An agv cross-scene task scheduling and multi-state linkage method and system

Abstract

The application relates to the technical field of automatic guided vehicle scheduling control, and discloses an AGV cross-scene task scheduling and multi-state linkage method and system, which comprises the following steps: a task management layer publishes a first target point as an initial pose and activates navigation; a core scheduling layer reads a behavior identifier bound to a current target point and issues a corresponding action package, a bottom layer execution layer receives the action package and executes corresponding transportation operations; the core scheduling layer respectively executes abnormal retry or task rollback operations according to state data; the bottom layer execution layer sets a sliding window for the front door and the back door of an elevator respectively to execute confidence degree evaluation; the core scheduling layer issues a mode number to the bottom layer execution layer; the bottom layer execution layer calculates an error between an actual pose of an AGV and a preset target pose; and real-time pose data output by the bottom layer execution layer is used as input data for pose legality verification when a task is connected and recovered. The application effectively improves the positioning accuracy and scheduling operation stability of AGV multi-scene operation.

Description

Technical Field

[0001] This application relates to the field of automated guided vehicle (AGV) scheduling and control technology, and in particular to a method and system for cross-scenario task scheduling and multi-state linkage of AGVs. Background Technology

[0002] In smart construction scenarios, AGVs often perform material transport tasks across floors and over long distances on construction sites. Unlike standardized factory buildings, construction site environments are characterized by high dynamism and strong interference, and frequent human intervention occurs during construction. Therefore, AGV scheduling and control systems in construction environments typically face the following problems: (1) Strong coupling and difficult system maintenance: Traditional scheduling schemes often use hard-coded finite state machines. When faced with complex scenarios such as loading and unloading, entering and exiting construction elevators, and abnormal interruptions, they are prone to triggering logical deadlocks. Furthermore, when adding new scenarios, the state transition matrix needs to be modified globally, making it difficult to expand new functions.

[0003] (2) Cross-floor positioning is easy to lose: When the robot switches floors in the elevator, if the new floor coordinates are directly issued without updating the map and positioning points in sync, the particle swarm of the global positioning algorithm is very easy to diverge, resulting in navigation errors.

[0004] (3) Sensor interference can easily lead to misoperation: The construction site is dusty, the light is variable, and the sensors in the elevator door area often generate instantaneous blocking noise. Existing solutions usually rely on single frame or a few frame signals to trigger the entry and exit of the elevator. They lack a confirmation mechanism for continuous stability and are prone to misjudging personnel movement or dust as the door being open, thus causing safety accidents of AGV hitting the elevator door. Summary of the Invention

[0005] In view of this, this application provides a method and system for cross-scenario task scheduling and multi-state linkage of AGVs. It adopts a behavior ID routing scheduling framework based on behavior tree organization, decoupling task parsing, state polling, and underlying execution, and supplements it with... Figure 1 Consistency comparison and state confidence assessment improve the system's execution reliability in complex environments.

[0006] This application discloses a method for cross-scenario task scheduling and multi-state linkage of AGVs, which includes: Step 1: The task management layer receives the task sequence carrying the sequence field through the message queue telemetry transmission protocol, sorts all target points according to the sequence field of each task point, publishes the first target point as the initial pose and activates navigation. Step 2: The core scheduling layer relies on periodic polling to detect the navigation execution progress. After the navigation is completed, it reads the behavior identifier bound to the current target point and sends out the corresponding action package. The underlying execution layer receives the action package and executes the corresponding transportation operation, and sends back the status data to the core scheduling layer. The core scheduling layer performs abnormal retry or task rollback operations for the three scenarios of loading and unloading, elevator entry and exit, and navigation, respectively, based on the status data. Step 3, Core scheduling layer execution location Figure 1 In the consistency verification process, after the elevator arrives at the target floor, the core scheduling layer performs the same operation. Figure 1 Consistency verification process, location Figure 1 The consistency verification process determines whether to continue issuing subsequent navigation target points; Step 4: The bottom execution layer collects sensor data from the front and rear doors of the elevator, and sets up sliding windows for confidence assessment for the front and rear doors respectively. When the sensor recognizes the door opening signal, the confidence built into the window increases by a fixed step size, and when interference signals appear, the confidence decreases by a fixed step size. The latched door opening status is output only when the confidence reaches the preset threshold. Step 5: The core scheduling layer executes the elevator interaction three-node closed-loop status confirmation process. Based on the current execution stage of the task, it issues a mode number to the lower execution layer. The lower execution layer stores the PID parameters of various working conditions, calculates the error between the AGV's current actual pose and the preset target pose. After the error converges to the corresponding tolerance range, the lower execution layer sends an action completion signal back to the core scheduling layer. Step 6: Upon receiving a navigation cancellation command, the underlying execution layer terminates navigation operation, the core scheduling layer disables all timed polling, clears status flags, and fully preserves the original task target point sequence and planned path; upon receiving a navigation continuation command, the underlying execution layer uses a temporary subscription method to collect real-time poses, and cancels the temporary subscription after obtaining valid real-time poses; the real-time pose data output by the underlying execution layer serves as input data for pose validity verification during task continuation and recovery.

[0007] Furthermore, the execution process of the behavior ID routing scheduling framework includes: After navigation is completed by periodic polling, the behavior identifier bound to the current target point is read. The behavior identifier is divided into five categories: loading, unloading, going up the elevator, going down the elevator, and rotating in place. There is a blank behavior identifier when no specific transportation action is specified. The core scheduling layer sends the corresponding action package to the underlying execution layer based on the behavior identifier. The underlying execution layer sends back three execution statuses to the core scheduling layer: execution in progress, execution completed, and execution error. The method of performing abnormal retry or task rollback operations based on status data for three scenarios—loading and unloading, elevator entry and exit, and navigation—includes: In loading and unloading scenarios, repeated retries are performed. Once the number of retries exceeds the maximum limit, an exception is reported and the current target point is skipped. In elevator entry and exit scenarios, the system returns to the initial standby state before the elevator call and re-executes the elevator call and alignment operations. In navigation scenarios, the current navigation target is canceled and an exception is reported, waiting for external devices to reissue task instructions. When adding new operation scenarios, only the corresponding behavior identifier needs to be extended, without modifying the core scheduling logic.

