Automatic calibration and adaptive autonomous operation method for AGV operation

Abstract

This invention provides an automatic calibration and adaptive autonomous operation method for AGVs. Applied to the field of industrial mobile robot technology, the method includes: An AGV responds to a scheduling command and arrives in the vicinity of a workstation; identifies the workstation type based on environmental perception data and calls a workstation template matching the workstation type; controls the AGV to perform active detection actions based on the active detection strategy in the workstation template, collects multiple frames of observation data, and constructs a pose solution model based on the multi-frame observation data and the observable constraint set in the workstation template; calculates the docking pose and pose confidence by minimizing the geometric residual of the pose solution model; determines the current operation status based on the docking pose and the current environmental perception data; and determines an operation control strategy based on the current operation status and the pose confidence. This improves the efficiency of AGV adaptive autonomous operation.

Description

Technical Field

[0001] This invention relates to the field of industrial mobile robot technology, and in particular to an automatic calibration and adaptive autonomous operation method for AGVs. Background Technology

[0002] As a core flexible material handling equipment in the fields of intelligent manufacturing and intelligent warehousing, the accuracy, stability and efficiency of Automated Guided Vehicles (AGVs) in docking with workstations directly determine the overall efficiency of production line flow and warehousing operations.

[0003] Existing AGV workstation operation solutions mostly rely on manually pre-calibrated fixed workstation parameters and preset running paths offline. This presents significant technical shortcomings when facing flexible operation scenarios involving multiple workstation switching, workstation pose drift, and dynamic environmental disturbances. Firstly, the manual offline calibration process is cumbersome and time-consuming, resulting in high adaptation costs for production line changes and workstation adjustments, hindering rapid deployment. Secondly, conventional pose calculations often rely on single-frame observation data, making them susceptible to environmental interference such as lighting and occlusion, leading to insufficient pose calculation accuracy and an inability to quantify result confidence, easily causing docking failures. Thirdly, the lack of real-time operation status determination and adaptive strategy switching mechanisms prevents dynamic adjustment of docking strategies based on pose deviation and solution confidence, resulting in repetitive alignment, lengthy operation cycles, and low AGV adaptive autonomous operation efficiency. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention provides an automatic calibration and adaptive autonomous operation method for AGVs. The method includes:

[0005] When the AGV responds to the scheduling command and arrives in the vicinity of the workstation, it identifies the workstation type based on environmental perception data and calls up the operation template that matches the workstation type.

[0006] Based on the active detection strategy in the operation template, the AGV is controlled to perform active detection actions, collect multiple frames of observation data, and construct a pose solution model based on the multiple frames of observation data and the observable constraint set in the operation template.

[0007] The docking pose and pose confidence are calculated by minimizing the geometric residual of the pose solution model, and the current operation status is determined based on the docking pose and the current environmental perception data.

[0008] The operation control strategy is determined based on the current operation status and the pose confidence level, including: when the current operation status is determined to be able to be directly docked and the pose confidence level meets the high confidence level condition, a direct fine docking strategy is executed; when the current operation status is determined to require fine-tuning or the pose confidence level does not meet the high confidence level condition, a layered fine-tuning strategy is determined based on the degree of deviation.

[0009] During the execution of the job control strategy, feedback data is monitored in real time to determine whether the success criteria in the job template are met; if the success criteria are met, the pose drift is calculated based on the final job state, and the online calibration parameters in the job template are updated based on the pose drift.

[0010] Further, the invocation of a job template matching the workstation type includes:

[0011] The prior information of the work template is parsed to obtain the workstation geometry type and allowed approach direction;

[0012] Based on the observable constraint set corresponding to the workstation geometry type, the observable constraint set includes at least one of laser structure constraints, visual semantic constraints, and contact mechanical constraints.

[0013] An active detection strategy is determined based on the detectability score of features in the observable constraint set, and a fine-tuning control strategy and success criteria are set according to the workstation operation accuracy requirements.

[0014] The active detection strategy specifies the trajectory actions that the AGV performs in the vicinity of the workstation to improve observability.

[0015] Furthermore, based on the active detection strategy in the work template, the AGV is controlled to perform active detection actions, including:

[0016] Obtain the feature occlusion rate and feature symmetry coefficient from the current observation perspective;

[0017] Based on the feature occlusion rate, it is determined whether the observation angle needs to be changed. If the feature occlusion rate is greater than the preset occlusion threshold, the AGV is controlled to perform a rotation scanning action to obtain multi-angle observation data.

[0018] Based on the characteristic symmetry coefficient, it is determined whether it is necessary to eliminate pose ambiguity. If the characteristic symmetry coefficient is greater than the preset symmetry threshold, the AGV is controlled to perform a fan-shaped corner sweep or a small lateral movement.

[0019] Environmental feature points corresponding to the observable constraint set are extracted based on the observation data collected during the execution of the active detection action.

[0020] Furthermore, the docking pose and pose confidence are calculated by minimizing the geometric residuals of the pose solution model, including:

[0021] Construct the geometric residual function between the observed environmental feature points and the feature model in the task template;

[0022] The presence of a contact feedback signal is detected. If it is present, the mechanical constraint corresponding to the contact feedback signal is added as a penalty term to the geometric residual function.

