A warehouse AGV conflict resolution system combining scenario coupling TOPSIS and shadow verification

Abstract

The application belongs to the technical field of warehouse AGV, and particularly relates to a warehouse AGV conflict resolution system based on scene coupling TOPSIS and shadow verification. The application generates an intervention request when the conflict exists; a local state snapshot and a conflict feature key are constructed in response to the request; consistency verification is performed on a reserved time window and a kinematic predicted occupancy time window; opposite edge conflicts, pursuit conflicts, point conflicts and cyclic waiting deadlocks are identified; resolution actions are constructed according to the conflict types; time cost and energy consumption cost are multi-target sorted and soft coupled with scene preferences; the resolution actions are simulated and verified in the local snapshot copy; the resolution actions that pass the verification are sent to the robot control system for execution; thereby, the calculation cost is reduced, the diagnosis accuracy and conflict resolution efficiency are improved, and the safety and optimality of the resolution actions are ensured through TOPSIS sorting and shadow verification.

Description

Technical Field

[0001] This invention belongs to the field of warehouse AGV technology, specifically relating to a warehouse AGV conflict resolution system that couples TOPSIS and shadow verification in a scenario. Background Technology

[0002] In warehouse automation systems, warehouse AGVs (Automated Guided Vehicles) automatically travel or tow cargo trolleys to designated locations along pre-set routes. The warehouse control system is responsible for task allocation and work scheduling, while the robot control system is responsible for issuing vehicle paths, traffic control, and executing movement commands.

[0003] In this system, the warehouse map is represented as a topological graph consisting of ground code nodes and directed edges. Vehicles scan the ground codes to report their discrete positions to the RCS and occupy nodes and edges sequentially along a predetermined path. To avoid vehicles using the same traffic resources simultaneously, existing warehouse AGV conflict resolution systems typically use global path planning, spatiotemporal A*, node or edge reservation, fixed priority, first-come-first-served, and local waiting methods to determine the passage order. However, the on-site operation status is affected by factors such as work stoppages, communication delays, differences in vehicle acceleration and deceleration, temporary congestion, positioning refresh, and lag in reservation table updates. The original planned path and reservation time window may deviate from the actual ground code position and movement status of the vehicle, resulting in local conflicts that were not eliminated in the planning stage or were re-formed in the execution stage. If only global planning is re-executed, not only will the computation and communication overhead be high, but unnecessary path disturbances may also occur to vehicles that did not participate in the conflict.

[0004] The current common conflict resolution method is to detect whether the next few nodes of a vehicle are occupied and make low-priority vehicles wait. The judgment distance is usually only a few yard intervals. For vehicles entering the same narrow alleyway from the opposite direction, there may not be enough room to back up safely by the time of detection. For vehicles chasing each other in the same direction, there may not be enough time to restore a safe distance by smooth deceleration by the time of detection, and they can only stop and wait. For work vehicles located on multiple subsequent path nodes, it is also difficult to complete the detour before the vehicle enters the narrow area. Therefore, detecting only the next node cannot simultaneously meet the requirements of conflict identification lead time and traffic processing efficiency.

[0005] Another common conflict resolution method is to directly determine whether a node or edge is in conflict based on the reservation time window. The reservation table can express the spatiotemporal occupancy relationship during planning, but its effectiveness depends on the vehicles still running according to the original plan. When a vehicle deviates from the reservation time due to loading and unloading operations, pauses, restarts, or other conflict resolution, continuing to use the original reservation table unconditionally will result in misjudgment. Completely discarding the reservation table and re-estimating all arrival times will result in the loss of the passage order that has been determined by the RCS.

[0006] Furthermore, existing conflict resolution methods often bind specific conflicts to fixed actions in their local traffic rules. While fixed rules are simple to implement, they are difficult to adapt to changes in buffer node availability, bypass length, vehicle speed, task priority, and reservation status. Moreover, if all actions are directly sorted using a general multi-attribute ranking without distinguishing conflict scenarios, actions with low numerical costs but not applicable to the current physical scenario may achieve higher rankings. If candidate actions are issued directly without checking node capacity, continuous edge occupancy, opposite physical edge occupancy, reservation conflicts, and waiting relationship recurrence, secondary conflicts are likely to occur.

[0007] In summary, this application provides a conflict resolution system for warehouse AGVs that couples TOPSIS and shadow verification in a scenario. Summary of the Invention

[0008] A scenario-coupled TOPSIS and shadow verification conflict resolution system for warehouse AGVs includes:

[0009] The condition triggering module is used to generate a conflict intervention request when the abnormal operation continues for a preset time threshold.

[0010] The snapshot parsing module is used to respond to conflict intervention requests, obtain the local topology of the warehouse, vehicle location, vehicle remaining path, vehicle movement status, appointment time window and trigger evidence, and construct a local state snapshot;

[0011] The time window verification module is used to verify the consistency between the scheduled time window and the kinematically estimated time window, and to select the estimated time window when the scheduled time window is not applicable.

[0012] The conflict diagnosis module is used to determine the conflict type and conflict node or conflict edge based on the local state snapshot and the time window selected after verification.

[0013] The candidate action generation module is used to construct a set of candidate resolution actions based on the conflict type.

[0014] The TOPSIS ranking module is used to calculate the basic proximity based on time cost and energy consumption estimation cost, and obtain a comprehensive ranking by weighting the scenario preference scores corresponding to the conflict type.

[0015] The shadow verification module is used to advance the vehicle state based on candidate actions and verify traffic resource occupancy constraints in a local state snapshot copy according to the comprehensive sorting.

[0016] The output and control handover module is used to send the verified candidate actions and execution constraints to the robot control system.

[0017] Preferably, the operational anomalies in the condition triggering module include: maintaining trigger evidence by the main path planner or motion execution controller for the same failed planning, reservation, or movement authorization object; and generating a conflict intervention request when subsequent executable paths continue to fail to be generated or submitted, resource reservations continue to fail, movement authorizations continue to fail, safe parking occurs, or no effective physical progress reaches the time threshold.

[0018] The snapshot parsing module is also used to obtain the vehicle location code and trigger evidence in response to conflict intervention requests, and to construct conflict feature keys with the identifiers of the requesting vehicle, the potentially blocked vehicle and the potential conflict resource, for association with diagnostic results, candidate actions and execution feedback.

[0019] Preferably, when the reservation end time plus allowable deviation is still earlier than the earliest kinematically predicted arrival time, or when the reservation start time is later than the kinematically predicted completion time plus allowable deviation and the vehicle is not in a reservation waiting state, the corresponding reservation time window is marked as inapplicable, and the predicted occupancy time window obtained based on the ground code location and kinematic parameters is used.

[0020] Preferably, two directed edges with opposite directions are mapped to the same physical edge, and the physical edge is identified as a conflict edge when two vehicles have already occupied the physical edge in opposite directions, or when the effective occupation time windows of the two vehicles for the two directed edges overlap within the safety clearance margin.