[0008] Furthermore, the land Figure 1 The consistency verification process includes: First, determine whether the current operation belongs to one of the two scenarios that require verification. The two scenarios are before performing the elevator up / down floor switching operation and before parsing and sending the next navigation target point. If the map belongs to either of the two scenarios, the map identifier of the currently loaded map is compared with the map identifier of the target scenario map. If the identifiers do not match, all navigation target instructions are suspended, and a map update instruction is sent to the underlying execution layer. The underlying execution layer loads the target map asynchronously. If map loading fails, the task flow is continuously suspended and an exception is reported. The entire task flow is resumed after the map is fully loaded.

[0009] Furthermore, the step of setting up independent sliding windows for the front door and the rear door to perform confidence assessment includes: The underlying execution layer independently collects sensor data from both the front and rear doors of the elevator through a dedicated topic, and configures a dedicated sliding window for each. When the sensor recognizes an opening signal, the built-in confidence level of the corresponding sliding window increases by a fixed step. When interference signals occur, the built-in confidence level of the sliding window decreases by a fixed step. Interference types include dust obstruction, personnel movement, and momentary door closing. Only when the built-in confidence level of the sliding window reaches a preset threshold is the latched valid opening status output. When the core scheduling layer issues elevator entry / exit completion instructions or switches task execution nodes, it automatically clears all confidence levels in the sliding window to eliminate latched data residue. If the waiting time for the door to open exceeds a preset threshold, the elevator door is directly determined to be faulty, and the elevator interaction task rollback process is initiated.

[0010] Furthermore, the execution process of the elevator interaction three-node closed-loop status confirmation process includes: First, the system calls the elevator and waits for it to arrive. If the waiting time exceeds a threshold or an anomaly is detected, it immediately returns to the initial standby state before the elevator call. Once the elevator is in place and there are no anomalies, three sequential process nodes are executed: The first process node is the AGV's coarse alignment of the passageway to the elevator perimeter. The navigation module of the bottom execution layer navigates to the elevator perimeter and switches the corresponding PID parameters. After the bottom execution layer reports that the alignment is complete, it proceeds to the second process node. If the execution times out, it returns to the initial standby state before the elevator call. The second process node is the AGV's fine docking at the elevator door area. The bottom execution layer performs precise alignment of the elevator feed door. After the bottom execution layer reports that the docking is complete, it proceeds to the third process node. The third process node is the AGV's entry into the elevator car. Before entering this stage, the latched door opening status must be verified. If a valid latched door opening status is not detected, entry into the elevator is prohibited. After all three process nodes are completed, the navigation coordinates are paused, and the AGV remains stationary, waiting for the elevator to arrive at the target floor.

[0011] Furthermore, the sequential execution of the three process nodes, which have a specific order, includes: The first process node is the execution stage where the AGV travels to the outside of the elevator and completes the coarse alignment of the passageway. After the bottom execution layer navigation module navigates to the outside of the elevator, it switches the corresponding PID parameters. Only after the bottom execution layer reports that the alignment is completed can it enter the second process node. If the execution times out, it returns to the initial standby state before the elevator call. The second process node is the execution stage where the AGV arrives at the elevator door area and completes the fine docking. The bottom execution layer performs the precise alignment operation of the elevator feed door. After the bottom execution layer reports that the docking is completed, it enters the third process node. The third process node is the execution stage where the AGV drives into the elevator car. Before entering this stage, the latched door opening status must be verified. If a valid latched door opening status is not detected, the AGV is prohibited from entering the elevator.

[0012] Furthermore, the step of sending the PID mode number to the underlying execution layer includes: Based on the execution stage of the task, the core scheduling layer issues the corresponding mode number for corridor driving, target tracking, and elevator entry to the lower execution layer. After receiving the mode number, the lower execution layer calls the PID parameters configured independently for that mode, and the PID parameters for different modes are isolated from each other. The lower execution layer calculates the error between the AGV's current actual pose and the preset target pose in real time. After the error converges to the corresponding tolerance range, the lower execution layer sends a dedicated action completion signal to the core scheduling layer. The core scheduling layer relies on this signal to advance the task flow within the behavior ID routing scheduling framework.

[0013] Furthermore, the task continuation recovery includes: Upon receiving a navigation cancellation command, the underlying execution layer terminates navigation, the core scheduling layer disables all timed polling, clears status flags, and fully preserves the original task target point sequence and planned path count. Upon receiving a continue navigation command, the underlying execution layer continuously collects real-time pose data via temporary subscription. After obtaining valid, anomaly-free real-time poses, the underlying execution layer cancels the temporary subscription and performs a two-dimensional pose validity check: First, the current pose is located in a passable area of ​​the map and is not in an obstacle or unknown area; second, the deviation distance between the current pose and the original planned path is less than a preset maximum deviation threshold. If both checks pass, the original path is locally stitched together based on the real-time pose, and the remaining tasks are continued by timed polling. If either check fails, the underlying execution layer replans the global path.