[0023] The geometric residual function is iteratively solved using a nonlinear optimization algorithm to obtain the optimal docking pose;

[0024] Calculate the optimized residual covariance matrix, and determine the trace of the residual covariance matrix as the pose confidence.

[0025] If the pose confidence level is less than the preset minimum confidence threshold, it is determined that the current observation is insufficient and the secondary detection logic is triggered.

[0026] Furthermore, based on the docking pose and current environmental perception data, the current operational status is determined, including:

[0027] Calculate the minimum Euclidean distance between the obstacle and the planned path. If the minimum Euclidean distance is less than the safe distance threshold, the current work status is determined to be that the workstation is occupied.

[0028] Identify the outline boundary of the tool and calculate its geometric center. Determine the tool placement status based on the offset between the geometric center and the work station reference center. If the offset is greater than a preset skew threshold, the current operation status is determined to be tool placement skew.

[0029] Acquire real-time sensor data of the docking mechanism, and determine whether the docking mechanism is in an abnormal state based on the real-time sensor data.

[0030] Furthermore, based on the current work status and the pose confidence level, a work control strategy is determined, including:

[0031] The pose confidence level is compared with a preset high confidence threshold.

[0032] If the pose confidence is greater than or equal to the preset high confidence threshold and the current operation status is normal, then the operation control strategy is determined to be direct precision docking.

[0033] If the pose confidence is less than the preset high confidence threshold but greater than the preset low confidence threshold, or if the current operation status is that the tools are placed at an angle, then the operation control strategy is determined to be layered fine-tuning.

[0034] If the pose confidence level is less than the preset low confidence threshold or the current work status is that the workstation is occupied, then the work control strategy is determined to be to execute rollback or retry.

[0035] Furthermore, a tiered fine-tuning strategy is implemented, including:

[0036] Start coarse alignment layer control, and guide the AGV into the docking channel based on navigation and positioning data and the docking posture until the distance to the target point is less than the preset coarse alignment distance;

[0037] Switch to fine alignment layer control, calculate lateral and angular errors in real time based on short-distance high-frequency observation data, and adjust the closed-loop control gain based on the convergence trend of the lateral and angular errors.

[0038] During the fine alignment process, if a contact mechanical feedback signal is detected, the contact operation layer control is activated, and end-effector flexibility correction is performed based on the mechanical feedback signal.

[0039] Furthermore, activate contact operation layer control and perform end-effector flexibility correction, including:

[0040] Real-time acquisition of the current value of the lifting motor and the status of the contact switch;

[0041] Calculate the slope of the change in the current value. If the slope exceeds a preset contact slope threshold, it is determined that physical contact has been established.

[0042] Based on the difference between the buffer compression amount after contact and the preset compression target, a fine-tuning command is generated to control the AGV to continue moving at low speed until the difference converges to zero.

[0043] If the trigger time of the contact switch exceeds the preset safety time limit during the fine-tuning process, the contact is determined to be abnormal and the movement is stopped immediately.

[0044] Further, updating the online calibration parameters in the job template based on the pose drift includes:

[0045] Obtain the final docking pose at the time of job completion and the prior pose recorded in the job template;

[0046] The absolute value of the difference between the final docking pose and the prior pose is calculated as the pose drift.

[0047] The pose drift is compared with a preset update threshold. If the pose drift is greater than the preset update threshold but less than the abnormal drift threshold, the prior pose is corrected using the final docking pose based on a weighted average algorithm.

[0048] If the pose drift is greater than or equal to the abnormal drift threshold, the prior pose remains unchanged and the workstation is marked as requiring manual review.

[0049] Furthermore, the job control strategy is determined to be to execute rollback or retry, including:

[0050] Identify the types of anomalies that prevent the operation from continuing, including insufficient observation, path obstruction, and mechanism failure;

[0051] If the anomaly type is insufficient observation, the active detection action is re-executed after adjusting the detection amplitude or changing the detection angle.

[0052] If the anomaly type is path obstruction or mechanism failure, or if the number of retries reaches the preset maximum number of retries, the AGV will be controlled to return to the safe waiting area and the dispatch system will be notified to switch to the alternative workstation.

[0053] This invention enables standardized pre-adaptation of different workstation operation processes after the AGV responds to scheduling commands and arrives in the vicinity of the workstation. Based on environmental perception data, it identifies the workstation type and calls the matching operation template, improving the efficiency of scene switching and initial action execution for AGV operations. By collecting multi-frame observation data through the active detection strategy in the operation template and constructing a pose solution model based on the observable constraint set, it minimizes the geometric residual to calculate the docking pose and pose confidence and determine the current operation status, achieving high-precision solution of the docking pose and accurate prediction of the operation status, thus improving the reliability of AGV pose decision-making and the pre-accuracy of docking actions. By matching the current operation status with the pose confidence, a differentiated operation control strategy is applied. Real-time monitoring and feedback data during the operation process verify the operation success criteria. After the operation is completed, the online calibration parameters of the operation template are updated based on the pose drift, achieving hierarchical and precise control of operation actions and online self-calibration iteration of the operation template, improving the execution efficiency of AGV docking operations and the continuous stability efficiency of long-term adaptive operations.