[0021] Intersection spatiotemporal conflicts are determined by comparing the arrival-to-departure time windows of multiple vehicles at intersection nodes or conflict zone nodes, and intersection spatiotemporal conflicts are suppressed when the non-overlapping passage order has been determined in the effective reservation time window.

[0022] Search along the future path of the requesting vehicle to find nodes that are stably occupied by other vehicles in the predicted time domain and cannot be released before the requesting vehicle arrives, thus identifying future node conflicts.

[0023] Preferably, continuous common directed edges with the same direction are identified from the remaining paths of the preceding and following vehicles. One-dimensional topological corridor coordinates are constructed according to the side length of the continuous common directed edges. In the prediction time domain, the states of the preceding and following vehicles are advanced according to vehicle speed, acceleration, deceleration, vehicle length, edge speed limit and turning delay. The moment when the effective distance between vehicles is not greater than the dynamic target safety distance is determined as the pursuit risk time.

[0024] A local waiting graph is constructed with vehicles as vertices and the relationship between waiting vehicles and blocking vehicles as directed edges. When a directed cycle exists, the loop waiting deadlock is checked based on the conflict duration, appointment failure, and trigger evidence of no effective physical progress.

[0025] Preferably, for pursuit conflicts, based on the dynamic target safety distance and the current effective distance, the upper limit of the following vehicle speed is determined so that the effective distance between vehicles is not less than the dynamic target safety distance within the control time domain, and the upper limit of the following vehicle speed is used as the target speed for the candidate speed adjustment action; wherein, the dynamic target safety distance includes the basic static distance, the travel distance of the following vehicle within the control link response time, the positive difference between the braking distance of the following vehicle and the braking distance of the preceding vehicle, and the safety margin;

[0026] When all other candidate resolution actions are not feasible, a safety hold action is generated to keep the selected vehicle stationary or prevent it from entering the conflicting resource within a bounded hold time, and the master path planner is requested to release, postpone or rebuild future appointments that have not yet been executed; based on the event-driven execution feedback, the latest conflicting resource status and action resolution conditions, if the conflict is not resolved, a candidate resolution action is regenerated or a system-level scheduling intervention is requested.

[0027] Preferably, a leading vehicle is determined, and waiting candidate actions are generated for low-priority vehicles that do not block the remaining path of the leading vehicle. When a low-priority vehicle blocks the remaining path of the leading vehicle and there is an available buffer node, a candidate action is generated to make the low-priority vehicle retreat to the buffer node. The buffer node includes a predefined buffer node and a dynamic escape node determined based on the direct adjacent edge of the current position, the node occupancy status, the reservation status, and the vehicle's retreat capability.

[0028] Using conflict nodes or conflict edges as restricted resources, prioritize the reachable merging point ahead of the original remaining path as the local planning target. If no reachable merging point is found, select the local window exit as the target, and generate a replanning constraint instruction that includes at least the restricted node or restricted edge, merging point or local window exit, executing vehicle, executing location, initial constraint, and control lease term.

[0029] Preferably, the TOPSIS decision matrix is ​​constructed using time cost and energy consumption estimated cost. Vector normalization and weighted normalization are performed on time cost and energy consumption estimated cost. Positive ideal solution and negative ideal solution are determined using cost-type indicators. The Euclidean distance from each candidate resolution action to the positive ideal solution and negative ideal solution is calculated, and the basic proximity is calculated based on the Euclidean distance.

[0030] After obtaining the basic proximity score, the preference scores corresponding to the basic proximity score and the scene adaptation order are weighted and combined according to the scene coupling coefficient to obtain the comprehensive ranking of candidate resolution actions.

[0031] Preferably, the local influence domain is determined based on the conflicting vehicles, blocked vehicles, conflicting resources, and safety margin in the local state snapshot. Only the vehicle status, reservation, and route data that need to be modified by the candidate action within the local influence domain are deeply copied, and the reservations and traffic resources outside the influence domain are treated as read-only boundary conditions.

[0032] Record the occupied range of a vehicle from entering a directed edge to reaching the destination node and being able to release the directed edge, forming a continuous edge occupancy table;

[0033] In each simulation step, the node capacity, the capacity of the same-direction edge, the simultaneous occupancy of directed edges in opposite directions on the same physical edge, the reservation time window conflict, and the recurrence of waiting relationships are verified. When any hard constraint fails, the current candidate verification is stopped and the next candidate is verified.

[0034] Preferably, the candidate resolution actions verified by shadow simulation are sorted in a comprehensive order to form a candidate queue, and a conflict identifier, snapshot timing identifier, reservation version, execution location and planning control lease period are attached, and then sent to the robot control system.

[0035] The robot control system accepts candidate actions when conflicts still exist, snapshots and reservation versions are applicable, the vehicle has not crossed the execution position, and there are no other valid control leases within the current scope; if path or reservation changes are involved, the main path planner generates and atomically submits the path and reservation; waiting, speed adjustment, and safety holding actions are executed by the motion execution controller within the lease and action constraints.

[0036] When a candidate action is rejected, the generation path fails, or the atomic submission fails, the original path and the original reservation are maintained and the next candidate is tried; event-driven execution feedback is received and conflict resources and action resolution conditions are reviewed.

[0037] Beneficial effects

[0038] This invention generates a conflict intervention request and triggers a snapshot parsing module to obtain a local state snapshot when route planning fails, reservations continue to fail, or the vehicle stops safely or fails to make effective physical progress to a preset intervention time threshold. The snapshot parsing module only constructs a local state snapshot at the current moment after receiving the conflict intervention request. The complete conflict diagnosis and resolution process is only initiated when a conflict actually occurs or persists. Through a conditional triggering mechanism, only regular route planning and reservation management need to be performed most of the time. It can use existing planning failures, reservation failures, and movement authorization results to confirm conflicts, and obtain local state snapshots only as needed after generating a conflict intervention request. This filters communication jitter and instantaneous planning failures, reduces computational overhead and communication load, and can intervene in advance before the vehicle reaches the potential conflict resource, improving scalability and real-time response capabilities.

[0039] This invention performs consistency verification between the original reservation time window for traffic resources and the estimated occupancy time window obtained based on the vehicle's current location, remaining path, and kinematic parameters. It solves the problem of inconsistency between the reservation time window and the actual occupancy time window caused by vehicle deviation from the original plan, control delay, or positioning error. By responding to conflict intervention requests, it calculates the estimated occupancy time window based on the current local state snapshot and verifies its consistency with the original reservation time window. This ensures that conflict diagnosis is always based on the time window that best reflects the current vehicle status. At the same time, the reliability of the time window is further improved through a deviation allowance mechanism, avoiding misjudgments and secondary conflicts caused by reservation failures.