[0014] This application also discloses an AGV cross-scenario task scheduling and multi-state linkage system for executing the above-described methods, including a task management layer, a core scheduling layer, and a bottom execution layer; The task management layer is used to receive task sequences through message queue telemetry transmission protocol, sort them by sequence field, publish the first target point as the initial pose and activate navigation, synchronously trigger the core scheduling layer to perform timed polling, and is responsible for task access, task sorting, task data distribution and task anomaly reporting. The core scheduling layer integrates behavior ID routing scheduling, location... Figure 1 Consistency verification, elevator door sliding window confidence assessment, elevator interaction three-node closed loop, multi-scenario PID parameter switching, task continuation recovery, are used to receive the sorted task sequence issued by the task management layer, split the corresponding actions of each target point and issue various control instructions to the underlying execution layer, and receive the status data fed back by the underlying execution layer to complete task flow, abnormal rollback, map verification, and elevator interaction management. The underlying execution layer is used to receive all control commands issued by the core scheduling layer and complete the corresponding motion, positioning, map loading, and elevator interaction operations, and to feed back pose data, action completion signals, elevator status, sensor signals, and task abnormality information to the core scheduling layer.

[0015] Furthermore, the underlying execution layer is internally configured with a PID controller, a navigation module, a map management module, an elevator interface module, and a perception and positioning module. The PID controller is used to store PID parameters for each working condition, receive the mode number issued by the core scheduling layer to output speed control commands, calculate the error between the current actual posture of the AGV and the preset target posture, and feed back the action completion signal to the core scheduling layer after the error converges to the tolerance range. The navigation module is used to receive navigation target instructions and perform navigation, local path stitching, and global path replanning operations. The map management module is used to receive map update instructions, asynchronously load and switch maps of different floors, and report an exception to the core scheduling layer when map loading fails. The elevator interface module is used to collect elevator positioning and door opening / closing status data and upload them to the core dispatch layer, and to receive elevator interaction control commands to execute elevator call actions. The perception and positioning module is used to collect real-time pose data of the AGV. It performs sliding window confidence assessment on the front and rear doors of the elevator through dedicated topics, and automatically cancels the temporary subscription after obtaining the valid pose.

[0016] Due to the adoption of the above technical solution, this application has the following advantages: (1) Improve execution accuracy in multiple scenarios. Mechanism 1 switches the underlying PID control mode according to the task stage to ensure control accuracy under different operating conditions.

[0017] (2) Reduce system coupling. Mechanism 2 encapsulates specific actions as local callbacks. When adding new scenarios, only the corresponding behavior ID needs to be extended, which avoids the problems of excessive coupling and difficulty in expansion of traditional state machines.

[0018] (3) Improve the security of cross-floor positioning. Mechanism 3 executes the location check before switching between floors and target points. Figure 1 Consistency verification ensures that the positioning coordinate system matches the actual floor, reducing the risk of navigation errors across floors.

[0019] (4) Improve anti-interference capability. Mechanism 4 adopts the "sliding window + confidence attenuation" mechanism to independently filter the front and back door states, thereby enhancing the stability of door state recognition in high dust and strong interference environments.

[0020] (5) Improve elevator entry and exit safety. Mechanism 5 reduces the risk of elevator docking failure and system deadlock through a three-stage closed-loop process, a timeout timer, and an abnormal rollback strategy.

[0021] (6) Improve operational efficiency. Mechanism 6 supports resuming execution from the current position after manual intervention without having to redistribute the entire task chain. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments recorded in the embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings.

[0023] Figure 1 This is a schematic diagram of the behavior ID routing and scheduling process according to an embodiment of this application; Figure 2 This is a schematic diagram of the elevator interaction process according to an embodiment of this application; Figure 3 This is a schematic diagram of the task continuation and recovery process according to an embodiment of this application.

[0024] Figure 4 This is a block diagram of an AGV cross-scenario task scheduling and multi-state linkage system according to an embodiment of this application. Detailed Implementation

[0025] The present application will be further described in conjunction with the accompanying drawings and embodiments. The described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments obtained by those skilled in the art should fall within the protection scope of the embodiments of the present application.

[0026] This application provides an embodiment of an AGV cross-scenario task scheduling and multi-state linkage method, which includes: Step 1: The task management layer receives the task sequence carrying the sequence field through the Message Queue Telemetry Transport Protocol (MQTT), sorts all target points according to the sequence field of each task point, publishes the first target point as the initial pose and activates navigation, and starts the timed polling executed by the core scheduling layer at the same time; the sorted task sequence is transmitted to the core scheduling layer as the basic data for scheduling processing. Step 2: The core scheduling layer operates on a behavior ID routing and scheduling framework that draws on the design principles of behavior trees but does not call the standard behavior tree library. It relies on periodic polling to continuously monitor the navigation execution progress. After navigation is completed, it reads the behavior identifier bound to the current target point and sends out the corresponding action package. The underlying execution layer receives the action package and executes the corresponding transportation operation, sending back three types of status data to the core scheduling layer: in progress, completed, and execution error. Based on the returned status data, the core scheduling layer performs error retry or task rollback operations for the three scenarios of loading / unloading, elevator entry / exit, and navigation, respectively. The action completion signal fed back by the underlying execution layer serves as the basis for the core scheduling layer to advance the task flow. Step 3: Before performing the elevator up / down floor switching operation and before parsing and sending the next navigation target point, the core scheduling layer executes the following: Figure 1 The consistency check process compares the map identifiers of the currently loaded map and the target scene map. If the identifiers do not match, navigation commands are paused, and a map update command is sent to the underlying execution layer. The underlying execution layer asynchronously loads the target map; if map loading fails, all tasks are paused and an exception is reported. After the elevator arrives at the target floor, the core scheduling layer executes the same process. Figure 1 The consistency verification process determines whether to continue sending subsequent navigation target points based on the map identifier matching results. Step 4: The bottom-level execution layer collects sensor data from the front and rear doors of the elevator through dedicated topics. Independent sliding windows are set up for each door to perform confidence assessment. When the sensor recognizes an opening signal, the confidence level built into the window increases by a fixed step. When interference signals such as dust obstruction, personnel movement, or momentary door closure occur, the confidence level decreases by a fixed step. Only when the confidence level reaches a preset threshold is the latched opening state output. The latched opening state is a prerequisite for entering the elevator after completing precise door area docking. When switching task execution nodes, all confidence levels in the sliding window are automatically cleared. If the waiting time for opening exceeds a preset threshold, an elevator door malfunction is determined, and task rollback is executed. Step 5: The core scheduling layer executes the elevator interaction three-node closed-loop status confirmation process. When executing this process, the latched door opening status is read. First, the operation of calling the elevator and waiting for the elevator to arrive is executed. If the elevator call wait timeout or an abnormality is detected, the standby initial state before the elevator call is returned. The three process nodes with a sequential order are executed in sequence. After the elevator entry operation is completed, the navigation coordinates are paused and the AGV is kept stationary while waiting for the elevator to arrive at the target floor. Step 6: The core scheduling layer issues a mode number to the lower execution layer based on the current execution stage of the task. The lower execution layer stores PID parameters for various working conditions, including corridor driving, target tracking, and elevator entry. The PID parameters include target pose, error tolerance, and driving speed limit. The PID parameters for different working conditions are isolated from each other. The lower execution layer calculates the error between the AGV's current actual pose and the preset target pose in real time. After the error converges to the corresponding tolerance range, the lower execution layer sends an action completion signal back to the core scheduling layer. The core scheduling layer uses this completion signal to advance the task flow within the behavior ID routing scheduling framework. S7. Upon receiving a navigation cancellation command, the underlying execution layer terminates navigation operation, the core scheduling layer disables all timed polling, clears the execution status flags, and fully preserves the original task target point sequence and planned path. Upon receiving a navigation continuation command, the underlying execution layer uses a temporary subscription method to collect real-time poses, and cancels the temporary subscription after obtaining valid real-time poses. The real-time pose data output by the underlying execution layer serves as input data for pose validity verification during task continuation and recovery. Pose validity is verified from two dimensions: grid traversability and path offset threshold. If both verifications pass, the original path is locally stitched together to continue executing the remaining tasks. If either verification fails, the underlying execution layer replans the global path.