[0054] It should be understood that the description in the Summary of the Invention is not intended to limit the key or essential features of the embodiments of the present invention, nor is it intended to restrict the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0055] The above and other features, advantages, and aspects of the various embodiments of the present invention will become more apparent from the accompanying drawings and the following detailed description. The drawings are provided for a better understanding of the invention and are not intended to limit the scope of the invention. In the drawings, the same or similar reference numerals denote the same or similar elements, wherein:

[0056] Figure 1 A flowchart illustrating an AGV operation automatic calibration and adaptive autonomous operation method according to an embodiment of the present invention is shown. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0059] Figure 1 A flowchart of an AGV operation automatic calibration and adaptive autonomous operation method according to an embodiment of the present invention is shown. The method includes:

[0060] S101, the AGV responds to the scheduling command and arrives in the workstation neighborhood, identifies the workstation type based on environmental perception data and calls the operation template that matches the workstation type;

[0061] S102, based on the active detection strategy in the operation template, control the AGV to perform active detection actions, collect multiple frames of observation data, and construct a pose solution model based on the multiple frames of observation data and the observable constraint set in the operation template;

[0062] S103, the docking pose and pose confidence are obtained by minimizing the geometric residual of the pose solution model, and the current operation status is determined based on the docking pose and the current environmental perception data.

[0063] S104, determine the operation control strategy based on the current operation status and the pose confidence level, including: when the current operation status is determined to be able to be directly docked and the pose confidence level meets the high confidence level condition, execute the direct fine docking strategy; when the current operation status is determined to require fine adjustment or the pose confidence level does not meet the high confidence level condition, determine the execution of the layered fine adjustment strategy based on the degree of deviation.

[0064] S105, during the execution of the operation control strategy, real-time monitoring of feedback data is performed to determine whether the success criteria in the operation template are met; if the success criteria are met, the pose drift is calculated based on the final operation state, and the online calibration parameters in the operation template are updated based on the pose drift.

[0065] In some embodiments, invoking a job template matching the workstation type includes: parsing prior information of the job template to obtain the workstation geometry type and allowed approach direction; associating a corresponding set of observable constraints based on the workstation geometry type, wherein the set of observable constraints includes at least one of laser structural constraints, visual semantic constraints, and contact mechanical constraints; determining an active detection strategy based on the detectability score of features in the set of observable constraints, and setting a fine-tuning control strategy and success criteria according to the workstation operation accuracy requirements; wherein the active detection strategy specifies the trajectory actions performed by the AGV in the vicinity of the workstation to improve observability. According to embodiments of the present invention, by parsing the prior information of the workstation template to obtain the workstation geometry type and allowed approach direction, precise limitation of the workstation work boundary and compliant approach path is achieved, preventing AGVs from illegally entering invalid work areas and improving the safety and compliance of work path planning. By associating the corresponding observable constraint set based on the workstation geometry type, adaptability matching of multi-source perception constraints is achieved, providing multi-dimensional constraint basis for subsequent pose solving and improving the robustness and accuracy of pose calculation. By determining the active detection strategy based on the detectability score of features in the observable constraint set, and setting fine-tuning control strategies and success criteria according to the workstation work accuracy requirements, personalized matching of detection actions, control logic and work accuracy requirements is achieved, avoiding the redundancy or insufficiency of accuracy in general strategies and improving the adaptability and work efficiency of AGV operations.

[0066] For example, for a standard pallet docking station identified by the system, the prior information of the work template is analyzed to obtain the station geometry as a 1200mm×1000mm standard pallet, with an allowable approach direction within ±15° of the pallet opening. Based on this pallet geometry, the corresponding set of observable constraints is associated, specifically including the linear structure constraints of the three pallet legs that can be identified by the LiDAR, the semantic constraints of the QR code on the pallet surface that can be identified by the vision camera, and the pressure contact mechanical constraints that can be collected by the lifting mechanism. The detectability scores of each feature in the set of observable constraints are calculated, resulting in a LiDAR leg feature score of 0.92, a vision QR code feature score of 0.85, and a contact mechanical constraint score of 0.78. Based on this, the active detection strategy is determined to be a combination of fixed-axis rotation and small forward and backward translation within a 1.5m radius of the station. At the same time, based on the station operation accuracy requirement of ±10mm, the fine-tuning control strategy is set as PID closed-loop control, with success criteria being docking posture deviation ≤8mm, angle deviation ≤0.5°, and lifting in place signal triggering.

