AGV global positioning controller based on double-filter arbitration and space-time verification

Abstract

The invention provides an AGV global positioning controller based on double-filter arbitration and space-time verification. The AGV global positioning controller comprises a visual motion estimator, a global absolute positioner and an intelligent arbitration center. The visual motion estimator generates relative pose and visual positioning confidence through data of a depth camera, an inertial measurement unit and a wheel type odometer; the global absolute locator outputs an absolute pose and a global positioning confidence coefficient through ultra-wideband base station ranging data; the intelligent arbitration center dynamically selects a fusion strategy according to the confidence degree, performs weighted fusion in a normal state, and adopts speed interpolation or drift compensation to realize positioning source switching in an abnormal state, so that the problem of positioning jump caused by failure of a single sensor in a complex environment is effectively solved; the positioning stability can be improved, smooth switching and abnormity correction are realized, and the AGV global positioning precision is further improved.

Description

Technical Field

[0001] This invention relates to the field of automated guided vehicle (AGV) positioning control technology, and in particular to an AGV global positioning controller based on dual-filter arbitration and spatiotemporal verification. Background Technology

[0002] In industrial automation scenarios, accurate positioning of Automated Guided Vehicles (AGVs) is crucial for their stable operation. Currently, the mainstream positioning methods for AGVs include Simultaneous Localization and Mapping (SLAM) and Ultra-Wide Band (UWB) positioning. SLAM relies on cameras to achieve high-precision pose estimation, but it is susceptible to occlusion and changes in lighting conditions, leading to failure. UWB positioning offers better stability and can provide absolute coordinates globally, but it is prone to multipath interference in dense metal environments, resulting in ranging errors.

[0003] Existing fusion solutions are mostly based on a single filter (such as Kalman filter). Although they can work under normal conditions, they lack a robust decision-making mechanism for sensor failure. When vision or UWB suddenly fails, it can easily cause positioning jumps, leading to unstable AGV control or even sudden stops.

[0004] Therefore, there is an urgent need for an AGV global positioning scheme that can maintain continuous positioning, suppress environmental interference, and achieve smooth switching when a single sensor fails, so as to improve the accuracy of AGV global positioning. Summary of the Invention

[0005] This invention provides an AG global positioning controller based on dual-filter arbitration and spatiotemporal verification, which can improve positioning stability in complex industrial environments, realize intelligent switching and fusion of multi-source positioning data, and improve the accuracy of AGV global positioning.

[0006] On one hand, the present invention provides an AGV global positioning controller based on dual-filter arbitration and spatiotemporal verification, which includes: The visual motion estimator is used to acquire depth camera images, inertial measurement unit data, and wheel odometer data, process them, and output the relative pose, relative speed, and visual positioning reliability of the AGV. A global absolute positioner is used to acquire ranging values ​​from at least three ultra-wideband base stations and data from the inertial measurement unit, process them, and output the absolute pose of the AGV, the absolute speed of the AGV, and the global positioning confidence level. Intelligent arbitration center, used for: The relative pose, relative velocity, visual positioning confidence, absolute pose, absolute velocity, and global positioning confidence are obtained. Data synchronization is achieved through hardware timestamps. After converting the relative pose and relative velocity into global pose and global velocity in the global coordinate system, data analysis is performed to obtain data analysis results. When the data analysis results do not meet the preset switching conditions, the global pose and the absolute pose are weighted and fused to obtain the first fused pose. When the data analysis results meet the preset switching conditions, if the preset switching condition is to switch to global positioning-dominated, a velocity-maintaining interpolation strategy is adopted, and the global velocity is used to perform motion integration on the absolute pose to obtain the second fused pose; if the preset switching condition is to switch to visual positioning-dominated, a drift compensation strategy is adopted, and the position deviation compensation amount between the absolute pose and the relative pose is superimposed on the global pose to obtain the third fused pose.

[0007] According to the present invention, an AGV global positioning controller based on dual-filter arbitration and spatiotemporal verification is provided, wherein the intelligent arbitration center comprises: The first fusion module is used for: When the data analysis results do not meet the preset switching conditions, calculate the visual positioning weight and the global positioning weight; The first product value is obtained by multiplying the visual localization weights with the global pose. The second product value is obtained by multiplying the global positioning weights and the absolute pose. The first fused pose is obtained by summing the first product value and the second product value.

[0008] According to the present invention, an AGV global positioning controller based on dual-filter arbitration and spatiotemporal verification is provided, wherein the first fusion module includes: The weight calculation unit is used for: Calculate the magnitude of the velocity deviation between the global velocity and the absolute velocity, and calculate a first quotient of the magnitude of the velocity deviation and the maximum permissible velocity; calculate a first difference between 1 and the first quotient, and calculate a third product of the first difference and the visual positioning confidence; calculate a first sum of the visual positioning confidence and the global positioning confidence, and calculate a second quotient of the third product and the first sum as the visual positioning weight; Calculate the magnitude of the pose deviation between the global pose and the absolute pose, and calculate the third quotient of the magnitude of the pose deviation and the maximum allowable position deviation; calculate the second difference between 1 and the third quotient, and calculate the fourth product of the second difference and the global positioning confidence; calculate the fourth quotient of the fourth product and the first sum as the global positioning weight.

