An agv vehicle obstacle closed-loop handling system and method

Abstract

The application discloses an AGV vehicle obstacle closed-loop treatment method and system. The system generates a task semantic configuration according to the region attribute passed by the AGV task path; constructs an obstacle event object with a number, evidence, credibility and life cycle state for a candidate obstacle in environmental perception data; selects a deceleration, detour, parking or manual takeover strategy based on the perception health degree and the credibility of the obstacle event object; after detouring, verifies whether the AGV restores global path tracking based on the regression credentials containing the reference path segment, the regression region, the allowed deviation, the speed range, the perception health degree and the event state requirement. The application can distinguish between temporary obstacles and long-term map changes, and make the safety treatment and path recovery after obstacle avoidance in different regions have traceable constraint conditions.

Description

Technical Field

[0001] This application relates to the field of mobile robot navigation and safety technology, and in particular to an AGV vehicle obstacle closed-loop handling system and method. Background Technology

[0002] Automated Guided Vehicles (AGVs) typically perform transportation tasks from a starting point to a destination based on pre-established maps and routes. In warehousing, flexible manufacturing, park delivery, training demonstrations, and personnel collaboration, AGVs need to acquire information about surrounding obstacles and take measures such as slowing down, avoiding them, or stopping during operation.

[0003] Existing AGV obstacle avoidance solutions mostly rely on single-frame ranging results, fixed safety sectors, or fixed distance thresholds for processing. When abnormal perception results are generated by lidar reflection, partial obstruction, short-term pedestrian passage, parking of movable goods, or positioning drift, the system may easily treat temporary anomalies as persistent obstacles, leading to unnecessary stopping or detours; conversely, it may also fail to take timely safety actions due to a single sensor malfunction.

[0004] Existing solutions typically treat each detected obstacle point, obstacle cluster, or occupied grid as a direct processing object, lacking a mechanism to correlate information about the same candidate obstacle across multiple sampling frames, different sensor evidence, and the handling process into a unified management object. Therefore, it is difficult to continuously record the formation, confirmation, handling, and disappearance of obstacles, and it is also difficult to reference the status of the same obstacle in subsequent path recovery or multi-vehicle collaboration.

[0005] Furthermore, the processing objectives for temporary obstacles differ from those for static map changes. Directly writing short-term personnel or goods into a static map will cause the map to continuously accumulate erroneous information; if detours are only based on the location of momentary obstacles without event status management, it is also difficult to record the process of obstacle appearance, confirmation, handling, and resolution.

[0006] After an AGV bypasses an obstacle, existing solutions typically resume path tracking when the vehicle approaches the original planned path. This approach does not simultaneously verify the vehicle's heading, speed, perception availability, and whether the obstacle has left its return position. This can easily lead to sudden changes in direction during path recovery, re-entry into the obstacle's influence area, or continued autonomous driving when perception capabilities are reduced.

[0007] Furthermore, the risks differ in personnel collaboration areas, narrow passages, loading / unloading areas, and intersections. Using the same safety threshold and detour strategy across all areas could lead to frequent detours in narrow passages, or compromise operational safety in personnel collaboration areas due to overly aggressive strategies. Therefore, a closed-loop AGV handling solution is needed that can differentiate task areas, manage obstacles as events, evaluate perceived availability, and verify path recovery conditions. Summary of the Invention

[0008] This application provides an AGV vehicle obstacle closed-loop handling system and method to solve the problems of AGV obstacle misjudgment, conflict between temporary obstacles and map updates, insufficient reliability of path recovery after obstacle avoidance, and inability to distinguish safety strategies in different areas.

[0009] The technical solution of this application is as follows: On the one hand, this application provides an AGV vehicle obstacle closed-loop handling system, including: The scheduling and control platform is configured to generate corresponding task semantic configurations based on the task path from the starting point to the end point and the regional attributes of the areas traversed by the task path. A navigation sensing unit, configured to acquire information about the surrounding environment of the AGV and the vehicle's pose information; The vehicle controller is configured to: establish obstacle event objects based on the environmental information, and perform lifecycle management on the obstacle event objects in the stages of pending verification, confirmed status, in progress, and pending resolution. The system performs a reliability evaluation on the obstacle event object to assess the authenticity of the obstacle, and evaluates the perception health of the navigation perception unit to assess the autonomous control capability of the AGV. Based on the task semantic configuration, reliability evaluation results, perception health evaluation results, and life cycle status, the system generates operation control commands for deceleration, partial detour, or stopping, or outputs a manual takeover request.

[0010] Furthermore, the task semantic configuration includes at least two or more of the following: area identifier, maximum permissible speed, safety margin, automatic detour permission, parking priority, manual takeover authority, and event reporting level; the area attributes include at least one of the following: personnel collaboration area, loading and unloading area, narrow passage, intersection, observation and display area, and emergency evacuation area.

[0011] Furthermore, the obstacle event object, as a dynamic obstacle state object for continuous state representation and management of dynamic obstacles, includes event number, candidate obstacle location, formation time, evidence data, credibility, handling strategy, life cycle state, handling process record, and obstacle event digital fingerprint; the handling process record is at least associated with life cycle state transition information, credibility evaluation results, perceived health evaluation results, operation control instructions, and regression credential verification results to form a handling evidence chain for the corresponding obstacle event object; the life cycle state includes at least a pending verification state, a confirmed state, a handling state, and a pending resolution state.