[0067] In some embodiments, the AGV is controlled to perform active detection actions based on the active detection strategy in the operation template, including: acquiring the feature occlusion rate and feature symmetry coefficient under the current observation view; determining whether to change the observation view based on the feature occlusion rate, and if the feature occlusion rate is greater than a preset occlusion threshold, controlling the AGV to perform a rotation scanning action to obtain multi-angle observation data; determining whether to eliminate pose ambiguity based on the feature symmetry coefficient, and if the feature symmetry coefficient is greater than a preset symmetry threshold, controlling the AGV to perform a fan-shaped corner sweep or a small lateral movement action; and extracting environmental feature points corresponding to the observable constraint set based on the observation data collected during the execution of the active detection action. According to embodiments of the present invention, by acquiring the feature occlusion rate and feature symmetry coefficient under the current observation viewpoint, a preliminary quantitative assessment of the validity of the current observation data is achieved, providing a precise basis for triggering active detection actions, avoiding the execution of invalid detection actions, and improving the rationality and efficiency of the detection process; by determining based on the feature occlusion rate and controlling the AGV to perform rotational scanning actions to obtain multi-angle observation data, effective completion of occluded features is achieved, eliminating the problem of observation gaps caused by occlusion and improving the completeness of environmental feature extraction; by determining based on the feature symmetry coefficient and controlling the AGV to perform fan-shaped corner sweeping or small-amplitude lateral movement, effective elimination of pose ambiguity is achieved, avoiding pose solution errors caused by symmetrical features, and improving the accuracy and uniqueness of docking pose calculation.

[0068] For example, the AGV is currently 1.8m directly in front of the pallet. The occlusion rate of the pallet legs is 38% and the feature symmetry coefficient is 0.78 from the current observation view. The preset occlusion threshold is 30%. Since 38% > 30%, the AGV is controlled to perform a 30° clockwise and 30° counterclockwise rotation scanning action to collect 6 frames of multi-angle laser observation data. The preset symmetry threshold is 0.7. Since 0.78 > 0.7, the AGV is controlled to perform a small lateral movement of ±100mm to eliminate pose ambiguity. From the observation data collected during the active detection action, 9 laser feature points and 1 visual QR code corner feature of the 3 pallet legs corresponding to the observable constraint set are extracted.

[0069] In some embodiments, the docking pose and pose confidence are obtained by minimizing the geometric residual of the pose solution model, including: constructing a geometric residual function between the observed environmental feature points and the feature model in the operation template; detecting whether there is a contact feedback signal, and if so, adding the mechanical constraint corresponding to the contact feedback signal as a penalty term to the geometric residual function; using a nonlinear optimization algorithm to iteratively solve the geometric residual function to obtain the optimal docking pose; calculating the optimized residual covariance matrix, and determining the trace of the residual covariance matrix as the pose confidence; if the pose confidence is less than a preset minimum confidence threshold, determining that the current observation is insufficient and triggering secondary detection logic. According to embodiments of the present invention, by constructing a geometric residual function between observed environmental feature points and feature models in the operation template, quantitative modeling of docking pose solution is realized, providing a core computational basis for pose optimization and improving the standardization and iterability of pose solution. By detecting contact feedback signals and adding the corresponding mechanical constraints as penalty terms to the geometric residual function, multi-source fusion constraints of visual / laser perception and mechanical perception are realized, making up for the insufficient accuracy of single geometric perception in close-range docking and improving the accuracy of close-range docking pose solution. By using a nonlinear optimization algorithm to iteratively solve for the optimal docking pose and determining the pose confidence based on the trace of the residual covariance matrix, the optimal solution of docking pose and the quantitative evaluation of the reliability of the solution results are realized. At the same time, by triggering secondary detection logic through confidence threshold determination, operation failure caused by low reliability pose is avoided, improving the success rate and stability of AGV docking operation.

[0070] For example, the geometric residual functions of point-to-line and point-to-point relationships between the extracted 9 laser feature points, 1 visual corner point, and the standard pallet feature model in the work template are constructed; no contact feedback signal is detected from the pressure sensor of the lifting mechanism, and no additional penalty term is added to the geometric residual function; the Gauss-Newton nonlinear optimization algorithm is used to solve the geometric residual function in 8 iterations to obtain the optimal docking pose in the world coordinate system as x=12.352m, y=8.624m, θ=35.28°; the trace of the optimized residual covariance matrix is ​​calculated to be 0.032, and this value is determined as the pose confidence level; the preset minimum confidence threshold is 0.05. Since 0.032 < 0.05, it is determined that the current observation is sufficient and the secondary detection logic is not triggered.

[0071] In some embodiments, determining the current operational status based on the docking pose and current environmental perception data includes: calculating the minimum Euclidean distance between the obstacle and the planned path; if the minimum Euclidean distance is less than a safe distance threshold, determining the current operational status as the workstation being occupied; identifying the outline boundary of the tool and calculating its geometric center; determining the tool placement status based on the offset between the geometric center and the workstation reference center; if the offset is greater than a preset skew threshold, determining the current operational status as the tool placement being skewed; acquiring real-time sensor data of the docking mechanism; and determining whether the docking mechanism is in an abnormal state based on the real-time sensor data. According to embodiments of the present invention, by calculating the minimum Euclidean distance between obstacles and the planned path and determining whether the workstation is occupied, a preliminary verification of the accessibility of the work path is achieved, avoiding collision risks in advance and improving the safety of AGV operations. By identifying the outline boundary of the tool and calculating the offset between its geometric center and the workstation reference center, a precise quantitative determination of the tool placement status is achieved, providing a direct basis for the selection of subsequent operation control strategies, avoiding docking failures caused by tool misalignment, and improving the fault tolerance of docking operations. By acquiring real-time sensor data of the docking mechanism to determine whether the docking mechanism is in an abnormal state, real-time status monitoring of the operation execution mechanism is achieved, identifying the risk of mechanism failure in advance, avoiding equipment damage or operation accidents caused by faulty operation, and improving the equipment safety and operational reliability of AGV operations.