[0009] According to the present invention, an AGV global positioning controller based on dual-filter arbitration and spatiotemporal verification is provided, wherein the intelligent arbitration center comprises: The second fusion module is used for: The global velocity is decomposed into global linear velocity and global angular velocity, and motion integration is performed on the absolute pose with the intelligent arbitration center control cycle as the step size to generate a temporary fused pose. The temporary fused pose is smoothed by a first-order low-pass filter, and the second fused pose is output.

[0010] According to the present invention, an AGV global positioning controller based on dual-filter arbitration and spatiotemporal verification is provided, wherein the intelligent arbitration center comprises: The third fusion module is used for: Calculate the current position deviation between the absolute pose and the relative pose; Calculate the fifth product of the current position deviation and the preset compensation coefficient; The third fused pose is obtained by calculating the second sum of the global pose and the fifth product value.

[0011] According to the present invention, an AGV global positioning controller based on dual-filter arbitration and spatiotemporal verification is provided. The third fusion module is further used for: The rate of change of position deviation is determined based on the current position deviation and the previous position deviation; Based on the correlation between position deviation and compensation coefficient, the initial compensation coefficient is determined; Based on the correlation between the rate of change of position deviation and the compensation adjustment ratio, the compensation adjustment ratio corresponding to the rate of change of position deviation is determined, so as to adjust the initial compensation coefficient and obtain the preset compensation coefficient.

[0012] According to the present invention, an AGV global positioning controller based on dual-filter arbitration and spatiotemporal verification is provided, wherein the visual motion estimator is specifically used for: After median filtering preprocessing of the depth camera image, a 3D point cloud is generated. The iterative nearest point algorithm is used to register two consecutive frames of point cloud, and the optimal rotation matrix and translation vector are solved to output the relative pose of the AGV, the local relative velocity of the AGV, and the visual positioning confidence. The process of determining the visual location confidence includes: Determine a fifth quotient between the desired accuracy and the accuracy of the relative pose; Determine the sixth quotient between the number of valid matching points in the 3D point cloud and the total number of points in the source point cloud that participated in the registration; A two-dimensional evaluation logic combining image features and motion state assistance is adopted to calculate the image feature score and the motion state correction score respectively, and then the image feature score and the motion state correction score are fused to obtain the motion blur score; The visual location reliability is obtained by weighted summation of the fifth quotient, the sixth quotient, and the motion blur score.

[0013] According to the present invention, an AGV global positioning controller based on dual-filter arbitration and spatiotemporal verification is provided, wherein the global absolute positioner is specifically used for: Anomalies are detected by historical rate of change. An overdetermined system of equations is constructed based on base station coordinates. The overdetermined system of equations is solved by the least squares method to obtain the absolute coordinates of the AGV, so as to output the absolute pose, absolute speed and global positioning confidence of the AGV. The process of determining the global location confidence includes: Determine the seventh quotient of the number of base stations that have successfully measured distances and 3, and determine the reciprocal of the geometric precision factor; The global location confidence is obtained by weighted summation of the seventh quotient and the reciprocal.

[0014] According to the present invention, an AGV global positioning controller based on dual-filter arbitration and spatiotemporal verification is provided, wherein the intelligent arbitration center is further used for: When the data analysis results meet the preset switching conditions, if the preset switching conditions are predictive switching, the weight of the fault source is reduced to a preset ratio, and the pose of the non-fault source at the previous moment is preloaded; wherein, when the fault source is a visual motion estimator, the non-fault source is a global absolute locator; when the fault source is a global absolute locator, the non-fault source is a visual motion estimator. The global pose, the absolute pose, and the previous pose are weighted and fused to obtain a fourth fused pose; wherein the weight of the previous pose is the weight by which the fault source weight is reduced.

[0015] According to the present invention, an AGV global positioning controller based on dual-filter arbitration and spatiotemporal verification is provided, wherein the intelligent arbitration center comprises: The spatiotemporal verification module is used for: Spatiotemporal verification is performed on the first fused pose, the second fused pose, the third fused pose, and the fourth fused pose; When the respective verification constraints are met, the first fused pose, the second fused pose, the third fused pose, or the fourth fused pose are output. If the respective verification constraints are not met, perform at least one of the following operations: reduce the confidence of the fault source, start cross-validation, or force entry into safe mode. The verification constraints of the first fused pose include: the positional deviation between the absolute pose and the global pose is less than a first preset deviation; the velocity deviation between the global velocity and the absolute velocity is less than a second preset deviation; the fusion velocity corresponding to the first fused pose is less than the maximum allowed velocity; and the displacement difference between the fused poses of adjacent frames in the first fused pose is less than a third preset deviation. The verification constraints for the second fused pose include: the fusion speed corresponding to the second fused pose is less than the maximum allowable speed; the displacement difference between the fused poses of adjacent frames in the second fused pose is less than the third preset deviation; The verification constraints of the third fused pose include: the fusion speed corresponding to the third fused pose is less than the maximum allowed speed; the displacement difference between the fused poses of adjacent frames in the third fused pose is less than the third preset deviation. The verification constraints of the fourth fused pose include: the fusion speed corresponding to the fourth fused pose is less than the maximum allowed speed; and the displacement difference between the fused poses of adjacent frames in the fourth fused pose is less than the third preset deviation.