[0012] Furthermore, the credible evaluation result is obtained from at least three of the following: cross-frame persistence, contour consistency, positioning quality, multi-source evidence consistency, and regional risk. When the credible evaluation result is lower than a first threshold, the on-board controller controls the AGV to decelerate and increase the retest frequency. When the credible evaluation result reaches a second threshold and meets the collision risk conditions, the on-board controller controls the AGV to perform local detour, stop and wait, or report anomalies, wherein the second threshold is greater than the first threshold.

[0013] Furthermore, the obstacle event object in the pending verification state is only used for immediate security control and does not update the static map; when the same obstacle event object meets the preset confirmation conditions in multiple independent tasks and is confirmed by manual verification or automatic verification as a long-term existing object, the scheduling control platform generates a static map patch for the obstacle event object.

[0014] Furthermore, the perceived health evaluation is based on at least three of the following: lidar point cloud efficiency, near-range ranging efficiency, fusion positioning residual, visual visibility, and communication link quality; the vehicle controller switches between normal mode, degraded mode, and protection mode according to the perceived health. The degraded mode limits the maximum speed, expands the safety margin, and increases the retest frequency. The protection mode prohibits automatic detours and controls the AGV to stop and report.

[0015] Furthermore, the regression credentials include at least a global reference path segment identifier, a regression region, allowable lateral deviation, allowable heading deviation, allowable speed range, minimum perception health, obstacle event status requirements, and validity period; the vehicle controller resumes global reference path tracking when the AGV is located in the regression region, the lateral deviation and heading deviation meet the allowable range, the current speed is within the allowable speed range, the perception health is not lower than the minimum perception health, and the obstacle event object meets the obstacle event status requirements.

[0016] Furthermore, the task interaction terminal constructs a digital twin monitoring interface, which displays the AGV's real-time pose, global reference path, local detour path, obstacle event number, obstacle event lifecycle status, perception health, and safety mode; the manual handling instructions issued by the digital twin monitoring interface carry the event number and validity period.

[0017] Furthermore, the scheduling and control platform includes an obstacle event history record management unit, which is used to store the event number, task number, candidate obstacle location, evidence data, lifecycle status, handling strategy, regression credential and handling result of the obstacle event object, and to query and trace according to the event number or task number.

[0018] Furthermore, the scheduling and control platform is used to generate the passage order or passage restriction of key areas when multiple AGVs are performing tasks, based on the spatial range, life cycle status and task semantic configuration of the obstacle event objects corresponding to the multiple AGVs; before entering the key area, the multiple AGVs respectively verify the corresponding passage conditions, perception health and obstacle event object status.

[0019] Furthermore, the obstacle event digital fingerprint includes at least spatial location, time information, contour features, motion trend, sensor source, and historical occurrence count; the scheduling control platform or vehicle controller uses the obstacle event digital fingerprint to perform cross-frame, cross-task, or cross-AGV association, query, and tracing of the same dynamic obstacle state object.

[0020] Furthermore, the vehicle controller forms a dual-confidence fusion safety decision-making mechanism based on the obstacle event credibility evaluation result and the perception health evaluation result; the dual-confidence fusion safety decision-making mechanism outputs a deceleration retest strategy when the obstacle event credibility is insufficient, outputs a local detour strategy corresponding to the task semantic configuration when the obstacle event credibility meets the confirmation condition and the perception health meets the autonomous control condition, and outputs a parking protection or manual takeover request when the perception health is in protection mode.

[0021] Furthermore, the obstacle event history record management unit obtains the obstacle risk evolution prediction results for the task area or path segment based on the spatial location, time information, historical occurrence frequency, and handling results of historical obstacle events; the scheduling control platform adjusts the maximum allowable speed, safety margin, automatic detour permission, event reporting level, or safety policy in the corresponding task semantic configuration according to the obstacle risk evolution prediction results.

[0022] Furthermore, the scheduling and control platform synchronizes obstacle event sharing information to multiple AGVs based on event numbers. The obstacle event sharing information includes at least the obstacle location or spatial range, life cycle status, and risk level. Before entering the key area corresponding to the obstacle event sharing information, each AGV verifies the timeliness of the obstacle event sharing information, the passage conditions, and its own perceived health status.

[0023] On another front, this application provides a closed-loop obstacle handling method for AGV vehicles, comprising the following steps: receiving AGV tasks from start to finish; generating task semantic configuration and global reference path based on the regional attributes of the areas traversed by the task path; acquiring environmental information and vehicle pose information around the AGV, forming candidate obstacle information and establishing obstacle event objects; managing the transition of the obstacle event objects between the states of pending verification, confirmed, being handled, and awaiting resolution based on the association results in continuous sampling frames; performing a credibility evaluation on the obstacle event objects to determine whether the candidate obstacles are real obstacles, and performing a perception health evaluation on the perception links used to acquire the environmental information and vehicle pose information to determine whether the AGV has autonomous control capabilities; executing deceleration, partial detour, stopping, or manual takeover requests based on the task semantic configuration, credibility evaluation results, and perception health evaluation results; generating regression credentials before executing partial detour, and resuming global reference path tracking when the regression conditions defined by the regression credentials are met.