[0072] For example, the minimum Euclidean distance between obstacles around the planned docking path and the path centerline is calculated to be 320mm, and the preset safety distance threshold is 200mm. Since 320mm > 200mm, the current working status is determined to be that the workstation is not occupied. The outline boundary of the cross pallet is identified, and its geometric center coordinates are calculated to be x=12.358m, y=8.627m. The offset from the workstation reference center x=12.350m, y=8.620m is 10.6mm, and the preset skew threshold is 15mm. Since 10.6mm < 15mm, the current working status is determined to be that the materials are placed normally. The real-time data of the displacement sensor and pressure sensor of the lifting mechanism are obtained. The values ​​are all within the rated working range, and the docking mechanism is determined to be in no abnormal state.

[0073] In some embodiments, determining a work control strategy based on the current work status and the pose confidence level includes: comparing the pose confidence level with a preset high confidence threshold; if the pose confidence level is greater than or equal to the preset high confidence threshold and the current work status is normal, then the work control strategy is determined to be direct fine docking; if the pose confidence level is less than the preset high confidence threshold but greater than a preset low confidence threshold, or the current work status is that the tooling is misaligned, then the work control strategy is determined to be layered fine-tuning; if the pose confidence level is less than the preset low confidence threshold or the current work status is that the workstation is occupied, then the work control strategy is determined to be to execute rollback or retry. According to embodiments of the present invention, by comparing the pose confidence level with preset high and low confidence thresholds, a graded determination of the reliability of the pose solution results is achieved, providing a quantitative standard for the differentiated selection of operation control strategies and improving the accuracy of strategy matching. By combining the determination based on the pose confidence level and the current operation status, differentiated control strategies such as direct fine docking, layered fine-tuning, rollback, or retry are matched respectively, realizing adaptive graded control of the operation process. In high-confidence scenarios, docking is directly executed to improve operation efficiency, while in low-reliability or abnormal scenarios, corresponding fault-tolerant strategies are executed, balancing operation efficiency and operation safety. By setting rollback or retry strategies for scenarios with insufficient pose confidence or occupied workstations, pre-emptive handling of operation anomalies is achieved, avoiding docking failures or equipment collisions caused by forced operation, and improving the fault tolerance and on-site adaptability of AGV operations.

[0074] For example, the pose confidence score of 0.032 is compared with the preset high confidence threshold of 0.04. Since 0.032 ≤ 0.04 and the current operation status is normal, the operation control strategy is determined to be direct fine docking. If the pose confidence score is 0.046 (less than the high confidence threshold of 0.04 but greater than the low confidence threshold of 0.05), or the tool offset is 18mm (greater than the preset skew threshold of 15mm), the operation control strategy is determined to be layered fine adjustment. If the pose confidence score is 0.062 (greater than the low confidence threshold of 0.05) or the minimum distance between the obstacle and the path is 150mm (less than the safe distance threshold of 200mm), the operation control strategy is determined to be to execute rollback or retry.

[0075] In some embodiments, a layered fine-tuning strategy is implemented, including: initiating coarse alignment layer control, guiding the AGV into the docking channel based on navigation and positioning data and the docking pose until the distance to the target point is less than a preset coarse alignment distance; switching to fine alignment layer control, calculating the lateral error and angle error in real time based on short-distance high-frequency observation data, and adjusting the closed-loop control gain based on the convergence trend of the lateral error and angle error; during the fine alignment process, if a contact mechanical feedback signal is detected, activating the contact operation layer control, and performing end-effector flexibility correction based on the mechanical feedback signal. According to embodiments of the present invention, by activating coarse alignment layer control to guide the AGV into the docking channel to a preset coarse alignment distance, the AGV achieves rapid and accurate positioning from the workstation neighborhood to the approach point, laying the foundation for subsequent fine alignment operations, shortening the overall docking operation time, and improving operational efficiency. By switching to fine alignment layer control, the error is calculated in real time based on short-distance high-frequency observation data, and the closed-loop control gain is adjusted, achieving dynamic high-precision control during close-range docking. The control parameters are adaptively adjusted according to the error convergence trend, avoiding overshoot or slow convergence caused by fixed gain, and improving the control accuracy and convergence speed of the fine alignment process. By activating contact operation layer control when a contact mechanical feedback signal is detected, and performing end-effector flexibility correction based on the mechanical feedback signal, compliant control during the hard contact stage is achieved, avoiding equipment impact and damage caused by rigid docking, and improving the operational safety and docking fit at the docking end.