[0016] The AGV global positioning controller provided by this invention, based on dual-filter arbitration and spatiotemporal verification, fuses multi-source data from a visual motion estimator and a global absolute positioner. Combined with an intelligent arbitration center that dynamically selects a fusion strategy based on confidence level, it performs weighted fusion under normal conditions and uses velocity interpolation or drift compensation to switch positioning sources under abnormal conditions. This effectively solves the problem of positioning jumps caused by the failure of a single sensor in complex environments. In this way, positioning stability can be improved, smooth switching and anomaly correction can be achieved, thereby improving the accuracy of AGV global positioning. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1This is a schematic diagram of the structure of the AGV global positioning controller based on dual-filter arbitration and spatiotemporal verification provided in an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0020] In existing technologies, AGV positioning mainly relies on a single sensor. Visual positioning methods are susceptible to environmental occlusion and changes in lighting, while ultra-wideband positioning is prone to multipath interference in densely metal areas. Traditional fusion schemes use a single filter for data integration, lacking an effective switching mechanism when a sensor suddenly malfunctions, leading to abrupt changes or failures in positioning results. In complex scenarios such as e-commerce warehouses, temporary obstacles and interference from metal shelves coexist, making it difficult for existing technologies to maintain positioning continuity.

[0021] To address the aforementioned issues, research has revealed that maintaining positioning continuity during sensor failure requires a dynamic multi-source data fusion mechanism. By analyzing the complementary characteristics of vision and ultra-wideband positioning, a dual-filter approach is proposed to process data from different sensors, and a confidence assessment is introduced to achieve dynamic weight allocation. For sudden anomaly scenarios, an arbitration strategy based on spatiotemporal verification is designed to automatically switch the dominant positioning source when a sensor fails, while maintaining a smooth trajectory transition through motion integration or deviation compensation.

[0022] Therefore, the present invention proposes the following technical solution: Figure 1 This is a schematic diagram of the structure of the AGV global positioning controller based on dual-filter arbitration and spatiotemporal verification provided in an embodiment of the present invention. Figure 1 As shown, the AGV global positioning controller based on dual-filter arbitration and spatiotemporal verification in this embodiment may include a visual motion estimator 11, a global absolute locator 12, and an intelligent arbitration center 13.

[0023] In a specific implementation, the visual motion estimator 11 is used to acquire depth camera images, inertial measurement unit data and wheel odometer data, process them, and output the relative pose of the AGV, the relative speed of the AGV and the visual positioning confidence. The global absolute locator 12 is used to acquire ranging values ​​from at least three ultra-wideband base stations and data from the inertial measurement unit, process them, and output the absolute pose of the AGV, the absolute speed of the AGV, and the global positioning confidence. Intelligent arbitration center 13, used for: The relative pose, relative velocity, visual positioning confidence, absolute pose, absolute velocity, and global positioning confidence are obtained. Data synchronization is achieved through hardware timestamps. After converting the relative pose and relative velocity into global pose and global velocity in the global coordinate system, data analysis is performed to obtain data analysis results. When the data analysis results do not meet the preset switching conditions, the global pose and the absolute pose are weighted and fused to obtain the first fused pose. When the data analysis results meet the preset switching conditions, if the preset switching condition is to switch to global positioning-dominated, a velocity-maintaining interpolation strategy is adopted, and the global velocity is used to perform motion integration on the absolute pose to obtain the second fused pose; if the preset switching condition is to switch to visual positioning-dominated, a drift compensation strategy is adopted, and the position deviation compensation amount between the absolute pose and the relative pose is superimposed on the global pose to obtain the third fused pose.

[0024] The visual motion estimator 11 is a module that processes depth images and inertial data using a point cloud registration algorithm. Specifically, it can use an iterative nearest-point algorithm to match continuous frame point clouds, combine inertial data to compensate for motion blur, and output the relative pose, relative speed, and visual positioning reliability of the AGV. The global absolute locator 12 is a module that constructs a position calculation model based on ultra-wideband ranging. Specifically, it can solve overdetermined equations using the least squares method, combine geometric accuracy factors to evaluate positioning reliability, and output the absolute pose, absolute speed, and global positioning reliability of the AGV. The intelligent arbitration center 13 is a decision-making module with multi-source data fusion and anomaly handling capabilities. Specifically, it can use time synchronization technology to align sensor data, trigger positioning source switching based on confidence thresholds, and use motion integration or deviation compensation to maintain pose continuity.

[0025] Specifically, the visual motion estimator 11 performs point cloud registration on the depth image to obtain relative motion parameters, such as the relative pose and relative speed of the AGV. Simultaneously, it evaluates the point cloud matching accuracy and data integrity to generate a confidence index, namely the visual positioning confidence. The global absolute locator 12 uses ultra-wideband ranging values ​​to calculate absolute coordinates, and combines the number and geometric distribution of base stations to evaluate positioning reliability, outputting the absolute pose, absolute speed, and global positioning confidence of the AGV. The intelligent arbitration center 13 receives pose data and confidence from two types of positioning sources in real time. When the visual positioning confidence is detected to be less than a first confidence threshold (e.g., 0.4) and the global positioning confidence is detected to be greater than a second confidence threshold (e.g., 0.6), the ultra-wideband positioning-dominated mode is adopted, maintaining pose updates through global velocity integration. When ultra-wideband positioning is interfered with, when the global positioning confidence is detected to be less than the first confidence threshold (e.g., 0.4) and the visual positioning confidence is detected to be greater than the second confidence threshold (e.g., 0.6), it switches to the visual-dominated mode and superimposes position deviation compensation amounts for absolute and relative poses for compensation. Under other normal operating conditions (i.e., when both sensors are functioning normally), the weights are dynamically allocated based on the confidence level to achieve optimal fusion.