[0024] Furthermore, establishing an obstacle event object includes: registering the current environmental perception data with a static map, extracting candidate residual clusters that are inconsistent with the static map; associating the candidate residual clusters with event numbers, locations, contours, timestamps, and relative motion estimates; and updating the lifecycle state of the obstacle event object based on the association results in consecutive sampling frames.

[0025] Furthermore, the credibility evaluation results are obtained using a weighted evaluation method, in which each evidence item is normalized, and different weights or thresholds are assigned to different areas for regional risk; for personnel collaboration areas, the credibility evaluation threshold for automatic detour permission is reduced or the threshold for entering local detours is increased.

[0026] Furthermore, when communication is interrupted, the fusion positioning quality is lower than the preset value, the perception health is lower than the protection threshold, the regression credentials are invalid, or a qualified local detour path is not obtained within the preset time, the AGV is controlled to decelerate and stop, and the current obstacle event object and regression credentials are cached for subsequent processing.

[0027] On another front, this application provides an AGV vehicle, including a processor, a memory, a lidar, a positioning device, a drive execution unit, and a communication device. The memory stores a program, which, when executed by the processor, implements an AGV vehicle obstacle closed-loop handling method as described above.

[0028] In another aspect, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the AGV vehicle obstacle closed-loop handling method described above.

[0029] This invention does not use a single obstacle point or a cluster of instantaneous obstacles as the final management object. Instead, it constructs candidate obstacles associated in continuous sampling frames into obstacle event objects with event numbers, evidence data, credibility, and lifecycle states. Obstacle event objects switch between states of pending verification, confirmed, in progress, and pending resolution. Objects in the pending verification state only participate in immediate security control and are not directly written into the static map.

[0030] This invention evaluates obstacle event credibility and perceived health separately. Obstacle event credibility is used to determine whether a candidate obstacle is a real obstacle, and accordingly decides whether the corresponding obstacle event needs to enter the confirmation or handling stage. Perceived health is used to determine whether the AGV's perception and positioning links have the ability to continue autonomous control, and accordingly limits the handling methods that can be adopted. Together with task semantic configuration, the two constitute a dual-confidence fusion safety decision-making mechanism: obstacle event credibility does not replace perceived health, and perceived health is not used to determine whether an obstacle actually exists, but rather determines the deceleration retest, partial detour, parking protection, or manual takeover strategy based on the combined state of the two.

[0031] This invention generates regression credentials before local detours and simultaneously verifies path segments, vehicle positions, lateral deviations, heading deviations, speeds, perception health, obstacle event status, and validity periods after the detour, thereby restoring global reference path tracking when conditions are met.

[0032] Compared to obstacle avoidance schemes that rely solely on fixed safety zones, this invention associates obstacle event objects with continuous perception evidence and handling processes, allowing these objects to be referenced across frames, tasks, and vehicles as dynamic obstacle state objects. Temporary obstacles awaiting verification do not directly update the static map, thus reducing the risk of them being mistakenly written into the map. By associating event numbers and obstacle event digital fingerprints with reliable evaluation, handling strategies, and regression credentials, the obstacle discovery, confirmation, detour, and disappearance processes can be queried and traced. Regression credentials simultaneously verify position, heading, speed, perception health, and event status, improving the reliability of restoring global path tracking after obstacle avoidance. Attached Figure Description

[0033] Figure 1 This is a system architecture diagram of the AGV vehicle obstacle closed-loop handling system of the present invention, showing the relationship between the scheduling and digital twin platform, communication, vehicle controller and drive execution layer, as well as the closed-loop relationship of drive execution feedback to event history management.

[0034] Figure 2 This is a schematic diagram of the hardware structure of the AGV vehicle of the present invention.

[0035] Figure 3 This is a schematic diagram of the lifecycle of the obstacle event object in this invention.

[0036] Figure 4 This is a schematic diagram illustrating the partial bypass, expansion of the regression credential region, and regression credential verification of the present invention.

[0037] Figure 5 This is a flowchart of the autonomous operation and obstacle event handling process of the present invention.

[0038] Figure 6 This is a schematic diagram of the digital twin monitoring interface of the present invention, showing event status fields such as event number, digital fingerprint, trustworthiness, health, lifecycle, risk level, and security mode.

[0039] Figure 7 This is a diagram illustrating the security degradation relationship driven by the perceived health status of this invention. Detailed Implementation

[0040] The specific embodiments of this application are described in detail below with reference to the accompanying drawings.