[0076] For example, coarse alignment layer control is initiated, based on AGV wheel odometer + laser SLAM navigation and positioning data and optimal docking posture, guiding the AGV to enter the pallet docking channel at a speed of 0.5m / s until the straight-line distance between the center of the AGV lifting mechanism and the target point of the pallet docking is 220mm, which is less than the preset coarse alignment distance of 300mm; switching to fine alignment layer control, based on short-range laser observation data at a frequency of 10Hz, the lateral error is calculated in real time to be ±4mm and the angle error to be ±0.2°, and the proportional gain of the PID closed-loop control is reduced from 1.2 to 0.8 according to the error convergence trend to ensure smooth convergence; during the fine alignment process, the contact mechanical feedback signal of the lifting mechanism pressure sensor is monitored, the contact operation layer control is activated, and end-effector flexibility correction is performed based on the mechanical feedback signal.

[0077] In some embodiments, activating contact operation layer control and performing end-effector flexibility correction includes: real-time acquisition of the current value of the lifting motor and the status of the contact edge switch; calculating the slope of the change in the current value; if the slope of the change exceeds a preset contact slope threshold, determining that physical contact has been established; generating a fine-tuning command based on the difference between the buffer compression amount after contact is established and a preset compression target, controlling the AGV to continue moving at low speed until the difference converges to zero; if the contact edge switch trigger time exceeds a preset safety time limit during the fine-tuning process, determining that the contact is abnormal and immediately stopping the movement. According to embodiments of the present invention, by real-time acquisition of the current value of the lifting motor and the status of the contact switch, real-time monitoring of multi-dimensional status signals during the docking process is realized, providing a comprehensive data source for contact determination and anomaly identification, and improving the comprehensiveness and reliability of contact status perception; by calculating the slope of the change in current value to determine the physical contact status, and generating fine-tuning instructions based on the difference between the buffer compression amount and the preset compression target, precise quantitative control of the docking stroke after contact is realized, ensuring that the docking fit meets the operational requirements and improving the operational accuracy of the end docking; by determining contact anomalies and immediately stopping movement when the contact switch trigger time exceeds the preset safety time limit, rapid response and safety protection for docking anomalies are realized, avoiding damage to equipment or materials caused by excessive compression, and improving the safety and fault tolerance of the contact operation process.

[0078] For example, the current value of the lifting motor and the on / off status of the contact switch at the front end of the AGV are collected in real time; the slope of the current value change is calculated to be 1.2A / s, which exceeds the preset contact slope threshold of 0.8A / s, indicating that the AGV lifting mechanism and the pallet have established stable physical contact; based on the difference of 1.8mm between the buffer compression amount of 3.2mm after the contact is established and the preset compression target of 5mm, a forward fine-tuning command of 0.05m / s is generated to control the AGV to continue moving at low speed until the difference between the buffer compression amount and the preset compression target converges to within ±0.2mm; if the trigger duration of the contact switch reaches 450ms during the fine-tuning process, which exceeds the preset safety time limit of 300ms, then the contact is determined to be abnormal and all AGV movements are stopped immediately.

[0079] In some embodiments, updating the online calibration parameters in the job template based on the pose drift includes: obtaining the final docking pose at the job completion time and the prior pose recorded in the job template; calculating the absolute value of the difference between the final docking pose and the prior pose as the pose drift; comparing the pose drift with a preset update threshold; if the pose drift is greater than the preset update threshold and less than an abnormal drift threshold, then correcting the prior pose using the final docking pose based on a weighted average algorithm; if the pose drift is greater than or equal to the abnormal drift threshold, then keeping the prior pose unchanged and marking the workstation as requiring manual review. According to embodiments of the present invention, by calculating the absolute value of the difference between the final docking pose and the prior pose at the time of job completion as the pose drift, the quantitative identification of long-term pose drift of the workstation is realized, providing accurate numerical basis for online updating of the job template and improving the rationality of template parameter iteration; by comparing the pose drift with a preset update threshold and an abnormal drift threshold, the prior pose is corrected based on a weighted average algorithm within a reasonable drift range, realizing online self-calibration and adaptive updating of job template parameters. It can adapt to the slow pose changes of the workstation without manual offline calibration, reducing manual maintenance costs and improving the long-term applicability of the job template; by keeping the prior pose unchanged and marking the workstation as needing manual review when the drift exceeds the abnormal drift threshold, the risk control of abnormal large drift is realized, avoiding template parameter failure caused by erroneous data, balancing the flexibility of template self-updating and the stability of the operation benchmark, and improving the reliability of AGV long-term operation.

[0080] For example, the final docking pose at the time of job completion is obtained as x = 12.354m, y = 8.625m, θ = 35.29°, and the prior pose recorded in the job template is x = 12.347m, y = 8.620m, θ = 35.26°. The absolute value of the difference in planar position between the two is calculated to be 8.6mm, and the absolute value of the difference in angle is 0.03°, determining the pose drift as 8.6mm. The pose drift is then compared with the preset update threshold of 8mm and the abnormal drift threshold of 25mm. Since 8mm < 8.6mm < 25mm, based on a weighted average algorithm, the weight of the current final pose is taken as 0.3, and the weight of the historical prior pose as 0.7. The corrected prior pose is then x = 12.349m, y = 8.622m, θ = 35.27°; If the pose drift reaches 32mm, which is greater than or equal to the abnormal drift threshold of 25mm, then the original pose remains unchanged, and the workstation is marked as requiring manual review.