[0026] This invention achieves smooth transitions through motion integration and deviation compensation. Existing single filters cannot distinguish sensor failure types; this invention employs dual filters to process data independently, combined with confidence assessment to achieve rapid anomaly detection. Furthermore, it enhances anti-interference capabilities through dynamic selection of ultra-wideband base stations and geometric accuracy assessment.

[0027] Through the above technical solutions, this invention automatically switches to ultra-wideband positioning mode when the visual sensor is obstructed, using velocity integration to maintain trajectory continuity; and switches to visual positioning mode when the ultra-wideband signal is interfered with by metal, eliminating accumulated errors through deviation compensation. The dual confidence assessment mechanism effectively identifies abnormal sensor states, and spatiotemporal verification constraints ensure that the fused pose conforms to the laws of physical motion, thereby achieving highly robust continuous positioning in complex industrial scenarios.

[0028] In some embodiments, the present invention further proposes an AGV global positioning controller based on dual-filter arbitration and spatiotemporal verification, including a visual motion estimator 11, a global absolute locator 12, and an intelligent arbitration center 13. The intelligent arbitration center 13 includes a first fusion module, which is used to calculate the visual positioning weight and the global positioning weight when the data analysis results do not meet the preset switching conditions, and then sum the weighted products of the global pose and the absolute pose to generate a first fused pose.

[0029] In a specific implementation, visual positioning weight refers to a weighting coefficient dynamically adjusted based on visual positioning confidence, velocity deviation, and pose deviation. Specifically, it can be implemented by combining the product of confidence and velocity deviation with the ratio of the maximum permissible velocity, reflecting the reliability of visual positioning during the fusion process. Global positioning weight refers to a weighting coefficient determined based on global positioning confidence, pose deviation, and the maximum permissible position deviation. Specifically, it can be implemented by combining the ratio of pose deviation to the maximum permissible deviation with confidence adjustment, characterizing the availability of global positioning.

[0030] Specifically, when the AGV is operating normally in the shelf aisle, if both vision and global positioning are in a valid state (i.e., normal condition), the first fusion module achieves the fusion of dual-source positioning data through a dynamic weight allocation mechanism. For example, when the visual positioning confidence is 0.95, the global positioning confidence is 0.88, and the speed deviation is below the threshold, the module calculates a visual positioning weight of 0.92 and a global positioning weight of 0.08, assigning 92% weight to the global pose converted from visual positioning and 8% weight to the absolute pose output by UWB, generating the fused pose through weighted summation. This process continuously monitors speed and pose deviations. When the confidence of the vision sensor drops to 0.35 due to temporary occlusion, the module automatically reduces the visual positioning weight to avoid the impact of low-quality data on the fusion result.

[0031] This invention introduces a two-dimensional weight calculation model for velocity deviation and pose deviation, and dynamically adjusts the fusion ratio based on confidence. For example, when the visual positioning point cloud matching rate decreases but the velocity estimation remains accurate, appropriate visual weights are maintained to participate in the fusion, thus avoiding sudden weight changes due to anomalies in a single indicator.

[0032] Through the above technical solution, this invention can achieve a smooth transition when both visual and global positioning data are valid but local deviations exist, by using dual-dimensional dynamic weight allocation, effectively suppressing the impact of short-term anomalies of a single sensor on the positioning results. For example, when the number of matching points in the visual point cloud temporarily decreases due to the AGV passing through an area of ​​changing lighting, the system maintains the continuity of positioning output by reducing the visual weight rather than completely switching to global positioning, thus avoiding sudden stops or path deviations.

[0033] In some embodiments, the present invention further proposes a weight calculation unit in the intelligent arbitration hub 13, comprising: calculating the magnitude of the velocity deviation between the global velocity and the absolute velocity, and calculating a first quotient of the magnitude of the velocity deviation and the maximum permissible velocity; calculating a first difference between 1 and the first quotient, and calculating a third product of the first difference and the visual positioning confidence; calculating a first sum of the visual positioning confidence and the global positioning confidence, and calculating a second quotient of the third product and the first sum as a visual positioning weight; calculating the magnitude of the pose deviation between the global pose and the absolute pose, and calculating a third quotient of the magnitude of the pose deviation and the maximum permissible position deviation; calculating a second difference between 1 and the third quotient, and calculating a fourth product of the second difference and the global positioning confidence; and calculating a fourth quotient of the fourth product and the first sum as a global positioning weight.

[0034] Specifically, the weight calculation formula can be found in equation (1): (1) in, Indicates visual positioning weight, Indicates global positioning weight. Indicates visual location reliability. Indicates global location confidence. Indicates global speed. Represents absolute velocity. Indicates the maximum permissible speed. Indicates the global pose. Indicates absolute pose. This indicates the maximum permissible positional deviation.