[0027] Optionally, the execution process of the behavior ID routing scheduling framework includes: After navigation is completed by periodic polling, the behavior identifier bound to the current target point is read. The behavior identifier is divided into five categories: loading, unloading, going up the elevator, going down the elevator, and rotating in place. There is a blank behavior identifier when no specific transportation action is specified. The core scheduling layer sends the corresponding action package to the underlying execution layer based on the behavior identifier. The underlying execution layer sends back three execution statuses to the core scheduling layer: execution in progress, execution completed, and execution error. The method of performing abnormal retry or task rollback operations based on status data for three scenarios—loading and unloading, elevator entry and exit, and navigation—includes: In loading and unloading scenarios, repeated retries are performed. Once the number of retries exceeds the maximum limit, an exception is reported and the current target point is skipped. In elevator entry and exit scenarios, the system returns to the initial standby state before the elevator call and re-executes the elevator call and alignment operations. In navigation scenarios, the current navigation target is canceled and an exception is reported, waiting for external devices to reissue task instructions. For new operation scenarios, only the corresponding behavior identifier needs to be extended, without modifying the core scheduling logic.

[0028] Optionally, the land Figure 1 The consistency verification process includes: First, determine whether the current operation belongs to one of the two scenarios that require verification. The two scenarios are before performing the elevator up / down floor switching operation and before parsing and sending the next navigation target point. If the map belongs to either of the two scenarios, the map identifier of the currently loaded map is compared with the map identifier of the target scenario map. If the identifiers do not match, all navigation target instructions are suspended, and a map update instruction is sent to the underlying execution layer. The underlying execution layer loads the target map asynchronously. If map loading fails, the task flow is continuously suspended and an exception is reported. The entire task flow is resumed after the map is fully loaded.

[0029] Optionally, the step of setting independent sliding windows for the front door and the rear door to perform confidence assessment includes: The underlying execution layer independently collects sensor data from both the front and rear doors of the elevator through a dedicated topic, and configures a dedicated sliding window for each. When the sensor recognizes an opening signal, the built-in confidence level of the corresponding sliding window increases by a fixed step. When interference signals such as dust obstruction, personnel movement, or momentary door closure occur, the built-in confidence level of the sliding window decreases by a fixed step. Only when the built-in confidence level of the sliding window reaches a preset threshold is the latched valid opening status output. When the core scheduling layer issues instructions to complete entering or exiting the elevator or switches task execution nodes, it automatically clears all confidence levels in the sliding window to eliminate residual latched data. If the waiting time for the door to open exceeds a preset threshold, the elevator door is directly determined to be faulty, and the elevator interaction task rollback process is initiated.

[0030] Optionally, the execution process of the elevator interaction three-node closed-loop status confirmation process includes: First, the system calls the elevator and waits for it to arrive. If the waiting time exceeds a threshold or an anomaly is detected, it immediately returns to the initial standby state before the elevator call. Once the elevator is in place and there are no anomalies, three sequential process nodes are executed: The first process node is the AGV's coarse alignment of the passageway to the elevator perimeter. The navigation module of the bottom execution layer navigates to the elevator perimeter and switches the corresponding PID parameters. After the bottom execution layer reports that the alignment is complete, it proceeds to the second process node. If the execution times out, it returns to the initial standby state before the elevator call. The second process node is the AGV's fine docking at the elevator door area. The bottom execution layer performs precise alignment of the elevator feed door. After the bottom execution layer reports that the docking is complete, it proceeds to the third process node. The third process node is the AGV's entry into the elevator car. Before entering this stage, the latched door opening status must be verified. If a valid latched door opening status is not detected, entry into the elevator is prohibited. After all three process nodes are completed, the navigation coordinates are paused, and the AGV remains stationary, waiting for the elevator to arrive at the target floor.