[0081] In some embodiments, determining the operation control strategy as rollback or retry includes: identifying the type of anomaly that prevents the operation from continuing, the anomaly types including insufficient observation, path obstruction, and mechanism failure; if the anomaly type is insufficient observation, then adjusting the detection amplitude or changing the detection angle and re-executing the active detection action; if the anomaly type is path obstruction or mechanism failure, or if the number of retries reaches the preset maximum number of retries, then controlling the AGV to return to the safe waiting area and reporting to the scheduling system to switch to the alternative workstation. According to embodiments of the present invention, by identifying the types of anomalies that cause the inability to continue operations, accurate classification and localization of operational faults are achieved, providing a basis for differentiated matching of anomaly handling strategies, avoiding the ineffectiveness of general handling strategies, and improving the pertinence and efficiency of anomaly handling; by adjusting the detection amplitude or changing the detection perspective for anomalies with insufficient observation and then re-executing the active detection action, self-repair handling of observation-related anomalies is achieved, which can supplement effective observation data and restore normal operation processes without manual intervention, thereby improving the autonomous fault tolerance capability of AGV operations; by controlling the AGV to return to the safe waiting area and reporting to the scheduling system to switch to the alternative workstation in scenarios such as path obstruction, mechanism failure, or reaching the maximum number of retries, a safety fallback handling for unrepairable anomalies is achieved, avoiding the AGV from staying at the faulty workstation for a long time and affecting the overall scheduling efficiency, while ensuring the safety of equipment and the site, and improving the overall scheduling efficiency and operational stability of AGV cluster operations.

[0082] For example, if the anomaly type that prevents the operation from continuing is identified as insufficient observation, the corresponding pose confidence level of 0.062 is greater than the preset low confidence threshold of 0.05. The original rotation scanning range is adjusted from ±30° to ±45°, the detection angle is changed to 1.2m to the left of the pallet, and the active detection action is re-executed. If the anomaly type is identified as path obstruction (static obstacles exist in the workstation) or lifting motor failure, or the number of active detection retries reaches the preset maximum number of retries (3), the AGV is controlled to return to the safe waiting area 3m away from the workstation, and the warehouse scheduling system is reported to switch to the backup workstation No. 2 in the same area to perform the operation.

[0083] It should be understood that the various processes described above can be used to rearrange, add, or delete steps. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0084] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for automatic calibration and adaptive autonomous operation of AGVs, characterized in that, include: When the AGV responds to the scheduling command and arrives in the vicinity of the workstation, it identifies the workstation type based on environmental perception data and calls up the operation template that matches the workstation type. Based on the active detection strategy in the operation template, the AGV is controlled to perform active detection actions, collect multiple frames of observation data, and construct a pose solution model based on the multiple frames of observation data and the observable constraint set in the operation template. The docking pose and pose confidence are calculated by minimizing the geometric residual of the pose solution model, and the current operation status is determined based on the docking pose and the current environmental perception data. The operation control strategy is determined based on the current operation status and the pose confidence level, including: when the current operation status is determined to be able to be directly docked and the pose confidence level meets the high confidence level condition, a direct fine docking strategy is executed; when the current operation status is determined to require fine-tuning or the pose confidence level does not meet the high confidence level condition, a layered fine-tuning strategy is determined based on the degree of deviation. During the execution of the job control strategy, feedback data is monitored in real time to determine whether the success criteria in the job template are met; if the success criteria are met, the pose drift is calculated based on the final job state, and the online calibration parameters in the job template are updated based on the pose drift.

2. The AGV operation automatic calibration and adaptive autonomous operation method according to claim 1, characterized in that, The process of calling a job template that matches the workstation type includes: The prior information of the work template is parsed to obtain the workstation geometry type and allowed approach direction; Based on the observable constraint set corresponding to the workstation geometry type, the observable constraint set includes at least one of laser structure constraints, visual semantic constraints, and contact mechanical constraints. An active detection strategy is determined based on the detectability score of features in the observable constraint set, and a fine-tuning control strategy and success criteria are set according to the workstation operation accuracy requirements. The active detection strategy specifies the trajectory actions that the AGV performs in the vicinity of the workstation to improve observability.

3. The AGV operation automatic calibration and adaptive autonomous operation method according to claim 2, characterized in that, Based on the active detection strategy in the aforementioned task template, the AGV is controlled to perform active detection actions, including: Obtain the feature occlusion rate and feature symmetry coefficient from the current observation perspective; Based on the feature occlusion rate, it is determined whether the observation angle needs to be changed. If the feature occlusion rate is greater than the preset occlusion threshold, the AGV is controlled to perform a rotation scanning action to obtain multi-angle observation data. Based on the characteristic symmetry coefficient, it is determined whether it is necessary to eliminate pose ambiguity. If the characteristic symmetry coefficient is greater than the preset symmetry threshold, the AGV is controlled to perform a fan-shaped corner sweep or a small lateral movement. Environmental feature points corresponding to the observable constraint set are extracted based on the observation data collected during the execution of the active detection action.