[0035] In a specific implementation, the magnitude of the velocity deviation refers to the Euclidean distance between the global velocity vector and the absolute velocity vector. Specifically, it can be achieved by calculating the square root of the sum of the squares of the differences between the two velocity components, which is used to quantify the degree of difference between the two positioning sources in motion.

[0036] The maximum permissible speed refers to the highest speed threshold that an AGV is allowed to reach in a specific working scenario, such as 1.5 meters per second. When the speed deviation exceeds this threshold, a weight adjustment mechanism is triggered.

[0037] The magnitude of pose deviation refers to the geometric distance between the position and attitude outputs of two positioning sources in the global coordinate system. Specifically, it can be calculated by taking the square root of the sum of the squares of the three-dimensional coordinate differences, and is used to measure the spatial consistency of the positioning results.

[0038] The maximum permissible position deviation refers to the maximum positional difference between two positioning sources that the system allows. For example, it can be 0.3 meters. If this value is exceeded, the positioning is judged to be abnormal.

[0039] Visual positioning reliability refers to the reliability score of the positioning result output by the visual motion estimator 11, which can be calculated by comprehensively considering the point cloud matching accuracy, the number of effective feature points, and the degree of motion ambiguity.

[0040] Global positioning reliability refers to the reliability score of the positioning result output by the global absolute locator 12, which can be calculated by the number of visible base stations and the geometric precision factor.

[0041] Specifically, when the AGV enters a metal interference area, the ranging error of the global absolute locator 12 may lead to an increase in pose deviation. At this time, the weight calculation unit generates adaptive weights by calculating the speed deviation and pose deviation in real time, combined with the dynamic changes in visual positioning confidence and global positioning confidence. For example, when the pose deviation of UWB positioning reaches 0.25 meters due to metal reflection, if the maximum allowable position deviation is set to 0.3 meters, the third quotient is 0.25 / 0.3≈0.83, and the second difference is 1-0.83=0.17. Assuming the global positioning confidence is 0.65, the fourth product is 0.17×0.65=0.11. When the visual positioning confidence is 0.88 and the first sum is 0.88+0.65=1.53, the final global positioning weight is 0.11 / 1.53≈0.07. Thus, the global positioning weight is significantly reduced, avoiding the impact of abnormal positioning data on the fusion results.

[0042] This invention introduces a dual bias assessment of velocity and pose, combined with dynamic adjustment of confidence, to more accurately identify sensor anomalies. For example, when the visual positioning confidence decreases due to a drop in point cloud matching rate, if the pose bias is still less than a threshold, a higher weight is retained through pose bias calculation, avoiding over-reliance on UWB positioning.

[0043] Through the above technical solution, this invention can achieve precise weight allocation by quantifying motion state deviation and spatial position deviation when local performance fluctuations occur in the vision and UWB positioning sources. For example, when the AGV passes through a metal shelf area, even if the UWB positioning confidence drops to 0.6 due to multipath interference, as long as its pose deviation does not exceed the maximum allowable value, it can still maintain a reasonable weight for fusion, avoiding jumps in positioning results. Simultaneously, when the vision sensor encounters temporary occlusion, dynamic evaluation of speed deviation maintains positioning continuity, ensuring stable operation of the AGV in complex industrial scenarios.

[0044] In some embodiments, the present invention further proposes that the intelligent arbitration center 13 includes a second fusion module, which is used to decompose the global velocity into global linear velocity and global angular velocity, and to perform motion integration on the absolute pose with the control cycle of the intelligent arbitration center 13 as the step size to generate a temporary fused pose, and to perform smoothing processing on the temporary fused pose through a first-order low-pass filter to output the second fused pose.

[0045] Global velocity decomposition refers to decomposing the AGV's motion into linear and rotational motion components. This can be achieved using orthogonal decomposition of velocity vectors in Cartesian coordinates, facilitating the independent handling of the effects of translation and rotation on pose. Motion integration refers to calculating future pose based on current velocity, which can be implemented using Euler integration or Runge-Kutta methods, approximating continuous motion using discrete time steps. First-order low-pass filtering is a signal processing technique to suppress high-frequency noise, specifically implemented using a Butterworth filter with adjustable cutoff frequency, controlling the smoothing degree by adjusting the time constant.

[0046] Specifically, when the confidence level of visual positioning drops sharply due to occlusion, the system switches to a global positioning-dominated mode. The second fusion module decomposes the global velocity into linear velocity and angular velocity. Then, a first-order low-pass filter with a cutoff frequency of 10Hz is used to smooth the temporary fused pose, eliminating the pose abrupt changes caused by UWB ranging jumps, and finally outputting a smooth and continuous second fused pose.

[0047] Traditional solutions typically use the raw pose data from the new sensor directly when switching sensors, which can easily cause the AGV to stop abruptly due to data jumps. This invention maintains speed continuity through motion integration and eliminates high-frequency interference through low-pass filtering, ensuring a smooth transition in pose output even when the UWB signal in the metal shelf area is affected by multipath effects.

[0048] Through the above technical solution, the present invention effectively solves the problem of pose jump during sensor switching. When visual positioning fails or UWB positioning is interfered with, the pose transition is achieved through speed maintenance and motion prediction, avoiding the AGV from triggering an emergency stop protection mechanism due to sudden positioning changes, thus ensuring the continuity of logistics transportation tasks.

