conflict detection and granting process within the same time step is entered again, forming a loop. This reselection process can continue until the vehicle is granted an occupancy permit or the preset reselection limit is reached; if there is still no action to take, then the vehicle waits.

[0041] Once all contested resources have been granted, the controller generates the grant result and broadcasts it to all vehicles. The vehicles that have been granted resources execute the corresponding action for the next time step and update the occupied set of that intelligent transport vehicle for the next time step. In occupying the set After the update, release the smart transfer vehicle at the current moment. The original resources are occupied, and the global occupancy table is updated synchronously. The updated position sequence of all vehicles at the current time step is output, thus forming a real-time evolving lane-level trajectory to complete the occupancy state switch from the current time step to the next time step.

[0042] The above process completes a closed-loop control process of "submitting occupancy request - mutual exclusion verification - granting right of way - executing action - updating occupancy table - entering the next time step and requesting again".

[0043] The aforementioned dynamic priority module can be inferred by a lightweight neural network model (e.g., a fully connected network with two hidden layers). Its input includes at least information such as task urgency, task type, load status, starting point and current destination, current occupied set, and historical operation information (e.g., execution status, completion time, and historical trajectory data of recent tasks). The output is a dynamic priority scalar of the vehicle.

[0044] Preferably, the input to the aforementioned dynamic priority module also incorporates the road network structure-aware coding result as an auxiliary context input. Specifically, the terminal vehicle road network is first represented as a directed mesh graph. , where nodes Indicates a passable mesh cell, edge This represents the adjacent feasible motion relationships that satisfy the lane direction constraints. Then, feature vectors are constructed for each node. The feature vector includes at least occupancy status, obstacle identification, and area type information. The area type information distinguishes between loading / unloading areas, storage yards, quay crane areas, approach bridge areas, regular lane areas, or intersection areas. Subsequently, the directed mesh graph is input into a graph encoding model, preferably a graph attention network, to perform neighborhood-based weighted aggregation on the node features, resulting in updated node embeddings: ,in For learnable weight matrix, For activation function, For nodes The set of adjacent nodes, These are the attention weights. Further, aggregation processing is performed on all node embeddings to form a global feature vector representing the current road network structure. This is then combined with the vehicle's local state to form an extended state input: ,in This represents a feasible reference path generated by heuristic search, which is used as prior knowledge. A dynamic priority scalar is obtained by processing the extended state in a lightweight neural network model. .

[0045] The above feasible reference paths At the start of the task, a heuristic search algorithm is used to generate a feasible reference path under the current static environmental constraints. Preferably, A is used. * Search algorithms in directed grid graphs Using the task origin and candidate destination as endpoints, and with obstacle nodes, forbidden grids, unreachable directional edges, and static capacity limits as constraints, the shortest feasible path from the current origin to the target location is searched. The feasible reference path is not directly used as the final execution trajectory for mandatory tracking, but rather embedded as prior knowledge in the vehicle state during "warm start" and participates in subsequent online decision-making. Specifically, the reference path can be represented as a sequence of grid cells, a relative heading sequence, or a sequence of recommended lanes for several future steps, and appended to the vehicle state input. During online decision-making, after the vehicle generates candidate actions at each time step, the controller or policy network aligns and compares the current action with the reference path: if a candidate action keeps the vehicle on the reference path or moves towards the next key node on the reference path, the sampling probability of that action is increased; if an action significantly deviates from the reference path and does not bring conflict avoidance benefits, its selection tendency is reduced. In the event of contention, local congestion, or the road segment ahead being occupied, the system allows the vehicle to temporarily deviate from the feasible reference path, choosing to wait, detour, or yield through an arbitration mechanism. After the conflict is resolved, the heuristic algorithm is invoked again based on the vehicle's current position and current destination to generate an updated feasible reference path, or the vehicle can return to a subsequent sub-path of the original reference path to continue moving forward. Based on the aforementioned extended state, when performing priority reasoning, the dynamic priority module not only considers the urgency of the vehicle's own task and its immediate location, but also combines the vehicle's structural position within the entire port road network, the availability of alternative routes, and potential congestion risks to output a priority scalar that better aligns with the overall traffic organization objectives. When running online, the controller can directly rely on priority scalars. The granting order can be determined from largest to smallest; alternatively, the priority result can be used as a bias score item corresponding to the "enter contention grid action" to provide contextual guidance for the choice of actions such as yielding, waiting, detouring, or going straight, thereby achieving more intelligent, task-aware, and structure-sensitive right-of-way coordination.