4. The AGV operation automatic calibration and adaptive autonomous operation method according to claim 3, characterized in that, The docking pose and pose confidence are obtained by minimizing the geometric residuals of the pose solution model, including: Construct the geometric residual function between the observed environmental feature points and the feature model in the task template; The presence of a contact feedback signal is detected. If it is present, the mechanical constraint corresponding to the contact feedback signal is added as a penalty term to the geometric residual function. The geometric residual function is iteratively solved using a nonlinear optimization algorithm to obtain the optimal docking pose; Calculate the optimized residual covariance matrix, and determine the trace of the residual covariance matrix as the pose confidence. If the pose confidence level is less than the preset minimum confidence threshold, it is determined that the current observation is insufficient and the secondary detection logic is triggered.

5. The AGV operation automatic calibration and adaptive autonomous operation method according to claim 4, characterized in that, Determining the operational status based on the docking pose and current environmental perception data includes: Calculate the minimum Euclidean distance between the obstacle and the planned path. If the minimum Euclidean distance is less than the safe distance threshold, the current work status is determined to be that the workstation is occupied. Identify the outline boundary of the tool and calculate its geometric center. Determine the tool placement status based on the offset between the geometric center and the work station reference center. If the offset is greater than a preset skew threshold, the current operation status is determined to be tool placement skew. Acquire real-time sensor data of the docking mechanism, and determine whether the docking mechanism is in an abnormal state based on the real-time sensor data.

6. The AGV operation automatic calibration and adaptive autonomous operation method according to claim 5, characterized in that, Based on the current work status and the pose confidence level, a work control strategy is determined, including: The pose confidence level is compared with a preset high confidence threshold. If the pose confidence is greater than or equal to the preset high confidence threshold and the current operation status is normal, then the operation control strategy is determined to be direct precision docking. If the pose confidence is less than the preset high confidence threshold but greater than the preset low confidence threshold, or if the current operation status is that the tools are placed at an angle, then the operation control strategy is determined to be layered fine-tuning. If the pose confidence level is less than the preset low confidence threshold or the current work status is that the workstation is occupied, then the work control strategy is determined to be to execute rollback or retry.

7. The AGV operation automatic calibration and adaptive autonomous operation method according to claim 6, characterized in that, Implement a tiered fine-tuning strategy, including: Start coarse alignment layer control, and guide the AGV into the docking channel based on navigation and positioning data and the docking posture until the distance to the target point is less than the preset coarse alignment distance; Switch to fine alignment layer control, calculate lateral and angular errors in real time based on short-distance high-frequency observation data, and adjust the closed-loop control gain based on the convergence trend of the lateral and angular errors. During the fine alignment process, if a contact mechanical feedback signal is detected, the contact operation layer control is activated, and end-effector flexibility correction is performed based on the mechanical feedback signal.

8. The AGV operation automatic calibration and adaptive autonomous operation method according to claim 7, characterized in that, Activate contact operation layer control and perform end-effector flexibility correction, including: Real-time acquisition of the current value of the lifting motor and the status of the contact switch; Calculate the slope of the change in the current value. If the slope exceeds a preset contact slope threshold, it is determined that physical contact has been established. Based on the difference between the buffer compression amount after contact and the preset compression target, a fine-tuning command is generated to control the AGV to continue moving at low speed until the difference converges to zero. If the trigger time of the contact switch exceeds the preset safety time limit during the fine-tuning process, the contact is determined to be abnormal and the movement is stopped immediately.

9. The AGV operation automatic calibration and adaptive autonomous operation method according to claim 8, characterized in that, Updating the online calibration parameters in the job template based on the pose drift includes: Obtain the final docking pose at the time of job completion and the prior pose recorded in the job template; The absolute value of the difference between the final docking pose and the prior pose is calculated as the pose drift. The pose drift is compared with a preset update threshold. If the pose drift is greater than the preset update threshold but less than the abnormal drift threshold, the prior pose is corrected using the final docking pose based on a weighted average algorithm. If the pose drift is greater than or equal to the abnormal drift threshold, the prior pose remains unchanged and the workstation is marked as requiring manual review.

10. The AGV operation automatic calibration and adaptive autonomous operation method according to claim 9, characterized in that, The job control strategy is determined to be either rollback or retry, including: Identify the types of anomalies that prevent the operation from continuing, including insufficient observation, path obstruction, and mechanism failure; If the anomaly type is insufficient observation, the active detection action is re-executed after adjusting the detection amplitude or changing the detection angle. If the anomaly type is path obstruction or mechanism failure, or if the number of retries reaches the preset maximum number of retries, the AGV will be controlled to return to the safe waiting area and the dispatch system will be notified to switch to the alternative workstation.

Images