[0040] This invention identifies four conflict types—opposing edge conflict, chasing conflict, point conflict, and cyclic waiting deadlock—and determines the corresponding conflict nodes or edges. Based on the current conflict type, it constructs a scenario-adapted set of candidate resolution actions. The most suitable resolution strategy is precisely matched according to the specific type and characteristics of the conflict. For opposing edge conflict systems, it prioritizes waiting and avoiding or retreating to a buffer node to avoid complex replanning in narrow passages. For chasing conflict systems, it adjusts the following speed to smoothly restore safe distances, avoiding efficiency losses caused by relying solely on stopping and waiting. For point conflicts, it achieves orderly passage through priority passage, waiting and avoiding, retreating to release resources, or local replanning. For cyclic waiting deadlock systems, it breaks loop dependencies through priority passage to avoid the system falling into deadlock. Through the scenario-adapted set of candidate resolution actions, it improves the efficiency and success rate of conflict resolution while ensuring safety, and reduces the impact of resolution actions on vehicle operating efficiency.

[0041] This invention determines the time cost and energy consumption estimation cost of each candidate elimination action in the candidate elimination action set. It then uses the time cost and energy consumption estimation cost to construct a TOPSIS decision matrix, performing vector normalization and weighted normalization on these costs. Positive and negative ideal solutions are determined using cost-type indicators. The basic proximity is calculated, and the preference scores corresponding to the basic proximity and scenario adaptation order are weighted and combined according to the scenario coupling coefficient to obtain a comprehensive ranking of the candidate elimination actions. The TOPSIS multi-objective ranking achieves a balance between the two optimization objectives of time cost and energy consumption estimation cost, avoiding suboptimal solutions caused by single-objective optimization. The scenario coupling coefficient incorporates scenario preferences into the ranking results, preserving both data-based objective ranking and experience-based scenario preferences. The ranking verification shows that the output elimination actions meet multi-objective optimization requirements and are practically feasible, reducing the risk of secondary conflicts after elimination action execution and improving the safety and efficiency of multi-vehicle collaborative operations in warehouse AGVs. Attached Figure Description

[0042] Figure 1 This is a flowchart of the overall system processing of the present invention;

[0043] Figure 2 This is a system operation flowchart of the present invention;

[0044] Figure 3 This is a flowchart of the conflict judgment and decision-making process of the present invention. Detailed Implementation

[0045] The technical solution of the present invention will be further described in detail below with reference to specific embodiments. The following embodiments are used to illustrate the present invention, but should not be used to limit the scope of protection of the present invention. The conditions in the embodiments can be further adjusted according to specific conditions, and simple improvements to the method of the present invention under the premise of the concept of the present invention are all within the scope of protection claimed by the present invention.

[0046] Please refer to Figures 1-3 This embodiment provides a warehouse AGV conflict resolution system that couples TOPSIS and shadow verification in a scenario, including the following modules:

[0047] The condition triggering module is used by the main path planner or motion execution controller to maintain trigger evidence for the same planning, reservation or movement authorization failure object. When the vehicle's subsequent executable path cannot be generated or submitted, traffic resource reservation continues to fail, movement authorization continues to fail, safe parking or vehicle has no effective physical progress, and the operation abnormality reaches the preset intervention time threshold, it confirms that there is a conflict that requires the intervention of the conflict resolution system and generates a conflict intervention request.

[0048] The snapshot parsing module is used to respond to conflict intervention requests. When the request is generated, it obtains the local topology of the warehouse, the vehicle's location code, the vehicle's remaining path, the vehicle's movement status, and the reservation time window. It initially identifies the requesting vehicle, potentially blocked vehicles, and potentially conflicting resources, and constructs a local state snapshot and conflict feature keys.

[0049] The time window verification module is used to verify the consistency between the reserved time window and the estimated occupancy time window obtained based on the vehicle's location code, remaining vehicle path, and kinematic parameters; when the reserved time window is not applicable to the current vehicle status, the estimated occupancy time window is selected.

[0050] The conflict diagnosis module is used to determine the current conflict type, the vehicles involved in the conflict, the conflict nodes or edges, the waiting relationships between vehicles, and the scope of the conflict based on local state snapshots, conflict feature keys, and time windows that have undergone consistency checks. The current conflict type includes at least one of the following: opposing edge conflict, chasing conflict, point conflict, and circular waiting deadlock.

[0051] The candidate action generation module is used to construct a set of candidate resolution actions that are adapted to the current conflict type.

[0052] The TOPSIS ranking module is used to determine the time cost and energy consumption estimation cost of each candidate resolution action in the candidate resolution action set; TOPSIS is used to calculate the basic proximity of the candidate resolution actions, and the basic proximity is softly coupled with the scene preference score corresponding to the conflict type to obtain the ranking of the candidate resolution actions.

[0053] The shadow verification module is used to perform shadow simulation verification of candidate resolution actions in a copy of the local influence domain determined by the local state snapshot, according to the candidate resolution action sorting.

[0054] The output and control handover module is used to output one or more candidate resolution actions and their execution constraints arranged in a comprehensive sort and verified by shadow simulation, and send the candidate action queue to the robot control system. For candidate actions involving local replanning or changes in path or reservation, the main path planner generates and atomically submits executable paths and reservations. For waiting, speed adjustment, and safety holding candidate actions, the motion execution controller executes them within the planning control lease and action constraints.

[0055] The above modules sequentially complete conflict condition confirmation and intervention request generation, local state snapshot parsing and conflict feature key construction, time window verification, conflict diagnosis, candidate action generation, comprehensive sorting, shadow verification, action issuance and execution result review, thus completing the conflict resolution of warehouse AGV vehicles;

[0056] Specifically, the condition triggering module generates a conflict intervention request when the vehicle's subsequent executable path cannot be generated or submitted, traffic resource reservations continue to fail, movement authorization continues to fail, the vehicle stops safely, or the vehicle has no effective physical progress and reaches a preset intervention time threshold; when the conflict is resolved or the conflict object changes, the corresponding status is cleared or updated.

[0057] Specifically, the main path planner or motion execution controller maintains information on requesting vehicles, potential conflicting resources, the time of the first failure, the time of the most recent failure, the number of consecutive failures, and the time of the most recent effective physical progress for planning, reservation, or movement authorization anomalies that have not yet recovered on their own. It updates the triggering evidence when new planning, reservation, or movement authorization results are generated. The duration is calculated based on simulation time or the system's monotonic clock. When the anomaly corresponding to the same requesting vehicle and potential conflicting resource continues to reach a preset intervention time threshold, and at least a preset number of failures occur within that duration, or the relevant vehicle does not generate effective physical progress exceeding a preset distance, and the main path planner still cannot generate or submit a subsequent executable path, complete resource reservation, or restore normal movement authorization, a conflict is confirmed and a conflict intervention request is generated. The preset intervention time threshold can be set to 1 to 5 seconds, the preset number of failures can be set to 2 to 5 times, and the preset distance can be set to a preset spatial radius of 20 to 40 meters. Specific values ​​can be dynamically adjusted based on the warehouse's vehicle density, communication latency, and operational urgency.