[0031] Optionally, the sequential execution of the three process nodes with a specific order includes: The first process node is the execution stage where the AGV travels to the outside of the elevator and completes the coarse alignment of the passageway. After the bottom execution layer navigation module navigates to the outside of the elevator, it switches the corresponding PID parameters. Only after the bottom execution layer reports that the alignment is completed can it enter the second process node. If the execution times out, it returns to the initial standby state before the elevator call. The second process node is the execution stage where the AGV arrives at the elevator door area and completes the fine docking. The bottom execution layer performs the precise alignment operation of the elevator feed door. After the bottom execution layer reports that the docking is completed, it enters the third process node. The third process node is the execution stage where the AGV drives into the elevator car. Before entering this stage, the latched door opening status must be verified. If a valid latched door opening status is not detected, the AGV is prohibited from entering the elevator.

[0032] Optionally, the step of sending the PID mode number to the underlying execution layer includes: Based on the execution stage of the task, the core scheduling layer issues the corresponding mode number for corridor driving, target tracking, and elevator entry to the lower execution layer. After receiving the mode number, the lower execution layer calls the PID parameters configured independently for that mode, and the PID parameters for different modes are isolated from each other. The lower execution layer calculates the error between the AGV's current actual pose and the preset target pose in real time. After the error converges to the corresponding tolerance range, the lower execution layer sends a dedicated action completion signal to the core scheduling layer. The core scheduling layer relies on this signal to advance the task flow within the behavior ID routing scheduling framework.

[0033] Optionally, the task continuation recovery includes: Upon receiving a navigation cancellation command, the underlying execution layer terminates navigation, the core scheduling layer disables all timed polling, clears status flags, and fully preserves the original task target point sequence and planned path count. Upon receiving a continue navigation command, the underlying execution layer continuously collects real-time pose data via temporary subscription. After obtaining valid, anomaly-free real-time poses, the underlying execution layer cancels the temporary subscription and performs a two-dimensional pose validity check: First, the current pose is located in a passable area of ​​the map and is not in an obstacle or unknown area; second, the deviation distance between the current pose and the original planned path is less than a preset maximum deviation threshold. If both checks pass, the original path is locally stitched together based on the real-time pose, and the remaining tasks are continued by timed polling. If either check fails, the underlying execution layer replans the global path.

[0034] See Figure 4 This application also provides an embodiment of an AGV cross-scenario task scheduling and multi-state linkage system for performing the methods described in the above embodiments, including a task management layer, a core scheduling layer, and a bottom execution layer; The task management layer is used to receive task sequences via the Message Queue Telemetry Transmission Protocol (MQTT), sort them by the order field, publish the first target point as the initial pose and activate navigation, and synchronously trigger the core scheduling layer to perform timed polling. It is responsible for task access, task sorting, task data distribution and task exception reporting. The core scheduling layer integrates behavior ID routing scheduling, location... Figure 1 Consistency verification, elevator door sliding window confidence assessment, elevator interaction three-node closed loop, multi-scenario PID parameter switching, task continuation recovery, receiving sorted task sequences issued by the task management layer, splitting the corresponding actions of each target point and issuing various control instructions to the underlying execution layer, receiving status data fed back by the underlying execution layer to complete task flow, abnormal rollback, map verification, and elevator interaction management. The underlying execution layer establishes a communication connection with the core scheduling layer, receives all control commands issued by the core scheduling layer and completes the corresponding motion, positioning, map loading, and elevator interaction operations, and feeds back pose data, action completion signals, elevator status, sensor signals, and task abnormal information to the core scheduling layer. The task management layer, core scheduling layer, and underlying execution layer transmit messages bidirectionally through dedicated topics. The output data of each layer serves as the basis for the execution actions of another layer. The three-layer architecture works together to complete the material transportation scheduling of AGVs across floors and in multiple scenarios.

[0035] Optionally, the underlying execution layer is internally configured with a PID controller, a navigation module, a map management module, an elevator interface module, and a perception and positioning module. The PID controller is used to store PID parameters for each working condition, receive the mode number issued by the core scheduling layer to output speed control commands, calculate the error between the current actual posture of the AGV and the preset target posture, and feed back the action completion signal to the core scheduling layer after the error converges to the tolerance range. The navigation module is used to receive navigation target instructions and perform basic navigation, local path stitching, and global path replanning operations. The map management module is used to receive map update instructions, asynchronously load and switch maps of different floors, and report exceptions to the core scheduling layer when map loading fails. The elevator interface module is used to collect elevator positioning and door opening / closing status data and upload them to the core dispatch layer, and to receive elevator interaction control commands to execute elevator call actions. The perception and positioning module is used to collect real-time pose data of the AGV. It performs sliding window confidence assessments on the front and rear doors of the elevator independently through dedicated topics, and automatically cancels the temporary subscription after obtaining the valid pose.

[0036] For ease of understanding, this application provides a more specific embodiment: This system includes the following 6 core mechanisms, and the system architecture diagram is as follows: Figure 1 As shown.

[0037] 1. Mechanism 1: Switching between PID control modes in multiple scenarios The system issues commands to the underlying PID control nodes based on the current task stage, enabling dynamic switching of control modes for different operating conditions such as corridor travel, elevator docking, and shelf alignment. This switching is driven by Mechanism Two, and the two form a closed loop through signal interaction between the upper and lower layers.

[0038] (1) Trigger switching The system sends mode numbers to the underlying PID control nodes via dedicated topics (e.g., mode 1 corresponds to corridor driving, mode 2 corresponds to target tracking, and mode 3 corresponds to elevator entry). Upon receiving the mode switching command, the control node activates the corresponding target pose parameters and tolerance range, and outputs speed control commands.

[0039] (2) Status feedback Once the error converges to the tolerance range, the underlying PID control node publishes the current mode action completion signal to the upper-level scheduling system through a dedicated topic, which triggers the next stage of task flow.

[0040] (3) Pattern isolation The PID parameters (target pose, tolerance, speed limit) for different scenarios are configured independently. The upper-level scheduling system is only responsible for issuing mode numbers and receiving feedback upon completion, thereby decoupling the scheduling logic from the control algorithm.