[0049] In some embodiments, the present invention further proposes that the intelligent arbitration center 13 includes a third fusion module, which is used to calculate the current position deviation between the absolute pose and the relative pose, calculate the fifth product of the current position deviation and the preset compensation coefficient, calculate the second sum of the global pose and the fifth product, and obtain the third fused pose.

[0050] The current position deviation refers to the difference in position between the absolute pose output by the global absolute locator 12 and the relative pose output by the visual motion estimator 11 at the same moment. Specifically, it can be calculated using Euclidean distance to determine the coordinate difference between the two in three-dimensional space, thus quantifying the degree of deviation between visual positioning and global positioning. The compensation coefficient is a proportional factor used to adjust the amount of position deviation compensation. It can be implemented using a preset fixed value or a variable dynamically adjusted according to the environment, balancing the relationship between compensation intensity and system stability to avoid over-compensation leading to sudden pose changes.

[0051] Specifically, when the intelligent arbitration center 13 determines that a switch to vision-based positioning is necessary, the third fusion module first acquires the current absolute and relative poses, unifies them to the same coordinate system through coordinate transformation, and then calculates their three-dimensional position deviation. This position deviation is multiplied by a preset compensation coefficient, for example, a compensation coefficient can be set to 0.5 to compromise between compensation effect and system stability. The product result is then superimposed on the global pose output by the visual motion estimator 11. The resulting third fused pose retains the high accuracy of vision positioning while suppressing the cumulative drift error that may exist in vision positioning by introducing the absolute position deviation compensation amount of global positioning, thus achieving a smooth transition between the two positioning modes.

[0052] This invention introduces a dynamic compensation mechanism based on position deviation to actively correct the drift of visual positioning during the switching process. For example, when the visual sensor recovers from occlusion, the visual pose is calibrated using the absolute position reference provided by global positioning, effectively eliminating the problem of discontinuous positioning at the moment of switching.

[0053] Through the above technical solution, the present invention can compensate for cumulative errors caused by temporary sensor failures or environmental interference in real time under vision-dominated mode, significantly improving the robustness of the positioning system in complex industrial scenarios. For example, when the visual positioning recovers after the AGV leaves the metal interference area, the visual drift generated during the UWB multipath effect can be quickly eliminated by superimposing the compensation amount, avoiding path tracking deviation caused by sudden positioning changes.

[0054] In some embodiments, the present invention further proposes a third fusion module, which is also used to determine the rate of change of position deviation based on the current position deviation and the previous position deviation, determine an initial compensation coefficient based on the correlation between position deviation and compensation coefficient, and determine the compensation adjustment ratio corresponding to the rate of change of position deviation based on the correlation between the rate of change of position deviation and compensation adjustment ratio, so as to adjust the initial compensation coefficient and obtain the preset compensation coefficient.

[0055] The position deviation change rate refers to the change in position deviation between absolute and relative poses per unit time. Specifically, it can be calculated by dividing the position deviation difference between adjacent control cycles by the control cycle duration, and is used to quantify the abrupt change in environmental interference or sensor error. The compensation adjustment ratio refers to the proportional factor that dynamically adjusts the compensation coefficient based on the position deviation change rate. Specifically, it can be implemented using a predefined change rate-adjustment ratio mapping table or a linear function relationship, and is used to enhance the compensation response speed during sudden environmental changes.

[0056] Specifically, when the AGV enters a metal interference area, causing ranging errors in UWB positioning, the third fusion module first calculates the position deviation between the current absolute pose and the visual relative pose, and subtracts it from the position deviation of the previous control cycle to obtain the position deviation change rate. Then, it queries the preset position deviation-compensation coefficient relationship curve to obtain the initial compensation coefficient; for example, when the position deviation exceeds 0.1 meters, the initial compensation coefficient is set to 0.8. This coefficient is further adjusted according to the position deviation change rate. If the change rate exceeds 0.05 meters per second, the compensation coefficient is increased to 0.9 through a compensation adjustment ratio, thereby enhancing the drift compensation force when the positioning system is interfered with. Finally, the adjusted compensation coefficient is multiplied by the position deviation to obtain the compensation amount, which is then superimposed on the global pose output fusion result.

[0057] This invention introduces a dynamic adjustment compensation coefficient based on the rate of change of position deviation, which can respond in real time to sudden changes in sensor error. For example, when an AGV suddenly enters a metal shelf area, causing a jump in UWB ranging, the positioning drift can be quickly suppressed by increasing the compensation coefficient.

[0058] Through the above technical solution, the present invention effectively solves the problem of positioning jump caused by sudden sensor abnormalities in complex industrial environments. It achieves a smooth transition during the switching process between vision and UWB positioning modes through a dynamic compensation mechanism, avoiding sudden stops or path deviations of AGVs caused by discontinuous positioning.