[0046] The system continuously monitors its operational status. When a new task arrives, a vehicle's resource contention waiting time exceeds a preset threshold (e.g., 20 seconds), or a deadlock risk is detected through the construction and analysis of the dependency graph, a replanning process is immediately triggered. The dependency graph uses vehicles as nodes. If intelligent transport vehicle A is waiting for intelligent transport vehicle B to release its requested resources (such as the target grid), a directed edge A→B is established. The presence of a directed cycle in the graph indicates a deadlock risk. This forms a closed-loop online planning process capable of rapidly responding to dynamic environmental changes.

[0047] For systems containing the aforementioned learnable modules (such as dynamic priority modules and graph coding models), a "centralized training, distributed execution" framework can be used for offline optimization and online deployment. During the training phase, interactive data is generated using a dock simulation environment to optimize model parameters with the goal of minimizing total system delay. During the deployment phase, each vehicle independently generates action proposals based on its locally perceived state and pre-trained fixed policies. The central controller centrally executes conflict detection, arbitration, and granting, and broadcasts the results, eliminating the need for direct communication between vehicles. This framework is a preferred implementation and does not affect the independent operation and feasibility of the aforementioned rule-based implementation method.

[0048] This embodiment is based on a real road network layout of a container terminal. The road network includes 8 sea-side operation lanes, 2 approach bridges, and 6 yard areas; the quay crane area contains 3 quay cranes, each with 6 P / D points; each yard area provides 15 P / D points along the operation lanes. The test uses different numbers of intelligent transfer vehicles, ranging from 20 to 100, to evaluate the online performance of this method under different traffic densities. The relevant results are available in […]. Figure 4 and Figure 5 The method is described in the document. Three baselines were used for comparison: Baseline A (heuristic path + fixed first-come, first-served rule), Baseline B (learning-based scheduling but conflict handling still uses fixed rules), and Baseline C (value decomposition-based learning method but conflict handling still uses fixed rules). Comparison metrics included task service quality (total latency), conflict control effectiveness (conflict rate), and real-time performance (average decision step latency). Experimental comparisons show that in high-density scenarios, because this invention integrates lane-level physical constraint modeling and task-oriented arbitration, while baselines B and C still use fixed rules for conflict handling, it performs better in terms of total latency and conflict rate, and the average decision step latency meets the timeliness requirements of online planning. The method in this embodiment can perform real-time planning of vehicle driving lanes based on road network occupancy status and demonstrates advantages in core metrics such as conflict rate, total delay time, and average decision step latency. The relevant results are as follows: Figure 4 (The darker the color, the more smart transfer vehicles have chosen that lane) and Figure 5 As shown.

[0049] In summary, the task-oriented intelligent transfer vehicle lane-level real-time route planning method proposed in this embodiment can effectively realize real-time route planning with controlled conflicts among multiple intelligent transfer vehicles, thereby improving system operation safety, real-time performance, and overall collaborative efficiency.

[0050] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0051] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention can be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0052] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0053] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0054] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0055] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0056] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A task-oriented method for lane-level route planning of intelligent transfer vehicles in a container terminal, characterized in that, Includes the following steps: S1. Obtain the directed mesh graph obtained by discretizing the lane-level model of the container terminal vehicle road network. It also obtains the task status information of each intelligent transfer vehicle at each time step, wherein each node in the directed mesh graph represents a passable mesh cell. For nodes, It is a directed edge; S2. At each time step, based on the directed mesh graph Based on task status information, a set of candidate actions is generated for each intelligent transfer vehicle, and based on directed edges... Map the candidate action set to the next time step occupancy request set; S3. Collect the next time step occupancy request set of all intelligent transfer vehicles, perform conflict detection and priority adjudication, obtain the occupancy grant result of each intelligent transfer vehicle for the next time step, and execute according to the occupancy grant result of the next time step; S4. Based on the execution of the occupancy grant result in the next time step, return to step S2 to advance and loop through the time steps until each intelligent transfer vehicle completes its task and obtains the real-time evolving lane-level trajectory.

2. The task-oriented intelligent transfer vehicle lane-level route planning method in a container terminal according to claim 1, characterized in that, The task status information includes the task start point. Current destination Remaining time for the task Task Type Load condition Current occupied set subscript Indicates the first Intelligent transfer vehicle, underlined Indicates intelligent transfer vehicle The number of grid cells occupied at each time step is expressed as: ， In the formula, For intelligent transfer vehicles The length of the car body, For driving distance, For vehicle width, This refers to the lane width.