[0041] 2. Mechanism Two: Behavior ID Routing and Scheduling See Figure 1 This system draws inspiration from behavior tree design principles, employing a behavior ID-based task scheduling framework instead of directly using a standard behavior tree library. Tasks are broken down into sequences of target points labeled with behavioral attributes, and state transitions are driven by periodic polling, effectively decoupling the scheduling logic from the underlying execution. The system supports adding new scenarios by extending behavior IDs without modifying the core scheduling logic.

[0042] (1) Task sorting and assignment After receiving the task sequence from MQTT, the system sorts the tasks according to the order field of each task point, then publishes the first target point as the initial pose and activates navigation.

[0043] (2) Behavioral routing branch When the feedback status is "Navigation successful", the system issues the corresponding action package based on the behavior attributes of the current target point, such as: loading (ID 1), unloading (ID 2), going up the elevator (ID 3), going down the elevator (ID 4), or rotating in place (ID 5). (3) Status feedback The underlying execution module returns status information to the upper-level scheduling framework, such as: in execution, execution completed, execution exception, etc. The upper layer triggers the next stage task switching based on the status information.

[0044] (4) Abnormal rollback and retry If the execution module returns an exception or times out, the system will handle it according to the current scenario type: Loading and unloading: The system re-executes the current action flow. If the maximum number of retries is exceeded, the task is issued as an exception and the current target point is skipped. Elevator entry and exit: The system returns to the starting point of this elevator task and re-triggers the elevator call and alignment process; Navigation: The system cancels the current navigation target and reports a task error, waiting for external instructions to be reissued.

[0045] 3. Mechanism Three: Land Figure 1 Consistency check and update To prevent location errors caused by map asynchrony, the system incorporates map comparison logic into the task workflow: (1) Pre-verification Floor switching and navigation target point distribution are two types of critical operations that may cause inconsistencies in the coordinate system. Therefore, before executing "up / down elevator" and "parse the next target point", the system will call a verification function to compare the current map identifier with the target map identifier.

[0046] (2) Map update If the identifiers do not match, the system will pause the issuance of the next navigation point and issue a map update command via a dedicated topic.

[0047] (3) Cooperative handover Upon receiving the update command, the map management module asynchronously loads the new map. If loading fails, the system remains paused and reports an error. The task flow resumes once the map is ready, thus ensuring that the AGV uses the correct map and coordinate system.

[0048] 4. Mechanism Four: Elevator Door Status Confirmation To address the issue of momentary false alarms from sensors caused by high-dust environments, the system introduces independent state confidence assessment mechanisms for both the front and rear doors of the elevator. (1) Confidence accumulation The system subscribes to elevator status topics. When an open door is detected, the confidence level in the sliding window increases by a set step size.

[0049] (2) Confidence decay When abnormal obstruction, momentary door closure, or interference signal is detected, the confidence level in the sliding window decreases by a set step size.

[0050] (3) Steady-state confirmation The system only outputs the status and triggers the subsequent elevator entry and exit process when the confidence level of the status within the sliding window reaches the set threshold.

[0051] This latch state is automatically reset when the upper-level scheduling framework issues a task switching instruction (i.e., the elevator entry / exit action is completed and the task flows to the next node), ensuring that the next elevator entry / exit process is re-evaluated in the initial state and avoiding false triggering caused by latch residue.

[0052] Note that the valid door-opening latch state of Mechanism 4 is one of the prerequisites for node flow in Mechanism 5. Before node 2 flows to node 3 in Mechanism 5, it will actively poll the door state in Mechanism 4: When the door in the corresponding direction is in a valid open latch state, the system allows the node to flow and issues an entry command; if the waiting time for opening exceeds the limit, the elevator door is determined to be abnormal, the system terminates the current node and triggers the abnormal rollback process.

[0053] 5. Mechanism Five: Confirmation of Elevator Interaction Closed-Loop Status See Figure 2 In the high-risk process of entering and exiting elevators, the system utilizes the completion feedback of the underlying PID controller nodes to construct a three-stage status confirmation process: (1) Node 1 (corridor coarse alignment) After navigating to the outside of the elevator, the system triggers the corridor alignment control mode and waits for a feedback signal: If the underlying feedback is "alignment of aisle completed" (state 1), the system issues a switching instruction and transitions to state 2; If the timeout occurs, the system will suspend the current elevator task and return to the initial state before the elevator call, and retry according to the abnormal rollback strategy of Mechanism 2.

[0054] (2) Node 2 (Gate Zone Fine-tuning) The system performs the docking operation and waits for a feedback signal: If the underlying feedback is "Alignment of elevator feed door completed" (state 2), the system issues a switching command and transitions to state 3; If the timeout occurs, the system will suspend the current elevator task and return to the initial state to retry.

[0055] (3) Node 3 (Enter the elevator) Perform the action of entering the elevator and wait for feedback signal: If the underlying system reports "Elevator entry complete" (state 3), the system pauses issuing subsequent coordinates, keeping the AGV in its current position and awaiting further scheduling until the elevator reaches the target floor. While passing through the ground... Figure 1 After consistency verification (mechanism 3), subsequent task flow is triggered.

[0056] If the timeout occurs, the system will suspend the current elevator task and return to the initial state to retry.

[0057] 6. Mechanism Six: Task Continuation and Recovery See Figure 3 To address manual intervention or emergency obstacle avoidance, the system provides a comprehensive navigation cancellation and task continuation mechanism: (1) Interruption cleanup Upon receiving a navigation cancellation command, the system immediately terminates navigation, cancels all polling timers, and clears the execution status flag. The original sequence of mission objective points and the number of planned paths are retained for subsequent recovery.

[0058] (2) Real-time pose acquisition When the system receives a command to continue navigation, it temporarily subscribes to the positioning module to obtain the current real-time pose information; after obtaining valid pose data, it cancels the temporary subscription.