[0058] When the master path planner successfully generates and submits an executable path, related resources are released, vehicles make effective physical progress, or the requesting vehicle or potential conflicting resources change, the corresponding triggering evidence is cleared or re-established. While a conflict intervention request is in the accepted or executing state, intervention requests for the same requesting vehicle and potential conflicting resources are not repeatedly generated; after a request is completed, fails, is replaced, or the control lease expires, a decision is made based on the latest planning and operational status to determine whether to regenerate the intervention request.

[0059] The snapshot parsing module responds to conflict intervention requests, acquires and constructs local state snapshots as needed, preliminarily identifies waiting vehicles, blocked vehicles, and conflicting resources based on the local state snapshots, and constructs corresponding conflict feature keys. Conflict feature keys include at least the requesting vehicle identifier, the potentially blocked vehicle identifier, and the potential conflicting resource identifier, used to associate the diagnostic results, candidate resolution actions, execution feedback, and resolution status of the same conflict. Pre-inclusion of the conflict type is not required; when multiple vehicles are involved, conflict component identifiers can be added after conflict diagnosis and grouping.

[0060] Specifically, the snapshot parsing module includes local state snapshots such as warehouse topology, vehicle location, remaining path, movement status, traffic resource reservation, waiting relationships, and conflict confirmation information; the warehouse topology includes nodes, directed edges and their coordinates, length, speed limit, and direction attributes.

[0061] The nodes of the warehouse are location points such as intersections, workstations, and charging stations. Each node has a unique node identifier and coordinates. The edges of the warehouse are feasible path segments connecting the nodes. Each edge contains the starting node, ending node, edge length, speed limit, and direction attributes. The directed edges of the warehouse are directed paths from one node to another. Edges with opposite directions correspond to different driving directions on the same physical edge.

[0062] The vehicle's location is identified by a QR code or magnetic nail, with each code corresponding to a unique coordinate position. The vehicle's current position is determined based on the most recently passed code, the current edge it is on, and the vehicle's movement information.

[0063] The remaining path of a vehicle consists of an ordered series of nodes and directed edges. The remaining path of a vehicle includes the starting node, the ending node, the sequence of nodes and edges traversed, and the estimated entry and exit times of each edge.

[0064] Vehicle motion status includes the vehicle's current speed, target speed, acceleration, deceleration, heading angle, steering status, vehicle size, mission status, movement authorization status, and current conflict resolution action status;

[0065] The local state snapshot is built on demand each time a conflict intervention request is generated. The scope of the local state snapshot includes vehicles and traffic resources within a preset spatial radius centered on the vehicle to be diagnosed or the candidate conflict resource, as well as the remaining paths, reserved resources and waiting dependent vehicles of the relevant vehicles in the prediction time domain. The preset spatial radius is 20 to 40 meters, and different spatial radii and prediction time domains can be preset according to the specific usage environment in actual use.

[0066] Specifically, the time window verification module calculates the estimated time period for a vehicle to occupy various traffic resources based on the vehicle's location, remaining path, and kinematic parameters. For each directed edge in the remaining path, the module calculates the estimated time when the vehicle enters and leaves the directed edge, using common vehicle state position calculation methods. The time range between these two times is used as the estimated time window. Similarly, for each node in the remaining path, the module calculates the time range between the estimated time of arrival and departure from the node. The turning delay refers to the time required for the vehicle to turn at the node, which depends on the vehicle's turning speed and turning angle.

[0067] The reservation time window is a time period that the robot control system pre-allocates to each node or directed edge representing a traffic resource that has been reserved by a vehicle. The reservation time window includes information such as vehicle identification, entry time, exit time, and reservation status, and is used to coordinate the order in which multiple vehicles occupy the same resource to avoid conflicts.

[0068] The consistency check between the scheduled time window and the expected occupied time window includes:

[0069] The reservation end time of the reservation time window is compared with the earliest arrival time of the expected occupancy time window. If the reservation end time plus the allowable deviation is still earlier than the kinematically expected earliest arrival time, it means that the vehicle cannot arrive at the traffic resource on time, the consistency check fails, and the reservation time window is marked as inapplicable. The reservation start time of the reservation time window is compared with the expected completion time of the expected occupancy time window. If the reservation start time is later than the kinematically expected completion time plus the allowable deviation and the vehicle is not in the reservation waiting state, it means that the vehicle has already passed the traffic resource, the consistency check fails, and the reservation time window is marked as inapplicable. The allowable deviation is a preset value used to tolerate normal operation errors, usually set to 2 to 5 seconds.

[0070] In other cases, if the consistency check between the appointment time window and the expected occupancy time window passes, the appointment time window is applicable;

[0071] When the scheduled time window is marked as inapplicable or not applicable to the current vehicle status, the estimated time window is selected. The estimated time window is only used as a valid time window for this conflict diagnosis and does not directly modify the central reservation table. The central reservation table is still updated by the main path planner or robot control system based on the latest vehicle status, actual planning results, and action execution results.

[0072] Among them, the reservation time window or the expected occupied time window determined after consistency verification is used as the valid reservation time window.

[0073] Specifically, the conflict diagnosis module is used to determine the current conflict type, the vehicles involved in the conflict, the conflict nodes or conflict edges, the waiting relationships between vehicles, and the scope of the conflict based on local state snapshots, conflict feature keys, and time windows that have undergone consistency checks; wherein, the current conflict type includes at least one of the following: opposing edge conflict, chasing conflict, point conflict, and circular waiting deadlock.

[0074] Opposing edge conflict identification includes mapping two directed edges with opposite directions to the same physical edge, and identifying the physical edge as a conflicting edge when two vehicles have already occupied the physical edge in opposite directions or when the effective occupation time windows of the two vehicles for the two directed edges overlap within the safety clearance margin.

[0075] Obtain the direction information of all directed edges in the local topology of the warehouse, and map two directed edges with opposite directions to the same physical edge. Specifically, a pair of directed edges with interchanged start and end nodes can be mapped to the same physical edge.

[0076] The system detects whether two vehicles have occupied the same physical edge in opposite directions. Specifically, a vehicle is considered to be occupying the same physical edge in opposite directions when it is currently located on the physical edge and is moving. If it is detected that two vehicles have occupied the physical edge in opposite directions, the conflict between the opposing edges is directly confirmed, and the physical edge is identified as the conflicting edge.

[0077] The system detects whether the effective occupancy time windows of the two directed edges of the physical edge overlap between the two vehicles. The effective occupancy time window is the time window selected by the time window verification module, which may include the scheduled time window or the expected occupancy time window.

[0078] If the effective occupancy time windows of two vehicles on two directed edges overlap within the safety clearance margin, then a conflict on the opposite edge is confirmed. The safety clearance margin is a preset time value that ensures that the vehicles have enough time to clear the conflict area, usually set to 2 to 5 seconds.