[0041] Example 1: An AGV vehicle obstacle closed-loop handling system, such as Figure 1 As shown, the system includes a task interaction terminal 1, a scheduling and control platform 2, an on-board controller 3, a drive execution unit 4, a LiDAR 5, a perception reliability assessment unit 6, and a status acquisition unit 7. The task interaction terminal 1 can be an industrial control computer, a fixed large screen, a web terminal, or a mobile terminal, used to select the start point, end point, task type, and manual takeover authority. The scheduling and control platform 2 stores static maps and regional attributes, and sends task semantic configurations to the on-board controller 3 via a communication link. It receives vehicle pose, obstacle event status, handling results, and drive execution feedback; the drive execution feedback is associated with the event number and written into the event history management system. The on-board controller 3 receives information from the LiDAR 5, the perception reliability assessment unit 6, and the status acquisition unit 7, and sequentially performs obstacle event object generation and lifecycle management, obstacle event reliability assessment, perception health assessment, safety decision-making, partial detour, and return credential verification on the on-board side, and outputs operation control commands to the drive execution unit 4. This forms a closed-loop relationship of task issuance, on-board handling, execution feedback, and platform history management feedback.

[0042] The scheduling and control platform is configured to generate corresponding task semantic configurations based on the task path from the origin to the destination and the regional attributes of the areas traversed by the task path. Task semantic configurations can be generated by region or path segment. For example, personnel collaboration zones are configured with lower maximum permissible speeds, larger safety margins, and higher event reporting levels; narrow passages are configured with priority parking or restricted automatic detours; loading and unloading zones are configured to allow low-speed detours and require confirmation upon leaving the area; and intersections are configured to verify passage conditions before entry. The configurations include a task number, region identifier, version number, and validity period to prevent older configurations from continuing to apply to subsequent tasks.

[0043] like Figure 2 As shown, the AGV vehicle may include a LiDAR 11, a camera 12, a near-range ranging unit 13, an encoder and inertial measurement unit 14, a collision detection and emergency stop unit 15, an edge controller 16, and a motor driver 17. The LiDAR 11 is used to acquire environmental point clouds, and the encoder and inertial measurement unit 14 are used to provide vehicle motion information. The edge controller 16 includes a processor, memory, and a communication interface for processing information from the sensors and controlling the motor driver 17. The collision detection and emergency stop unit 15 is independent of local detour calculations and directly triggers the motor driver 17 to enter a safe stopping state in the event of an emergency.

[0044] The task semantic configuration includes at least two or more of the following: area identifier, maximum permissible speed, safety margin, automatic detour permission, parking priority, manual takeover authority, and event reporting level; the area attributes include at least one of the following: personnel collaboration area, loading and unloading area, narrow passage, intersection, observation and display area, and emergency evacuation area.

[0045] The navigation sensing unit is configured to acquire information about the AGV's surrounding environment and vehicle position.

[0046] The vehicle controller is configured to: establish obstacle event objects based on the environmental information, and perform lifecycle management on the obstacle event objects in the states of pending verification, confirmed, in progress, and pending resolution; perform a credibility evaluation of the obstacle event objects to assess their authenticity, and perform a perception health evaluation of the AGV's autonomous control capability on the navigation perception unit; and generate operation control commands for deceleration, partial detour, or stopping, or output a manual takeover request, based on the task semantic configuration, credibility evaluation results, perception health evaluation results, and lifecycle status.

[0047] The onboard controller 3 aligns the current environmental perception data to the static map coordinate system and extracts candidate residual clusters that are inconsistent with the static map. Each candidate residual cluster includes at least location, contour, number of points, distance, timestamp, and relative motion estimate. To prevent transient disturbances from directly causing map updates, the controller assigns event numbers to the candidate residual clusters and establishes obstacle event objects. These obstacle event objects serve as dynamic obstacle state objects, providing continuous state representation for the same dynamic obstacle during its formation, confirmation, handling, and resolution phases.

[0048] like Figure 3As shown, the controller first associates candidate obstacles and generates obstacle event objects, which then enter a pending verification state 21. The controller tracks the same candidate residual cluster in consecutive sampling frames and obtains a credible evaluation result using cross-frame persistence, contour consistency, positioning quality, and optional camera or near-range ranging evidence. The controller generates an obstacle event digital fingerprint for the obstacle event object, which associates at least spatial location, temporal information, contour features, motion trends, sensor source, and historical occurrence counts, and updates with the association results and lifecycle state transitions. When the credible evaluation result meets the confirmation conditions, the obstacle event object enters a confirmed state 22; when the AGV starts to decelerate, detour, stop and wait, or requests manual intervention, the obstacle event object enters a handling state 23; when the obstacle no longer occupies the corresponding area and meets the fading conditions, the obstacle event object enters a pending fading state 24 and the event is closed. Each state transition, corresponding evidence, and operation control command are recorded in the handling process record; objects in the pending verification state 21 do not update the static map, while obstacle event objects that have been confirmed over a long period through multiple tasks generate static map patches.

[0049] The obstacle event object, as a dynamic obstacle state object for continuous state representation and management of dynamic obstacles, includes event number, candidate obstacle location, formation time, evidence data, credibility, handling strategy, life cycle state, handling process record, and obstacle event digital fingerprint; the handling process record is at least associated with life cycle state transition information, credibility evaluation results, perceived health evaluation results, operation control instructions, and regression credential verification results to form a handling evidence chain for the corresponding obstacle event object; the life cycle state includes at least the pending verification state, confirmed state, handling state, and pending resolution state.