3. The task-oriented intelligent transfer vehicle lane-level route planning method in a container terminal according to claim 1, characterized in that, The steps for generating the candidate action set include: Based on the task status information, as well as road network connectivity constraints and vehicle kinematic constraints, each intelligent transfer vehicle... Generate a set of candidate actions , Indicates intelligent transfer vehicle The driving action, Indicates intelligent transfer vehicle At this moment of decision-making The selected current destination, where the road network connectivity constraint represents the intelligent transfer vehicle. The vehicle can only move along the directions permitted in the directed mesh graph, and the vehicle kinematic constraints include at least the intelligent transport vehicle. Move between adjacent reachable grid cells according to the current orientation and the selected candidate action.

4. The task-oriented intelligent transfer vehicle lane-level route planning method in a container terminal according to claim 1, characterized in that, The mapping steps for the next time step occupancy request set include: Obtain a motion envelope template, which is indexed by {action type, orientation, step number}; According to the directed edge And the candidate actions selected from the candidate action set, to calculate the intelligent transfer vehicle The position and orientation of the reference point at the next time step are calculated using the following expression: ， In the formula, For intelligent transfer vehicles At any moment Location and orientation It is determined by the action type, vehicle orientation, and a preset turning envelope template. For intelligent transfer vehicles At any moment Location and orientation For intelligent transfer vehicles The driving action; Query the intelligent transfer vehicle The motion envelope template corresponding to the reference point, the currently selected candidate action, orientation, and number of steps executed is queried to obtain the offset set relative to the reference point. ,in, This refers to the stagger sequence number of the current candidate action during continuous execution; The offset set By translating the position and orientation to the next time step, we obtain the intelligent transfer vehicle. The next time step occupies the request set , For intelligent transfer vehicles The next driving action in time step offset set The single offset relative to the reference point in the middle.

5. The task-oriented intelligent transfer vehicle lane-level route planning method in a container terminal according to claim 1, characterized in that, The types of conflicts include: Conflict of Occupation at the Same Point: There are requests to occupy the same grid cell in the next time step occupancy request sets of different intelligent transfer vehicles; Opposing conflict: There are adjacent grid cell regions in front of each other that are requesting to enter in opposite directions in the next time step occupancy request set of two intelligent transfer vehicles. Cross-over conflict: The next time step occupancy request sets of different intelligent transfer vehicles overlap in the cross-over area, or they request to enter the same cross-over area from different directions.

6. The task-oriented intelligent transfer vehicle lane-level route planning method in a container terminal according to claim 1, characterized in that, The steps for conflict detection and priority determination include: Initialize the semaphore state of each grid cell , Indicates the current time , For grid cells The maximum capacity per grid cell Satisfy capacity constraints; Define a global occupancy table, which is used to record the current time step. The occupancy status of each grid cell and the planned occupancy status for the next time step; Determine whether the next time step occupancy request sets of all intelligent transfer vehicles will overlap on any grid cell. If so, mark the overlapping grid cells as contentious resources and form a contentious vehicle set. If not, directly grant occupancy rights to the next time step occupancy request sets of intelligent transfer vehicles that will not overlap, wherein occupancy mutual exclusion constraints are used for the determination. Based on the remaining time of the task Task Type Load condition Task Start Point Current destination Current occupied set In addition to historical operational information, the system calculates the number of intelligent transfer vehicles in the disputed vehicle set. Dynamic priority scalar ; Based on the current state of each competing resource semaphore states Combined with the next time step The reserved occupancy status, when the intelligent transfer vehicle in the set of competing vehicles... The next time step's occupancy request set satisfies the capacity constraint of the corresponding contested resource. This indicates that the intelligent transfer vehicle It has the qualification to grant occupancy, and it has authority over all intelligent transfer vehicles that have the qualification to grant occupancy. According to the dynamic priority scalar Granting a contested resource, and updating the semaphore state of the corresponding resource after each grant. ;when When the time step is reached, it indicates that the resource being contested has no remaining capacity in the next time step, and new requests for occupancy will no longer be granted. For intelligent transfer vehicles that have not been granted occupancy authorization The system either remains in its current position and waits, or removes conflicting candidate actions from the candidate action set and regenerates a new set of occupancy requests for the next time step, then re-enters the conflict detection and priority adjudication process within the same time step, continuing this process until the intelligent transfer vehicle... Obtain the right to occupy, or reach the preset limit for the number of reselections; For intelligent transfer vehicles that have been granted occupancy rights The intelligent transfer vehicle will then execute the corresponding action for the next time step. The occupancy set is updated in the next time step. Then release the intelligent transfer vehicle. The system handles contention for resources at the current time step and updates the global resource allocation table synchronously to complete the switching of resource allocation status from the current time step to the next time step.