[0079] The pursuit conflict identification includes identifying continuous common directed edges with the same direction from the remaining paths of the preceding and following vehicles, constructing one-dimensional topological corridor coordinates according to the side length of the continuous common directed edges, advancing the state of the preceding and following vehicles in the prediction time domain based on vehicle speed, acceleration, deceleration, vehicle length, edge speed limit and turning delay, and determining the moment when the effective distance between vehicles is first no greater than the dynamic target safety distance as the pursuit risk time.

[0080] Identify continuous common directed edges with the same direction in the remaining paths of the preceding and following vehicles. The preceding vehicle is the vehicle traveling ahead of the path, and the following vehicle is the vehicle traveling behind the path. Continuous common directed edges are the sequence of directed edges with the same direction and continuous overlap in the remaining paths of the preceding and following vehicles.

[0081] Construct a one-dimensional topological corridor coordinate system based on the length of the continuous common directed edges; map the position of each vehicle to a coordinate value in the linear coordinate system with the starting point of the continuous common directed edges as the origin and the vehicle driving direction as the positive direction.

[0082] Within the prediction time domain, the states of the preceding and following vehicles are advanced based on vehicle speed, acceleration, deceleration, vehicle length, edge speed limit, and turning delay. The prediction time domain is a preset time period used to predict the future state of the vehicles, typically set to 10 to 30 seconds. The advancement of the vehicle state uses a kinematic model, updating the vehicle's position, speed, and acceleration at each simulation step. The turning delay is the time required to turn at a node, depending on the vehicle's turning speed and turning angle, typically set to 1 to 3 seconds.

[0083] The effective vehicle spacing is calculated as the distance between the rear of the preceding vehicle and the front of the following vehicle. The formula for calculating the effective vehicle spacing is: Effective vehicle spacing = (one-dimensional topological corridor coordinates of the preceding vehicle - half the length of the preceding vehicle) - (one-dimensional topological corridor coordinates of the following vehicle + half the length of the following vehicle).

[0084] The moment when the effective distance between vehicles is first no greater than the dynamic target safety distance is defined as the overtaking risk time. The dynamic target safety distance is a preset safety distance that changes dynamically over time.

[0085] Among them, the dynamic target safety distance is the sum of the basic static distance, the travel distance of the following vehicle within the control link response time, the positive difference between the braking distance of the following vehicle and the braking distance of the preceding vehicle, and the safety margin.

[0086] The basic static distance is the minimum safe distance between vehicles when they are stopped, and it is usually set to 1 to 2 meters.

[0087] The distance traveled by the following vehicle within the control link response time represents the distance traveled by the following vehicle between receiving the control command and starting to execute the response; the control link response time includes communication delay, calculation delay and actuator response time, which is usually 100ms to 500ms; the distance traveled by the following vehicle within the control link response time is equal to the current speed of the following vehicle multiplied by the control link response time.

[0088] A positive difference between the braking distance of the following vehicle and the braking distance of the preceding vehicle indicates the safety compensation caused by the difference in braking capabilities between the two vehicles; if the braking distance of the following vehicle is greater than that of the preceding vehicle, the difference is positive; otherwise, the difference is 0.

[0089] Safety margin is an additional safety buffer distance, typically set to 0.5 meters to 1 meter.

[0090] Point conflict identification includes determining intersection spatiotemporal conflicts by comparing the arrival-to-departure time windows of multiple vehicles at intersection nodes or conflict zone nodes, and suppressing intersection spatiotemporal conflicts when the non-overlapping passage order has been determined in the effective reservation time window; and searching for nodes that are stably occupied by other vehicles in the predicted time domain and cannot be released before the arrival of the requesting vehicle along the future path of the requesting vehicle to determine future occupation node conflicts. Here, future occupation node conflicts and intersection spatiotemporal conflicts are parallel subtypes.

[0091] Compare the arrival-to-departure time windows of multiple vehicles at intersection nodes or conflict zone nodes to determine whether there is time window overlap; where intersection nodes are nodes where multiple paths intersect, conflict zone nodes are nodes where there may be resource competition, and the arrival-to-departure time window is the time period between the time when the vehicle is expected to arrive at the node and the time when the vehicle is expected to leave the node.

[0092] If the arrival-to-departure time windows of two or more vehicles at the same intersection node or conflict zone node overlap, then the intersection spatiotemporal conflict is confirmed; if within the valid reservation time window, a non-overlapping passage order has been determined and a non-overlapping time window sequence has been assigned to the node, then the spatiotemporal conflict at the intersection is suppressed and no further diagnosis is needed.

[0093] Along the future path of the requesting vehicle, search for nodes that are stably occupied by other vehicles in the predicted time domain and cannot be released before the arrival of the requesting vehicle to determine future node occupancy conflicts; where stable occupancy means that the time window of other vehicles occupying the node has been confirmed and cannot be released before the arrival of the requesting vehicle; specifically, if the departure time of other vehicles occupying a node in the time window is later than the expected arrival time of the requesting vehicle in the node, then a future node occupancy conflict is determined.

[0094] Circular wait deadlock identification includes constructing a waiting relationship graph by expanding outwards from vehicles as vertices and the relationship between waiting vehicles and blocking vehicles as directed edges, and confirming a circular wait deadlock when a directed cycle exists in the local waiting relationship graph.

[0095] Using the vehicles involved in the conflict as vertices and the relationships between waiting vehicles and blocking vehicles as directed edges, a waiting relationship graph is constructed. Here, waiting vehicles are those that have stopped moving because resources are occupied, and blocking vehicles are those that occupy the resources. Specifically, if vehicle A is waiting for vehicle B to release resource R, then a directed edge is constructed from A to B.

[0096] Detect whether there is a directed cycle in the waiting relationship graph. A directed cycle is a closed path that starts from a certain vehicle and returns to the vehicle itself along a directed edge. If a directed cycle exists, record the set of vehicles and the set of conflicting resources of the directed cycle.

[0097] Specifically, after determining the conflict type, conflicting vehicles, and conflicting resources, the candidate action generation module generates candidate resolution actions that are suitable for the current conflict scenario. The candidate resolution actions include priority passage, buffer retreat, local replanning, following speed adjustment, and safety maintenance.

[0098] Priority passage is used to identify vehicles that should proceed first, and to allow other vehicles to wait without obstructing the path ahead. The priority vehicles are determined based on task priority, waiting time, resource availability, and task urgency.

[0099] When a waiting vehicle blocks the path of an oncoming vehicle and the conflict cannot be resolved within a preset time, a buffer retreat action is generated, causing the vehicle to retreat to an available buffer node. Buffer nodes include predefined buffer nodes and dynamically escaped nodes determined based on the vehicle's current position, adjacency, node occupancy, reservation status, and the vehicle's retreat capability.

[0100] Local replanning actions use conflict nodes or conflict edges as restricted resources, prioritizing planning to the merging point ahead of the original remaining path. If the merging point cannot be reached, planning is done to the local window exit, generating planning constraint instructions that include the executing vehicle, restricted resources, merging point or local window exit, execution location, initial constraint, action release condition, and control lease period. The actual executable path and corresponding reservation are generated and submitted by the main path planner.