[0059] In some embodiments, the present invention further proposes a visual motion estimator 11 to generate a 3D point cloud after median filtering preprocessing of depth camera images, and to register two consecutive frames of point clouds using an iterative nearest-point algorithm to solve for the optimal rotation matrix and translation vector, so as to output the relative pose, local relative velocity, and visual positioning reliability of the AGV. The process of determining the visual positioning reliability includes: determining the quotient of the expected accuracy and the accuracy of the relative pose; determining the quotient of the number of valid corresponding points successfully matched in the target point cloud and the total number of points participating in the registration in the source point cloud; the target point cloud is the point cloud of the previous time step in the 3D point cloud, and the source point cloud is the point cloud of the current time step in the 3D point cloud; using a two-dimensional evaluation logic that combines image features and motion state assistance, the image feature score and motion state correction score are calculated respectively, and the image feature score and motion state correction score are fused to obtain the motion blur score; the quotient is calculated, and the quotient and motion blur score are weighted and summed to obtain the visual positioning reliability.

[0060] In a specific implementation process, the formula for calculating visual location reliability can be found in equations (2) and (3): (2) (3) This indicates that a valid corresponding point was successfully found in the target point cloud; This represents the total number of points in the source point cloud that participated in ICP registration; It represents the motion blur score, a metric that quantifies the quality of depth images; Weighting coefficients are used to balance the importance of accuracy, completeness, and data quality in the overall confidence score; the parameters are determined by manual tuning after extensive testing and data statistics in actual working scenarios. Expected weight error; Optimal rotation matrix; Translation vector; Source point cloud, point cloud; Point cloud within a 3D point cloud; In a specific implementation, median filtering preprocessing refers to noise suppression of the raw images acquired by the depth camera. This can be achieved by using a sliding window to sort depth values ​​within the pixel neighborhood and taking the median, effectively eliminating the impact of impulse noise on point cloud quality. The iterative nearest-neighbor algorithm solves for the optimal rigid body transformation parameters by minimizing the distance between corresponding points in two consecutive point cloud frames. This can be achieved by using a KD-tree to accelerate the nearest neighbor search and setting a convergence threshold, used to accurately calculate the relative motion of the AGV. Visual positioning reliability is an indicator that quantifies the reliability of visual positioning results. It can be achieved by weighting and calculating three factors: registration error, effective matching point ratio, and motion blur score, used to dynamically evaluate the reliability of the visual positioning system. Motion blur score is a quantitative value reflecting the degree of blurring in the depth image due to AGV motion. This can be achieved using edge gradient analysis or high-frequency component energy detection methods, used to identify image quality degradation caused by rapid motion.

[0061] Specifically, the visual motion estimator 11 first performs median filtering on the raw images captured by the depth camera to remove noise points caused by environmental interference, generating high-quality 3D point cloud data. Next, it registers the point clouds of the current frame with those of the previous frame using an iterative nearest-neighbor algorithm, accelerates the corresponding point search process using a KD-tree, and iteratively optimizes the rotation matrix and translation vector until convergence, finally outputting the relative pose change of the AGV. When calculating the visual positioning reliability, the statistical characteristics of the registration error, the proportion of effective matching points, and the impact of motion blur on image quality are considered simultaneously. For example, when the AGV operates in a turning area with drastic lighting changes, the depth camera may experience increased point cloud registration error due to motion blur, significantly increasing the motion blur score. Weighted calculations can dynamically reduce the visual positioning reliability, triggering a subsequent positioning mode switching mechanism.

[0062] This invention innovatively introduces motion blur scoring as a dimension, combining image feature analysis and motion state correction to form a multi-factor fusion confidence assessment system. For example, when temporary cardboard box obstruction causes a decrease in the proportion of effective matching points in the point cloud, if the AGV is in a low-speed linear motion state and there is no motion blur, the system can still maintain a high confidence level, avoiding unnecessary positioning mode switching.

[0063] Through the above technical solutions, this invention can achieve accurate assessment of visual positioning reliability in complex industrial scenarios. For example, when the AGV is temporarily obstructed by a cardboard box, causing 50% of the point cloud to fail, the system accurately calculates the decrease in visual positioning reliability by detecting the decrease in the proportion of effective matching points and the increase in registration error, combined with the feature that the motion blur score does not increase significantly. This provides a decision-making basis for the smooth switching of the subsequent arbitration center. At the same time, when UWB positioning fails in the metal shelf area due to multipath interference, the visual motion estimator 11 suppresses noise through median filtering, maintains stable pose output capability, and ensures the continuous operation of the AGV in the hybrid positioning mode.

[0064] In some embodiments, the present invention further proposes a global absolute locator 12 that detects outliers by historical rate of change, constructs an overdetermined system of equations based on base station coordinates, solves the overdetermined system of equations by least squares method to obtain the absolute coordinates of the AGV, and outputs the absolute pose, absolute speed and global positioning confidence of the AGV; wherein, the process of determining the global positioning confidence includes: determining the seventh quotient of the number of base stations with successful ranging and 3, and determining the reciprocal of the geometric precision factor; and performing a weighted summation of the seventh quotient and the reciprocal to obtain the global positioning confidence.

[0065] In a specific implementation, the overdetermined system of equations is as follows: ; By minimizing the sum of squared residuals The optimal solution is obtained, which effectively suppresses the ranging error of a single base station.

[0066] in, These are the absolute coordinates of the AGV. The absolute coordinates of the three base stations; , , The distance to the base station; The formula for calculating global location reliability can be found in equation (4): (4) This indicates the number of visible base stations, and the number of base stations that have been successfully measured so far. Then set directly They believe that UWB positioning is invalid.