[0059] (3) Dynamic coverage and continuity The current real-time pose is used as a candidate starting point for recovery, and pose validity is verified in conjunction with the original planned path. Validity verification includes the following two dimensions: 1) The grid cell where the current pose is located is a passable area in the current map and is not in an obstacle or unknown area; 2) The deviation between the current pose and the original planned path does not exceed the set maximum deviation threshold.

[0060] If both conditions are met, the current pose is determined to be valid, and local path stitching can be performed; otherwise, the current pose is determined to be invalid, and global path replanning is triggered.

[0061] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application and not to limit them. Although this application has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of this application. Any modifications or equivalent substitutions that do not depart from the spirit and scope of this application should be covered within the protection scope of the claims of this application.

Claims

1. A method for cross-scenario task scheduling and multi-state linkage of AGVs, characterized in that, include: Step 1: The task management layer receives the task sequence carrying the sequence field through the message queue telemetry transmission protocol, sorts all target points according to the sequence field of each task point, publishes the first target point as the initial pose and activates navigation. Step 2: The core scheduling layer relies on periodic polling to detect the navigation execution progress. After the navigation is completed, it reads the behavior identifier bound to the current target point and sends out the corresponding action package. The underlying execution layer receives the action package and executes the corresponding transportation operation, and sends back the status data to the core scheduling layer. The core scheduling layer performs abnormal retry or task rollback operations for the three scenarios of loading and unloading, elevator entry and exit, and navigation, respectively, based on the status data. Step 3: The core scheduling layer executes the map consistency verification process. After the elevator arrives at the target floor, the core scheduling layer executes the same map consistency verification process. The verification result of the map consistency verification process determines whether to continue to send subsequent navigation target points. Step 4: The bottom execution layer collects sensor data from the front and rear doors of the elevator, and sets up sliding windows for confidence assessment for the front and rear doors respectively. When the sensor recognizes the door opening signal, the confidence built into the window increases by a fixed step size, and when interference signals appear, the confidence decreases by a fixed step size. The latched door opening status is output only when the confidence reaches the preset threshold. Step 5: The core scheduling layer executes the elevator interaction three-node closed-loop status confirmation process. Based on the current execution stage of the task, it issues a mode number to the lower execution layer. The lower execution layer stores the PID parameters of various working conditions, calculates the error between the AGV's current actual pose and the preset target pose. After the error converges to the corresponding tolerance range, the lower execution layer sends an action completion signal back to the core scheduling layer. Step 6: Upon receiving a navigation cancellation command, the underlying execution layer terminates navigation operation, the core scheduling layer disables all timed polling, clears status flags, and fully preserves the original task target point sequence and planned path; upon receiving a navigation continuation command, the underlying execution layer uses a temporary subscription method to collect real-time poses, and cancels the temporary subscription after obtaining valid real-time poses; the real-time pose data output by the underlying execution layer serves as input data for pose validity verification during task continuation and recovery.

2. The AGV cross-scenario task scheduling and multi-state linkage method according to claim 1, characterized in that, The execution process of the behavior ID routing scheduling framework includes: After navigation is completed by periodic polling, the behavior identifier bound to the current target point is read. The behavior identifier is divided into five categories: loading, unloading, going up the elevator, going down the elevator, and rotating in place. There is a blank behavior identifier when no specific transportation action is specified. The core scheduling layer sends the corresponding action package to the underlying execution layer based on the behavior identifier. The underlying execution layer sends back three execution statuses to the core scheduling layer: execution in progress, execution completed, and execution error. The method of performing abnormal retry or task rollback operations based on status data for three scenarios—loading and unloading, elevator entry and exit, and navigation—includes: In loading and unloading scenarios, repeated retries are performed. Once the number of retries exceeds the maximum limit, an exception is reported and the current target point is skipped. In elevator entry and exit scenarios, the system returns to the initial standby state before the elevator call and re-executes the elevator call and alignment operations. In navigation scenarios, the current navigation target is canceled and an exception is reported, waiting for external devices to reissue task instructions. When adding new operation scenarios, only the corresponding behavior identifier needs to be extended, without modifying the core scheduling logic.

3. The AGV cross-scenario task scheduling and multi-state linkage method according to claim 1, characterized in that, The map consistency verification process includes: First, determine whether the current operation belongs to one of the two scenarios that require verification. The two scenarios are before performing the elevator up / down floor switching operation and before parsing and sending the next navigation target point. If the map belongs to either of the two scenarios, the map identifier of the currently loaded map is compared with the map identifier of the target scenario map. If the identifiers do not match, all navigation target instructions are suspended, and a map update instruction is sent to the underlying execution layer. The underlying execution layer loads the target map asynchronously. If map loading fails, the task flow is continuously suspended and an exception is reported. The entire task flow is resumed after the map is fully loaded.

4. The AGV cross-scenario task scheduling and multi-state linkage method according to claim 1, characterized in that, The method of setting up independent sliding windows for the front door and the rear door to perform confidence assessment includes: The underlying execution layer independently collects sensor data from both the front and rear doors of the elevator through a dedicated topic, and configures a dedicated sliding window for each. When the sensor recognizes an opening signal, the built-in confidence level of the corresponding sliding window increases by a fixed step. When interference signals occur, the built-in confidence level of the sliding window decreases by a fixed step. Interference types include dust obstruction, personnel movement, and momentary door closing. Only when the built-in confidence level of the sliding window reaches a preset threshold is the latched valid opening status output. When the core scheduling layer issues elevator entry / exit completion instructions or switches task execution nodes, it automatically clears all confidence levels in the sliding window to eliminate latched data residue. If the waiting time for the door to open exceeds a preset threshold, the elevator door is directly determined to be faulty, and the elevator interaction task rollback process is initiated.