[0101] The following speed adjustment action is applicable to pursuit conflicts where the vehicle in front is still moving and the vehicle behind is continuously approaching. Based on the current effective distance, the dynamic target safe distance, the vehicle's acceleration and deceleration capabilities, and the control time domain, the upper limit of the safe speed of the following vehicle is determined, so that the following vehicle can decelerate smoothly and restore the safe distance. When the vehicle in front has stopped, it is handled as a point conflict.

[0102] The safety hold action is used to keep the selected vehicle stationary or prevent it from entering the conflicting resource for a bounded hold time when other candidate resolution actions are not feasible, and to request the master path planner to release, postpone or rebuild future appointments that have not yet been executed; the safety hold action is only a temporary safety measure and is not directly used as a basis for determining that the conflict has been resolved.

[0103] Conflicts on opposite sides correspond to priority passage, buffer backoff, or local replanning actions; priority in chasing conflicts corresponds to car-following speed adjustment actions; spatiotemporal conflicts at intersections correspond to priority passage actions; conflicts involving future occupied nodes correspond to waiting, buffer backoff, or local replanning actions; cyclic waiting deadlock generates priority passage, buffer backoff, or local replanning actions by selecting key waiting relationships in the waiting loop, and adopts a safe hold action when none of the above actions are feasible.

[0104] Specifically, the TOPSIS ranking module determines two indicators for each candidate elimination action: time cost and estimated energy cost. The time cost is the additional time required to perform the candidate elimination action compared to the original plan, and the estimated energy cost is the additional energy consumption required to perform the candidate elimination action compared to the original plan.

[0105] Time costs include waiting time, detour time, and acceleration / deceleration time. Waiting time is the time a vehicle spends waiting at a conflict node or edge; detour time is the additional travel time incurred by a vehicle due to rerouting; and acceleration / deceleration time is the additional time incurred by a vehicle due to changing its speed.

[0106] Energy consumption cost estimates include acceleration energy consumption, braking energy consumption, and cornering energy consumption. Acceleration energy consumption is the energy consumed by the motor when the vehicle accelerates, braking energy consumption is the energy consumed when the vehicle brakes, and cornering energy consumption is the energy consumption that increases due to the change of direction when the vehicle turns.

[0107] The calculation of time cost and energy consumption estimated cost is based on the vehicle's kinematic model and path information; for waiting actions, the time cost equals the waiting time, and the energy consumption estimated cost equals the energy consumed by the vehicle to maintain its state while waiting; for detour actions, the time cost equals the estimated travel time of the detour path, and the energy consumption estimated cost equals the estimated energy consumption of the detour path.

[0108] The TOPSIS decision matrix is ​​constructed using time cost and energy consumption estimated cost. For n candidate elimination actions, the TOPSIS decision matrix is ​​determined to be an n-row, 2-column matrix, with each row corresponding to a candidate elimination action and each column corresponding to a cost index.

[0109] Vector normalization is applied to the estimated costs of time and energy consumption to ensure that the normalized values ​​of each cost indicator are on the same dimension. The formula for vector normalization is as follows:

[0110]

[0111] In the formula, For the first The first candidate elimination action Normalized values ​​of each cost indicator; This is only for normalized summation indexes, to avoid merging with candidate resolution action indices. Confusion; Indicates the first The first candidate elimination action One cost indicator value, among which Indicates time cost, This indicates the estimated cost of energy consumption;

[0112] When the normalized denominator of a certain indicator is zero, division is not performed, and all normalized values ​​of the indicator are set to zero so that it is not included in the differentiation and ranking.

[0113] The normalized matrix is ​​then subjected to weighted normalization. Specifically, the weights of time cost and energy consumption estimation cost can be preset or dynamically adjusted according to scenario requirements. In one embodiment, the weights of both time cost and energy consumption estimation cost are set to 0.5, meaning that time cost and energy consumption estimation cost are equally important.

[0114] In another implementation, in a high-efficiency production scenario, the weight of time cost is set to 0.7 and the weight of energy consumption estimation cost is set to 0.3, with time cost taking priority; in an energy-saving scenario, the weight of time cost is set to 0.3 and the weight of energy consumption estimation cost is set to 0.7, with energy cost taking priority.

[0115] Positive and negative ideal solutions are determined using cost-type indicators. Cost-type indicators refer to those with smaller values, while time cost and estimated energy cost are both cost-type indicators. The positive ideal solution is the set of minimum values ​​for each cost indicator, which represents the minimum time cost and the minimum estimated energy cost among all candidate elimination actions. The negative ideal solution is the set of maximum values ​​for each cost indicator, which represents the maximum time cost and the maximum estimated energy cost among all candidate elimination actions.

[0116] Calculate the Euclidean distance from each candidate resolution action to the positive and negative ideal solutions, and calculate the basic proximity based on the Euclidean distance. Basic proximity = Euclidean distance to the negative ideal solution / (Euclidean distance to the negative ideal solution + Euclidean distance to the positive ideal solution).

[0117] The basic proximity score ranges from 0 to 1. The closer the basic proximity score is to 1, the farther the candidate solution is from the negative ideal solution and the closer it is to the positive ideal solution, resulting in better overall performance in terms of time cost and energy consumption estimation cost. The closer the basic proximity score is to 0, the farther the candidate solution is from the positive ideal solution and the closer it is to the negative ideal solution, resulting in worse overall performance.

[0118] After obtaining the basic proximity score, the preference scores corresponding to the basic proximity score and the scene adaptation order are weighted and combined according to the scene coupling coefficient to obtain the comprehensive ranking of candidate resolution actions. The preference score corresponding to the scene adaptation order is a preset preference value based on the conflict type and application scenario. The scene adaptation order is the applicability priority order of each candidate resolution action pre-set according to the current conflict type. The order of the candidate resolution action in the scene adaptation order is used to determine the scene preference score, but does not directly replace the TOPSIS ranking result. The preference score reflects the degree of preference of a certain type of resolution action relative to other types of resolution actions in a specific scenario.

[0119] The formula for calculating preference scores includes:

[0120]

[0121] In the formula, As a scene type, The current number of candidate actions, Candidate actions In the order of scene adaptation This represents the preference score for this scenario type;

[0122] The formula for calculating the scene coupling comprehensive score includes:

[0123]

[0124] In the formula, Candidate actions Basic proximity, The preference score for this scenario type. The scene coupling coefficient, This is the overall score for scene coupling in this scene type;

[0125] The scenario coupling coefficient is a weighting parameter used to balance basic proximity and scenario preference. Its value ranges from 0 to 1. The value of the scenario coupling coefficient determines the weight ratio of basic proximity and scenario preference in the final ranking. When the scenario coupling coefficient is 0, the ranking degenerates into a pure two-index TOPSIS. When the scenario coupling coefficient is greater than 0, scenario preference affects the ranking in a soft constraint manner, but does not directly cover the time and energy consumption evaluation results.