7. The task-oriented intelligent transfer vehicle lane-level route planning method in a container terminal according to claim 6, characterized in that, The capacity constraint is: ， In the formula, A collection of intelligent transfer vehicles. This is an indicator function; it takes the value 1 when the condition inside the parentheses is true, and 0 otherwise. The occupancy mutual exclusion constraint is as follows: ， In the formula, For intelligent transfer vehicles The next time step occupies the request set. For intelligent transfer vehicles The next time step occupies the request set. It is an empty set.

8. The task-oriented intelligent transfer vehicle lane-level route planning method in a container terminal according to claim 6, characterized in that, In calculating the dynamic priority scalar At the same time, the road network structure perception coding results are also introduced, and the road network structure perception coding results are compared with those of the intelligent transfer vehicle. The local states are concatenated to form the extended state input. To utilize the extended state input Calculate the dynamic priority scalar ,in, The steps for obtaining the road network structure-aware coding results, which are feasible reference paths generated by a heuristic search algorithm, include: The directed mesh graph Each node in the construct features vector , where subscript For the first The feature vector of the node At a minimum, it should include information on occupancy status, obstacle identification, and area type; The directed mesh graph In the input graph coding model, the feature vectors of each node are used as the basis for input. Perform neighborhood-based weighted aggregation to obtain the updated node embeddings. , is represented as: ， In the formula, For activation function, For attention weights, For learnable weight matrix, For nodes The set of adjacent nodes, For the first Feature vectors of each node; Embed the updated nodes corresponding to all nodes Perform aggregation processing to form a global feature vector representing the current road network structure. This serves as the result of the road network structure perception coding.

9. A task-oriented intelligent transfer vehicle lane-level route planning method in a container terminal according to claim 8, characterized in that, The steps for obtaining the feasible reference path include: In the directed mesh graph Using the task start point and candidate destination as endpoints, and with obstacle nodes, prohibited grid cells, unreachable direction edges, and static capacity limits as constraints, a heuristic search algorithm is used to find the shortest feasible path from the current start point to the target location, which serves as the feasible reference path. ; The feasible reference path mentioned above As prior knowledge, it is embedded into the task state information so that after generating a set of candidate actions at each time step, the current candidate action is compared with the feasible reference path. Perform alignment comparison: If a candidate action causes the intelligent transfer vehicle to... If the candidate action is kept on the feasible reference path or moves toward the next key node on the feasible reference path, the sampling probability of the candidate action is increased. If a candidate action deviates from the feasible reference path by more than a set threshold and does not bring conflict avoidance benefits, the selection tendency of the candidate action is reduced. When a conflict is detected, the intelligent transfer vehicle is allowed. The path may temporarily deviate from the feasible reference path, and once the conflict is resolved, it may revert to the feasible reference path to continue moving forward, or it may be based on an intelligent transfer vehicle. The heuristic search algorithm is re-invoked to generate an updated feasible reference path based on the current starting point and the current target position.

10. A task-oriented intelligent transfer vehicle lane-level route planning system in a container terminal, characterized in that, include: Input Acquisition Unit: Used to acquire the directed mesh graph obtained by lane-level discretization modeling of the container terminal vehicle road network. It also obtains the task status information of each intelligent transfer vehicle at each time step, wherein each node in the directed mesh graph represents a passable mesh cell. For nodes, It is a directed edge; Occupancy request generation unit: used to generate requests based on the directed mesh graph at each time step. Based on task status information, a set of candidate actions is generated for each intelligent transfer vehicle, and based on directed edges... Map the candidate action set to the next time step occupancy request set; Conflict Detection and Priority Resolution Unit: Collects the next time step occupancy requests from all intelligent transfer vehicles, performs conflict detection and priority resolution, obtains the occupancy grant result for each intelligent transfer vehicle for the next time step, and executes according to the occupancy grant result for the next time step; Loop Execution Unit: Based on the execution of the occupancy grant result in the next time step, it returns to the occupancy request generation unit to advance and loop through the time steps until each intelligent transfer vehicle completes its task, thus obtaining the real-time evolving lane-level trajectory.

Images