[0050] The credible evaluation result is obtained from at least three of the following: cross-frame persistence, contour consistency, positioning quality, multi-source evidence consistency, and regional risk. When the credible evaluation result is lower than a first threshold, the on-board controller controls the AGV to decelerate and increase the retest frequency. When the credible evaluation result reaches a second threshold and meets the collision risk conditions, the on-board controller controls the AGV to perform local detour, stop and wait, or report anomalies, wherein the second threshold is greater than the first threshold.

[0051] The obstacle event object in the pending verification state is only used for immediate security control and does not update the static map; when the same obstacle event object meets the preset confirmation conditions in multiple independent tasks and is confirmed by manual confirmation or automatic verification as a long-term existing object, the scheduling control platform generates a static map patch for the obstacle event object.

[0052] The perception health evaluation is based on at least three of the following: LiDAR point cloud effectiveness, near-range ranging effectiveness, fusion positioning residual, visual visibility, and communication link quality. The vehicle controller switches between normal mode, degraded mode, and protection mode according to the perception health. The degraded mode limits the maximum speed, increases the safety margin, and increases the retest frequency. The protection mode prohibits automatic detours and controls the AGV to stop and report.

[0053] The credibility evaluation result is obtained by weighting multiple normalized evidence items, including cross-frame persistence, contour consistency, positioning quality, multi-source evidence consistency, and regional risk. For personnel collaboration areas, the credibility evaluation threshold for triggering automatic detours can be increased, or automatic detours can be directly prohibited; for loading and unloading areas, local detour paths can be generated at low speeds and with a large safety margin. When the credibility evaluation result is below the first threshold, the AGV slows down and undergoes retesting; when it reaches the second threshold and there is a collision risk, the AGV detours, stops, or reports for handling, where the second threshold is greater than the first threshold.

[0054] Perception health reflects whether the current perception and positioning link is suitable for continued autonomous control. It can be obtained based on at least three of the following: lidar point cloud effectiveness, near-range ranging effectiveness, fusion positioning residual, camera visibility, and communication link quality. Obstacle event credibility is used to evaluate the existence of obstacles, while perception health is used to evaluate whether the system has the conditions for automatic handling. The two have different roles and jointly participate in strategy selection. In one implementation, the onboard controller maps the credibility level and health mode to the dual-confidence fusion safety decision result: when the credibility is insufficient, deceleration and retesting are prioritized; when the credibility meets the confirmation and the health meets the conditions for autonomous control, partial detour is allowed according to the task semantics; when the health enters the protection mode, even if the credibility is high, stopping for protection or requesting manual takeover is prioritized.

[0055] The regression credentials include at least a global reference path segment identifier, a regression region, allowable lateral deviation, allowable heading deviation, allowable speed range, minimum perception health, obstacle event status requirements, and validity period. The vehicle controller resumes global reference path tracking when the AGV is located in the regression region, the lateral deviation and heading deviation meet the allowable range, the current speed is within the allowable speed range, the perception health is not lower than the minimum perception health, and the obstacle event object meets the obstacle event status requirements.

[0056] The task interaction terminal constructs a digital twin monitoring interface, which displays the AGV's real-time pose, global reference path, local detour path, obstacle event number, obstacle event lifecycle status, perception health, and safety mode; the manual handling instructions issued by the digital twin monitoring interface carry the event number and validity period.

[0057] The scheduling and control platform includes an obstacle event history record management unit, which stores the event number, task number, candidate obstacle location, evidence data, lifecycle status, handling strategy, regression credential, and handling result of the obstacle event object, and performs querying and tracing according to the event number or task number.

[0058] The scheduling and control platform is used to generate the passage order or passage restriction of key areas when multiple AGVs are performing tasks, based on the spatial range, life cycle status and task semantic configuration of the obstacle event objects corresponding to the multiple AGVs; before entering the key area, the multiple AGVs respectively verify the corresponding passage conditions, perception health and obstacle event object status.

[0059] The obstacle event digital fingerprint includes at least spatial location, time information, contour features, motion trend, sensor source, and historical occurrence count; the scheduling control platform or vehicle controller uses the obstacle event digital fingerprint to perform cross-frame, cross-task, or cross-AGV association, query, and tracing of the same dynamic obstacle status object.

[0060] The vehicle controller forms a dual-confidence fusion safety decision mechanism based on the obstacle event credibility evaluation result and the perception health evaluation result. When the obstacle event credibility is insufficient, the dual-confidence fusion safety decision mechanism outputs a deceleration retest strategy. When the obstacle event credibility meets the confirmation condition and the perception health meets the autonomous control condition, it outputs a local detour strategy corresponding to the task semantic configuration. When the perception health is in protection mode, it outputs a parking protection or manual takeover request.

[0061] The obstacle event history record management unit obtains the obstacle risk evolution prediction results for the task area or path segment based on the spatial location, time information, historical occurrence frequency and handling results of historical obstacle events; the scheduling control platform adjusts the maximum allowable speed, safety margin, automatic detour permission, event reporting level or safety policy in the corresponding task semantic configuration according to the obstacle risk evolution prediction results.

[0062] The scheduling and control platform synchronizes obstacle event sharing information to multiple AGVs based on event numbers. The obstacle event sharing information includes at least the obstacle location or spatial range, life cycle status, and risk level. Before entering the key area corresponding to the obstacle event sharing information, each AGV verifies the timeliness of the obstacle event sharing information, the passage conditions, and its own perceived health status.