[0126] Candidate elimination actions are sorted from highest to lowest according to their comprehensive ranking scores to obtain the candidate elimination action ranking; the candidate elimination action with the highest comprehensive ranking score is the optimal elimination action, which will be submitted to the shadow verification module for simulation verification first.

[0127] Specifically, the shadow verification module sorts the candidate resolution actions and performs shadow simulation verification on the candidate resolution actions in the copy of the local influence domain determined by the local state snapshot;

[0128] The local impact domain is determined based on conflicting vehicles, congested vehicles, conflicting resources, the impact path of candidate actions, and safety margin; only the vehicle status, path, reservation, and waiting relationship that may be modified by candidate actions within the local impact domain are deeply copied, and the effective reservations and traffic resources related to candidate actions outside the local impact domain are treated as read-only boundary conditions;

[0129] Among them, the local impact domain is the range of vehicles and traffic resources within the area centered on the conflicting vehicles or conflicting resources, plus a preset spatial radius and a safety margin. The safety margin is a preset value used to expand the coverage of the impact domain.

[0130] The local influence domain copy is independent of the original local state snapshot. The vehicle state advancement, candidate path preview, and scheduled trial of the copy will not modify the actual robot control system. The topology data adopts read-only caching, and the candidate actions reuse the unchanged read-only data to reduce copying and memory overhead.

[0131] In the local influence domain copy, the relevant vehicle status is advanced according to the verification step size, and the speed, position, candidate path, and reservation trial of the relevant vehicles are updated according to the candidate actions. For vehicles outside the local influence domain, their dynamic status is not advanced one by one, but their effective reservations, boundary entry time windows, and resource occupancy status are used as external constraints to verify whether the candidate actions will conflict with the traffic status outside the local influence domain. The above verification step size can be dynamically adjusted according to the real-time speed of the vehicle. It is usually set to the same value or smaller than the control cycle of the vehicle control system of 100ms. When the vehicle is at low speed or stationary, a larger step size such as 500ms can be used to improve the verification efficiency and ensure that the step size is not greater than the minimum time interval at which the vehicle may undergo significant state changes.

[0132] The continuous edge occupancy table is used to quickly check in each simulation step whether the same directed edge is occupied by multiple vehicles at the same time, and whether directed edges in opposite directions correspond to the same physical edge being occupied at the same time.

[0133] The continuous edge occupancy table is a data structure used to record the simulation steps when each vehicle enters a directed edge and the simulation steps when it can release the directed edge after reaching the endpoint node. Each row of the continuous edge occupancy table corresponds to a vehicle's occupancy of a directed edge. The continuous edge occupancy table includes fields such as vehicle identifier, directed edge identifier, entry simulation step, and expected departure simulation step. The entry simulation step is the simulation step number corresponding to the moment when the vehicle enters the directed edge; the expected departure simulation step is the simulation step number corresponding to the moment when the vehicle can release the directed edge after reaching the endpoint node.

[0134] Shadow verification verifies node capacity, same-direction edge capacity, simultaneous occupancy of directed edges in opposite directions on the same physical edge, reservation time window conflict, and recurrence of waiting relationships in each simulation step. If any constraint is not satisfied, the current candidate verification stops and the next candidate is verified until a preset number of safe candidates are obtained or all candidates are verified.

[0135] The output lists the candidate actions that pass the shadow verification according to the comprehensive sorting, along with the execution constraints corresponding to each candidate action. If all candidate resolution actions fail to be verified, the shadow verification module outputs a signal that there are no available resolution actions, triggering system-level scheduling intervention, global replanning, or manual alarm.

[0136] Specifically, the output and control handover module sends a queue of candidate actions, arranged in a comprehensive sort and verified by shadow, to the robot control system. It then requests a planning control lease with a time limit and scope. For candidate actions involving partial replanning or changes to paths or reservations, the main path planner generates and atomically submits executable paths and reservations. For waiting, speed adjustment, and safety holding candidate actions, the motion execution controller executes them within the planning control lease and action constraints. The candidate action queue includes at least the action type, target vehicle, execution constraints, action release conditions, sorting position, and verification results, along with conflict flags, snapshot timing flags, reservation versions, and lease durations.

[0137] Before accepting a candidate action, the robot control system verifies whether conflicts still exist, whether snapshots and reservation versions are applicable, whether the relevant vehicle has crossed the execution position, and whether there are other valid control leases in the current scope. Once the verification is successful, a planning control lease is granted. During the lease's validity period, the conflict resolution system provides action semantics and planning constraints, while the master path planner maintains sole control over executable paths, reservation allocations, and movement authorizations.

[0138] For candidate actions that require replanning or adjustment of appointments, the main path planner generates and atomically submits the path and appointment based on the real-time appointment status; if the current candidate action cannot be executed, the original path and appointment are maintained, and the next candidate action is tried according to the candidate queue order.

[0139] Once a candidate action is accepted, the robot control system issues speed, waiting, movement authorization, or path instructions to the relevant vehicle and tracks the action status locally. Event-based execution feedback is sent only when the action is accepted, rejected, executed, completed, failed, or substituted. The feedback includes action-related information, status, reason for failure, actual vehicle status, and effective physical progress.

[0140] The output and control handover module reviews conflicting resources and action resolution conditions based on event-driven execution feedback. When conflicting resources are released and resolution conditions are met, it records the actual executed action and resolution time, releases the planning control lease, and resumes normal planning. If a candidate action fails but the conflict still exists, it tries the next candidate action. If a snapshot expires or the conflicting object changes, it terminates the current intervention and re-evaluates. If all candidate actions fail, it triggers system-level scheduling, global replanning, or manual alarm.

[0141] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A conflict resolution system for warehouse AGVs that couples TOPSIS and shadow verification in a scenario, characterized in that, include: The condition triggering module is used to generate a conflict intervention request when the abnormal operation continues for a preset time threshold. The snapshot parsing module is used to respond to conflict intervention requests, obtain the local topology of the warehouse, vehicle location, vehicle remaining path, vehicle movement status, appointment time window and trigger evidence, and construct a local state snapshot; The time window verification module is used to verify the consistency between the scheduled time window and the kinematically estimated time window, and to select the estimated time window when the scheduled time window is not applicable. The conflict diagnosis module is used to determine the conflict type and conflict node or conflict edge based on the local state snapshot and the time window selected after verification. The candidate action generation module is used to construct a set of candidate resolution actions based on the conflict type. The TOPSIS ranking module is used to calculate the basic proximity based on time cost and energy consumption estimation cost, and obtain a comprehensive ranking by weighting the scenario preference scores corresponding to the conflict type. The shadow verification module is used to advance the vehicle state based on candidate actions and verify traffic resource occupancy constraints in a local state snapshot copy according to the comprehensive sorting. The output and control handover module is used to send the verified candidate actions and execution constraints to the robot control system.