5. The AGV cross-scenario task scheduling and multi-state linkage method according to claim 1, characterized in that, The execution process of the elevator interaction three-node closed-loop status confirmation process includes: First, the system calls the elevator and waits for it to arrive. If the waiting time exceeds a threshold or an anomaly is detected, it immediately returns to the initial standby state before the elevator call. Once the elevator is in place and there are no anomalies, three sequential process nodes are executed: The first process node is the AGV's coarse alignment of the passageway to the elevator perimeter. The navigation module of the bottom execution layer navigates to the elevator perimeter and switches the corresponding PID parameters. After the bottom execution layer reports that the alignment is complete, it proceeds to the second process node. If the execution times out, it returns to the initial standby state before the elevator call. The second process node is the AGV's fine docking at the elevator door area. The bottom execution layer performs precise alignment of the elevator feed door. After the bottom execution layer reports that the docking is complete, it proceeds to the third process node. The third process node is the AGV's entry into the elevator car. Before entering this stage, the latched door opening status must be verified. If a valid latched door opening status is not detected, entry into the elevator is prohibited. After all three process nodes are completed, the navigation coordinates are paused, and the AGV remains stationary, waiting for the elevator to arrive at the target floor.

6. The AGV cross-scenario task scheduling and multi-state linkage method according to claim 1, characterized in that, The sequential execution of the three process nodes, which have a specific order, includes: The first process node is the execution stage where the AGV travels to the outside of the elevator and completes the coarse alignment of the passageway. After the bottom execution layer navigation module navigates to the outside of the elevator, it switches the corresponding PID parameters. Only after the bottom execution layer reports that the alignment is completed can it enter the second process node. If the execution times out, it returns to the initial standby state before the elevator call. The second process node is the execution stage where the AGV arrives at the elevator door area and completes the fine docking. The bottom execution layer performs the precise alignment operation of the elevator feed door. After the bottom execution layer reports that the docking is completed, it enters the third process node. The third process node is the execution stage where the AGV drives into the elevator car. Before entering this stage, the latched door opening status must be verified. If a valid latched door opening status is not detected, the AGV is prohibited from entering the elevator.

7. The AGV cross-scenario task scheduling and multi-state linkage method according to claim 1, characterized in that, The step of sending the PID mode number to the underlying execution layer includes: Based on the execution stage of the task, the core scheduling layer issues the corresponding mode number for corridor driving, target tracking, and elevator entry to the lower execution layer. After receiving the mode number, the lower execution layer calls the PID parameters configured independently for that mode, and the PID parameters for different modes are isolated from each other. The lower execution layer calculates the error between the AGV's current actual pose and the preset target pose in real time. After the error converges to the corresponding tolerance range, the lower execution layer sends a dedicated action completion signal to the core scheduling layer. The core scheduling layer relies on this signal to advance the task flow within the behavior ID routing scheduling framework.

8. The AGV cross-scenario task scheduling and multi-state linkage method according to claim 1, characterized in that, The task continuation and recovery includes: Upon receiving a navigation cancellation command, the underlying execution layer terminates navigation, the core scheduling layer disables all timed polling, clears status flags, and fully preserves the original task target point sequence and planned path count. Upon receiving a continue navigation command, the underlying execution layer continuously collects real-time pose data via temporary subscription. After obtaining valid, anomaly-free real-time poses, the underlying execution layer cancels the temporary subscription and performs a two-dimensional pose validity check: First, the current pose is located in a passable area of ​​the map and is not in an obstacle or unknown area; second, the deviation distance between the current pose and the original planned path is less than a preset maximum deviation threshold. If both checks pass, the original path is locally stitched together based on the real-time pose, and the remaining tasks are continued by timed polling. If either check fails, the underlying execution layer replans the global path.

9. An AGV cross-scenario task scheduling and multi-state linkage system, characterized in that, The method for performing any one of claims 1-8 includes a task management layer, a core scheduling layer, and a bottom execution layer; The task management layer is used to receive task sequences through message queue telemetry transmission protocol, sort them by sequence field, publish the first target point as the initial pose and activate navigation, synchronously trigger the core scheduling layer to perform timed polling, and is responsible for task access, task sorting, task data distribution and task anomaly reporting. The core scheduling layer integrates behavior ID routing scheduling, map consistency verification, elevator door sliding window confidence assessment, elevator interaction three-node closed loop, multi-scenario PID parameter switching, and task continuation recovery. It is used to receive the sorted task sequence issued by the task management layer, split the corresponding actions of each target point and issue various control instructions to the underlying execution layer, and receive the status data fed back by the underlying execution layer to complete task flow, abnormal rollback, map verification, and elevator interaction management. The underlying execution layer is used to receive all control commands issued by the core scheduling layer and complete the corresponding motion, positioning, map loading, and elevator interaction operations, and to feed back pose data, action completion signals, elevator status, sensor signals, and task abnormality information to the core scheduling layer.

10. The AGV cross-scenario task scheduling and multi-state linkage system according to claim 9, characterized in that, The underlying execution layer is internally configured with a PID controller, a navigation module, a map management module, an elevator interface module, and a perception and positioning module. The PID controller is used to store PID parameters for each working condition, receive the mode number issued by the core scheduling layer to output speed control commands, calculate the error between the current actual posture of the AGV and the preset target posture, and feed back the action completion signal to the core scheduling layer after the error converges to the tolerance range. The navigation module is used to receive navigation target instructions and perform navigation, local path stitching, and global path replanning operations. The map management module is used to receive map update instructions, asynchronously load and switch maps of different floors, and report an exception to the core scheduling layer when map loading fails. The elevator interface module is used to collect elevator positioning and door opening / closing status data and upload them to the core dispatch layer, and to receive elevator interaction control commands to execute elevator call actions. The perception and positioning module is used to collect real-time pose data of the AGV. It performs sliding window confidence assessment on the front and rear doors of the elevator through dedicated topics, and automatically cancels the temporary subscription after obtaining the valid pose.

Images