[0063] like Figure 7As shown, when the perceived health level is not lower than the normal threshold, the system is in normal mode 61; when the perceived health level is lower than the normal threshold but not lower than the protection threshold, the system enters degraded mode 62, which limits the speed, increases the safety margin, increases the retest frequency, and tightens the path recovery conditions; when the perceived health level is lower than the protection threshold, the system enters protection mode 63, which prohibits automatic detours and controls the AGV to stop and report.

[0064] like Figure 4 As shown, before selecting a local detour, the vehicle controller 3 creates a regression credential 34 and reserves an extended regression credential region near the global reference path 31. This extended regression credential region covers the spatial range used for entering, verifying, and resuming path tracking, providing margin for trajectory convergence and speed adjustment of the local detour path 33. The regression credential 34 is associated with a path segment in the global reference path 31 and includes at least position conditions, heading conditions, speed conditions, minimum perception health, obstacle event status requirements, and validity period. The position conditions include the regression region and allowed lateral deviation, and the heading conditions include allowed heading deviation. After the local detour path 33 avoids obstacle event 32, the controller does not immediately resume tracking as the vehicle approaches the global reference path 31.

[0065] The controller resumes tracking of the global reference path 31 only when the vehicle enters the return zone, the lateral deviation is no greater than the allowable lateral deviation, the heading deviation is no greater than the allowable heading deviation, the speed is within the allowable speed range, the perception health is no less than the minimum perception health, and the obstacle event has subsided or no longer occupies the return zone. If any condition is not met, local handling, replanning, or manual takeover is requested.

[0066] like Figure 5 As shown, after the AGV receives the task from the starting point to the destination, the system first performs task semantic parsing and global path planning. During operation, it continuously performs fusion perception and status acquisition, and establishes obstacle event objects after discovering candidate obstacles. When the confidence of the obstacle event is insufficient, the AGV slows down and retests, and the retest results are fed back to the fusion perception and event object update. When the perception health is insufficient, the AGV enters a stop protection state and waits for manual handling or manual takeover instructions, while caching the current obstacle event object and regression credentials for subsequent processing. When the confidence, perception health, and task semantic configuration all meet the conditions, the AGV performs a partial detour. After the detour is completed, regression credentials are verified, and global path tracking is restored when the verification is successful and the obstacle event fades or no longer occupies the regression area; otherwise, partial handling, replanning, or requesting manual takeover continues. This forms a closed loop of discovery, confirmation, handling, fading, and recovery.

[0067] Taking the AGV transporting materials from point A to point B as an example, the task path passes through the personnel collaboration area. The scheduling and control platform 2 configures this area as low speed, with a large safety margin and a high event reporting level. After the AGV enters the personnel collaboration area, the lidar 11 detects a moving target ahead. The onboard decision controller 3 obtains candidate residual clusters with similar positions and contours in multiple consecutive frames, and combines them with evidence from the camera 12 or the near-range ranging unit 13 to form an obstacle event object. If the credibility evaluation result does not reach the detour threshold, the AGV decelerates and retests; if the credibility evaluation result reaches the detour threshold and the perceived health is in normal mode, the controller generates a local detour path 33 to avoid personnel, and simultaneously generates regression evidence 34.

[0068] After the AGV bypasses the moving target, the controller checks whether the obstacle event has disappeared, whether the AGV has entered the return area, whether the vehicle's heading and speed meet the return credential 34, and whether the perception health remains normal. If all conditions are met, the AGV resumes the global reference path 31 and continues to travel towards point B; if the target still occupies the return area, the communication quality deteriorates, or the positioning quality is insufficient, the AGV maintains a low speed or stops, and reports the event number and current status to the digital twin platform 2.

[0069] In the industrial implementation, the AGV transports materials from point A in the raw material storage area to point B next to the processing equipment, passing through the pallet storage area, personnel passage area, and equipment operation area in sequence. When the LiDAR detects temporary pallet, personnel, or equipment obstruction ahead of the path, the onboard controller first groups the candidate obstacle information associated with the location in consecutive sampling frames into the same obstacle event object, rather than treating each frame's detection result as an independent obstacle. The controller obtains the obstacle event confidence level based on cross-frame persistence, contour consistency, and optional near-range ranging evidence; this confidence level is used to determine whether the temporary pallet, personnel, or equipment obstruction is a real obstacle. Simultaneously, the controller obtains perception health based on point cloud effectiveness, positioning residual, and communication quality; this perception health level is used to determine whether the AGV still possesses autonomous deceleration and detour capabilities.

[0070] When the credibility of an obstacle event meets the confirmation criteria and the perceived health level meets the requirements of normal mode, the AGV first slows down, and then selects either partial detour or stopping to wait based on the task semantic configuration corresponding to the personnel passage area or equipment operation area; a return credential is generated before partial detour. If the temporary pallet is removed, personnel leave, or equipment operation is completed, and the AGV meets the requirements for return area, heading, speed, perceived health level, and event status, the AGV resumes its original global path; if the obstacle still exists or the perceived health level decreases, it remains stopped, replans, or requests manual intervention. Thus, the temporary pallet, personnel, and equipment occupation are all handled in a closed loop from discovery to path restoration using traceable obstacle event objects.