2. The warehouse AGV conflict resolution system according to claim 1, characterized in that: The operational anomalies in the condition triggering module include: maintaining trigger evidence by the main path planner or motion execution controller for the same failed planning, reservation or movement authorization object; and generating a conflict intervention request when subsequent executable paths continue to fail to be generated or submitted, resource reservations continue to fail, movement authorizations continue to fail, safe parking occurs or no effective physical progress reaches the time threshold. The snapshot parsing module is also used to obtain the vehicle location code and trigger evidence in response to conflict intervention requests, and to construct conflict feature keys with the identifiers of the requesting vehicle, the potentially blocked vehicle and the potential conflict resource, for association with diagnostic results, candidate actions and execution feedback.

3. The warehouse AGV conflict resolution system according to claim 1, characterized in that, The time window verification module is also used for: If the reservation end time plus allowable deviation is still earlier than the earliest kinematically predicted arrival time, or if the reservation start time is later than the kinematically predicted completion time plus allowable deviation and the vehicle is not in a reservation waiting state, the corresponding reservation time window will be marked as inapplicable, and the predicted occupancy time window obtained based on the ground code location and kinematic parameters will be used.

4. The warehouse AGV conflict resolution system according to claim 1, characterized in that, The conflict diagnosis module is also used for: Map two directed edges with opposite directions to the same physical edge, and identify the physical edge as a conflict edge when two vehicles have already occupied the physical edge in opposite directions, or when the effective occupation time windows of the two vehicles for the two directed edges overlap within the safety clearance margin. Intersection spatiotemporal conflicts are determined by comparing the arrival-to-departure time windows of multiple vehicles at intersection nodes or conflict zone nodes, and intersection spatiotemporal conflicts are suppressed when the non-overlapping passage order has been determined in the effective reservation time window. Search along the future path of the requesting vehicle to find nodes that are stably occupied by other vehicles in the predicted time domain and cannot be released before the requesting vehicle arrives, thus identifying future node conflicts.

5. The warehouse AGV conflict resolution system according to claim 1, characterized in that, The conflict diagnosis module is also used for: Identify continuous common directed edges with the same direction from the remaining paths of the preceding and following vehicles, construct one-dimensional topological corridor coordinates according to the side length of the continuous common directed edges, and advance the state of the preceding and following vehicles in the prediction time domain based on vehicle speed, acceleration, deceleration, vehicle length, edge speed limit and turning delay, and determine the moment when the effective distance between vehicles is not greater than the dynamic target safety distance as the pursuit risk time. A local waiting graph is constructed with vehicles as vertices and the relationship between waiting vehicles and blocking vehicles as directed edges. When a directed cycle exists, the loop waiting deadlock is checked based on the conflict duration, appointment failure, and trigger evidence of no effective physical progress.

6. The warehouse AGV conflict resolution system according to claim 1, characterized in that, The candidate action generation module is also used for: In response to pursuit conflicts, based on the dynamic target safety distance and the current effective distance, the upper limit of the following vehicle speed is determined to ensure that the effective distance between vehicles is not less than the dynamic target safety distance within the control time domain, and the upper limit of the following vehicle speed is used as the target speed for the candidate speed adjustment action; wherein, the dynamic target safety distance includes the basic static distance, the travel distance of the following vehicle within the control link response time, the positive difference between the braking distance of the following vehicle and the braking distance of the preceding vehicle, and the safety margin. When all other candidate resolution actions are not feasible, a safety hold action is generated to keep the selected vehicle stationary or prevent it from entering the conflicting resource within a bounded hold time, and the master path planner is requested to release, postpone or rebuild future appointments that have not yet been executed; based on the event-driven execution feedback, the latest conflicting resource status and action resolution conditions, if the conflict is not resolved, a candidate resolution action is regenerated or a system-level scheduling intervention is requested.

7. The warehouse AGV conflict resolution system according to claim 1, characterized in that, The candidate action generation module is also used for: The system identifies the leading vehicle and generates waiting candidate actions for low-priority vehicles that do not block the remaining path of the leading vehicle. When a low-priority vehicle blocks the remaining path of the leading vehicle and there is an available buffer node, a candidate action is generated to make the low-priority vehicle retreat to the buffer node. The buffer node includes a predefined buffer node and a dynamic escape node determined based on the direct adjacent edge of the current position, the node occupancy status, the reservation status, and the vehicle's retreat capability. Using conflict nodes or conflict edges as restricted resources, prioritize the reachable merging point ahead of the original remaining path as the local planning target. If no reachable merging point is found, select the local window exit as the target, and generate a replanning constraint instruction that includes at least the restricted node or restricted edge, merging point or local window exit, executing vehicle, executing location, initial constraint, and control lease term.

8. The warehouse AGV conflict resolution system according to claim 1, characterized in that, The TOPSIS sorting module is also used for: The TOPSIS decision matrix is ​​constructed using time cost and energy consumption estimated cost. Vector normalization and weighted normalization are performed on time cost and energy consumption estimated cost. Positive ideal solution and negative ideal solution are determined using cost-type indicators. The Euclidean distance from each candidate resolution action to the positive ideal solution and negative ideal solution is calculated, and the basic proximity is calculated based on the Euclidean distance. After obtaining the basic proximity score, the preference scores corresponding to the basic proximity score and the scene adaptation order are weighted and combined according to the scene coupling coefficient to obtain the comprehensive ranking of candidate resolution actions.

9. The warehouse AGV conflict resolution system according to claim 1, characterized in that, The shadow verification module is also used for: The local influence domain is determined based on the conflicting vehicles, congested vehicles, conflicting resources, and safety margins in the local state snapshot. Only the vehicle status, reservation, and route data that need to be modified by the candidate action within the local influence domain are deeply copied, and the reservations and traffic resources outside the influence domain are treated as read-only boundary conditions. Record the occupied range of a vehicle from entering a directed edge to reaching the destination node and being able to release the directed edge, forming a continuous edge occupancy table; In each simulation step, the node capacity, the capacity of the same-direction edge, the simultaneous occupancy of directed edges in opposite directions on the same physical edge, the reservation time window conflict, and the recurrence of waiting relationships are verified. When any hard constraint fails, the current candidate verification is stopped and the next candidate is verified.

10. The warehouse AGV conflict resolution system according to claim 1, characterized in that, The output and control handover module is also used for: Candidate resolution actions verified through shadow simulation are sorted into a candidate queue, and a conflict identifier, snapshot timing identifier, reservation version, execution location and planning control lease period are attached and sent to the robot control system. The robot control system accepts candidate actions when conflicts still exist, snapshots and reservation versions are applicable, the vehicle has not crossed the execution position, and there are no other valid control leases within the current scope; if path or reservation changes are involved, the main path planner generates and atomically submits the path and reservation; waiting, speed adjustment, and safety holding actions are executed by the motion execution controller within the lease and action constraints. When a candidate action is rejected, the path generation fails, or the original submission fails, the original path and the original appointment are maintained, and the next candidate is tried; event-driven execution feedback is received, and conflicting resources and action resolution conditions are reviewed.

Images