[0071] like Figure 6As shown, the digital twin monitoring interface includes a map and vehicle pose display area 51 and an event and status display area 52. The map and vehicle pose display area 51 can display the AGV position, global reference path, local detour path, and key task areas; the event and status display area 52 can display the obstacle event number, obstacle event digital fingerprint, trust level, lifecycle status, health level, risk level, safety mode, and the validity period of manual intervention instructions. Manual pause, resume, confirm road closure, or rerouting instructions are associated with the event number, thereby preventing manual instructions from being mistakenly applied to other tasks or other obstacle events.

[0072] In the multi-vehicle implementation, the scheduling and control platform 2 can issue task semantic configurations to multiple AGVs respectively and establish an obstacle event sharing mechanism based on event numbers. This sharing mechanism synchronizes the obstacle location or spatial range, lifecycle status, risk level, and necessary digital fingerprint digests to the relevant AGVs. Before entering a critical area, the relevant AGVs verify the latest status of the event, corresponding passage conditions, and their own perceived health status based on the shared event number. The scheduling and control platform 2 can also coordinate the AGV entry order based on the spatial range, lifecycle status, and passage status of multiple obstacle event objects and critical areas. This implementation does not change the basic method by which a single AGV completes closed-loop processing based on obstacle event objects, perceived health status, and regression credentials.

[0073] The dispatch control platform 2 can also manage historical records of obstacle events. These records are linked to at least the event number, task number, candidate obstacle location, evidence data, obstacle event digital fingerprint, lifecycle status, handling strategy, regression credential, regression credential verification result, and handling result. It also retains records of the handling process corresponding to state transitions, reliability assessments, perceived health assessments, and operational control commands. Administrators can query the corresponding handling evidence chain by event number, task number, or digital fingerprint to review frequent obstacle events in specific areas, the reasons for manual takeover, and path recovery results.

[0074] Based on the spatial location, temporal information, digital fingerprints, and historical occurrence counts in historical obstacle event records, the scheduling and control platform 2 statistically analyzes the patterns of obstacle occurrence in task areas or path segments to obtain obstacle risk evolution prediction results. These prediction results can represent the probability of obstacle occurrence or risk level in the corresponding area during subsequent task periods. Without altering the basic closed-loop handling method, the platform can adjust the task semantic configuration accordingly before task issuance or during task execution, such as reducing the maximum permissible speed, increasing the safety margin, tightening automatic detour permissions, or increasing the event reporting level.

[0075] The aforementioned thresholds and parameters can be calibrated based on vehicle size, braking performance, maximum speed, sensor configuration, and site type. The core of this invention lies not in a fixed threshold, but in the correlated control of task semantic configuration, obstacle event reliability assessment, perceived health, and regression credential verification.

[0076] Example 2: An AGV vehicle obstacle closed-loop handling method, implemented using the AGV vehicle obstacle closed-loop handling system described in Example 1, includes the following steps: receiving AGV tasks from start to finish; generating task semantic configuration and global reference path based on the regional attributes of the areas traversed by the task path; acquiring environmental information and vehicle pose information around the AGV, forming candidate obstacle information and establishing obstacle event objects; managing the transition of the obstacle event objects between the states of pending verification, confirmed, handling, and pending resolution based on the association results in continuous sampling frames; performing a credibility evaluation on the obstacle event objects to determine whether the candidate obstacles are real obstacles, and performing a perception health evaluation on the perception link used to acquire the environmental information and vehicle pose information to determine whether the AGV has autonomous control capabilities; executing deceleration, partial detour, stopping, or manual takeover requests based on the task semantic configuration, credibility evaluation results, and perception health evaluation results; generating regression credentials before executing partial detour, and restoring global reference path tracking when the regression conditions defined by the regression credentials are met.

[0077] Establishing an obstacle event object includes: registering the current environmental perception data with a static map and extracting candidate residual clusters that are inconsistent with the static map; associating the candidate residual clusters with event numbers, locations, contours, timestamps, and relative motion estimates; and updating the lifecycle state of the obstacle event object based on the association results in consecutive sampling frames.

[0078] The credibility evaluation results are obtained using a weighted evaluation method. In the weighted evaluation method, each evidence item is normalized, and different weights or thresholds are assigned to different areas of regional risk. For personnel collaboration areas, the credibility evaluation threshold for automatic detour permission is reduced or the threshold for entering local detours is increased.

[0079] When communication is interrupted, the fusion positioning quality is lower than the preset value, the perception health is lower than the protection threshold, the regression credentials are invalid, or a qualified local detour path is not obtained within the preset time, the AGV is controlled to decelerate and stop, and the current obstacle event object and regression credentials are cached for subsequent processing.

[0080] Example 3: An AGV vehicle includes a processor, a memory, a lidar, a positioning device, a drive execution unit, and a communication device. The memory stores a program, and when the program is executed by the processor, it implements an AGV vehicle obstacle closed-loop handling method as described in Example 2.

[0081] Example 4: A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements an AGV vehicle obstacle closed-loop handling method as described in Example 2.

[0082] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several modifications and improvements can be made without departing from the inventive concept of this application, and these all fall within the protection scope of this application.

Claims

1. An AGV vehicle obstacle closed-loop handling system, characterized in that, include: The scheduling and control platform is configured to generate corresponding task semantic configurations based on the task path from the starting point to the ending point and the regional attributes of the areas traversed by the task path. A navigation sensing unit, configured to acquire information about the surrounding environment of the AGV and the vehicle's pose information; The vehicle controller is configured to: establish obstacle event objects based on the environmental information, and perform lifecycle management on the obstacle event objects in the stages of pending verification, confirmed status, in progress, and pending resolution. A reliable evaluation of the authenticity of the obstacle event object is performed, and a perception health evaluation of the AGV's autonomous control capability is performed on the navigation perception unit. Based on the task semantic configuration, trust evaluation results, perceived health evaluation results, and lifecycle status, generate operation control commands for deceleration, partial detour, or stopping, or output a manual takeover request.

2. The AGV vehicle obstacle closed-loop handling system according to claim 1, characterized in that, The task semantic configuration includes at least two or more of the following: area identifier, maximum permissible speed, safety margin, automatic detour permission, parking priority, manual takeover authority, and event reporting level; the area attributes include at least one of the following: personnel collaboration area, loading and unloading area, narrow passage, intersection, observation and display area, and emergency evacuation area.

3. The AGV vehicle obstacle closed-loop handling system according to claim 1, characterized in that, The obstacle event object, as a dynamic obstacle state object for continuous state representation and management of dynamic obstacles, includes event number, candidate obstacle location, formation time, evidence data, credibility, handling strategy, life cycle status, handling process record, and obstacle event digital fingerprint; The handling process records at least the life cycle state transition information, credible evaluation results, perceived health evaluation results, operation control instructions, and regression credential verification results to form a chain of evidence for the handling of the corresponding obstacle event object; The lifecycle status includes at least the status of pending verification, confirmed status, status of being processed, and status of pending expiration.

4. The AGV vehicle obstacle closed-loop handling system according to claim 3, characterized in that, The credible evaluation result is obtained from at least three of the following: cross-frame persistence, contour consistency, positioning quality, multi-source evidence consistency, and regional risk. When the credible evaluation result is lower than a first threshold, the vehicle controller controls the AGV to decelerate and increase the retest frequency. When the credible evaluation result reaches a second threshold and meets the collision risk conditions, the vehicle controller controls the AGV to perform local detour, stop and wait, or report anomalies, wherein the second threshold is greater than the first threshold.

5. The AGV vehicle obstacle closed-loop handling system according to claim 3, characterized in that, The obstacle event object in the pending verification state is only used for immediate security control and does not update the static map; when the same obstacle event object meets the preset confirmation conditions in multiple independent tasks and is confirmed by manual confirmation or automatic verification as a long-term existing object, the scheduling control platform generates a static map patch for the obstacle event object.

6. The AGV vehicle obstacle closed-loop handling system according to claim 1, characterized in that, The perception health evaluation is based on at least three of the following: LiDAR point cloud effectiveness, near-range ranging effectiveness, fusion positioning residual, visual visibility, and communication link quality. The vehicle controller switches between normal mode, degraded mode, and protection mode according to the perception health. The degraded mode limits the maximum speed, increases the safety margin, and increases the retest frequency. The protection mode prohibits automatic detours and controls the AGV to stop and report.

7. The AGV vehicle obstacle closed-loop handling system according to claim 1, characterized in that, The regression credentials include at least a global reference path segment identifier, a regression region, allowable lateral deviation, allowable heading deviation, allowable speed range, minimum perception health, obstacle event status requirements, and validity period. The vehicle controller resumes global reference path tracking when the AGV is located in the regression region, the lateral deviation and heading deviation meet the allowable range, the current speed is within the allowable speed range, the perception health is not lower than the minimum perception health, and the obstacle event object meets the obstacle event status requirements.

8. A closed-loop method for handling obstacles in AGV vehicles, characterized in that, Includes the following steps: Receive AGV tasks from the starting point to the end point, and generate task semantic configuration and global reference path based on the regional attributes of the areas traversed by the task path; The system acquires environmental information and vehicle pose information surrounding the AGV, forms candidate obstacle information, and establishes obstacle event objects. Based on the correlation results in continuous sampling frames, it manages the transitions of these obstacle event objects between states of pending verification, confirmed status, being handled, and awaiting resolution. It performs a credibility evaluation on the obstacle event objects to determine whether candidate obstacles are real obstacles, and evaluates the perception health of the perception links used to acquire the environmental and vehicle pose information to determine whether the AGV possesses autonomous control capabilities. Based on the task semantic configuration, credibility evaluation results, and perception health evaluation results, it executes requests for deceleration, partial detour, stopping, or manual takeover. Before executing a partial detour, it generates regression credentials and resumes global reference path tracking when the regression conditions defined by the regression credentials are met.

9. An AGV vehicle, characterized in that, It includes a processor, a memory, a lidar, a positioning device, a drive execution unit, and a communication device. The memory stores a program, which, when executed by the processor, implements the method of claim 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of claim 8.

Images