[0067] GDOP, or Geometric Precision Factor, measures the impact of base station geometry on accuracy. Generally, increasing the number of base stations can reduce GDOP, while avoiding base stations being on the same line should be avoided.

[0068] The weighting coefficient is used to balance the importance of the number of base stations and geometric accuracy in the overall confidence level. The parameters are determined by manual adjustment after a large number of tests and data statistics in actual working scenarios.

[0069] In a specific implementation process, historical rate of change detection of outliers refers to identifying abrupt changes in ranging data that exceed a reasonable range by analyzing the trend of ranging data changes over multiple consecutive time periods. This can be achieved using a sliding window statistical method, such as calculating the standard deviation of the last five ranging values; if the deviation of the current ranging value from the mean exceeds three times the standard deviation, it is considered an outlier. Overdetermined equations refer to a system of equations where the number of equations exceeds the number of unknown variables. This can be constructed using redundant base station ranging data, for example, establishing four distance equations to solve for two-dimensional coordinates in an environment with four base stations. Least squares is a mathematical method that finds the optimal solution by minimizing the sum of squared residuals. This can be implemented using a Gauss-Newton iterative algorithm to suppress the impact of a single base station's ranging error on the overall positioning. Geometric precision factor (GDOP) is a quantitative indicator reflecting the impact of the spatial distribution of base stations on positioning accuracy. It can be calculated using the covariance matrix of the base station coordinates; for example, the GDOP value is lower when the base stations are arranged in an equilateral triangle. Manual parameter tuning refers to adjusting the weighting coefficients based on test data from the actual scenario. For example, in areas with severe metallic interference, increasing the weight of ω2 can enhance the contribution of the geometric precision factor to the confidence level.

[0070] Specifically, when the AGV enters the metal shelf area, the global absolute locator 12 first detects outliers based on historical ranging data. For example, if the ranging value of a certain base station suddenly increases by 1.2 meters within three consecutive cycles, it is determined to be abnormal data caused by multipath interference and is excluded. Subsequently, an overdetermined system of equations is constructed using the ranging values ​​of the remaining three base stations, and the absolute coordinates of the AGV are solved using the least squares method. In this process, minimizing the sum of squared residuals can reduce the impact of errors from a single base station. At the same time, the number of currently visible base stations and the geometric precision factor are calculated. For example, when N_ANCORS=3 and GDOP=2.1, the calculation is based on preset weights. =0.6、 =0.4 yields a global positioning confidence score of 0.65. If only valid signals from two base stations are received at a given time, the confidence score is immediately reset to zero, triggering a positioning mode switch.

[0071] This invention actively eliminates abnormal base stations by detecting historical rate of change and dynamically evaluates positioning reliability by combining geometric accuracy factor. For example, when a sudden change in the ranging value of a base station is detected, it is excluded and the equation system is reconstructed. Compared with the fixed base station selection method, this invention can improve positioning stability in metal interference environments.

[0072] Through the above technical solution, this invention can achieve reliable global positioning in complex industrial scenarios. For example, in the metal shelf aisles of an e-commerce warehouse, when some UWB base stations are affected by multipath interference, the faulty base stations are eliminated through an anomaly detection mechanism, and the remaining base station data is used to maintain effective positioning; at the same time, the confidence level is dynamically adjusted by combining geometric layout evaluation, providing an accurate basis for subsequent arbitration decisions and avoiding positioning jumps caused by the failure of a single sensor.

[0073] In some embodiments, the present invention further proposes an implementation of an AGV global positioning controller based on dual-filter arbitration and spatiotemporal verification. When the data analysis result meets the preset switching conditions and the switching conditions are predictive switching, the weight of the fault source is reduced to a preset ratio, and the previous pose of the non-fault source is preloaded. Wherein, when the fault source is the visual motion estimator 11, the non-fault source is the global absolute locator 12; when the fault source is the global absolute locator 12, the non-fault source is the visual motion estimator 11; the global pose, absolute pose, and previous pose are weighted and fused to obtain a fourth fused pose; wherein, the weight of the previous pose is the weight of the reduced fault source weight.

[0074] Predictive switching refers to triggering switching logic in advance before the sensor completely fails by monitoring the changing trend of sensor confidence. For example, when the global positioning confidence is detected to be decreasing at a rate of 7% per second, switching preparation can be started 200 milliseconds in advance. The preset proportion for reducing the weight of the fault source can be 70% of the original weight, for example, adjusting it to 0.49 when the original weight is 0.7. Preloading the previous pose of the non-fault source means caching the last valid positioning data of the non-fault source before switching for later use. For example, when the global absolute locator 12 is about to fail, the pose data of the visual motion estimator 11 in the previous control cycle can be saved in advance. During the weighted fusion process, the weight allocation of the previous pose can be the proportion of the reduced weight of the fault source. For example, after the original fault source weight of 0.7 is reduced to 0.21, 0.21 is allocated to the previous pose.

[0075] Specifically, when the system detects a continuous downward trend in the confidence level of a certain sensor, such as the global absolute positioner 12 decreasing at a rate of 7% per second, the intelligent arbitration center 13 will initiate a predictive switching mechanism before the sensor completely fails. First, the weight of the fault source is reduced by a preset ratio, for example, from 0.7 to 0.49, while simultaneously extracting the effective pose data from the previous control cycle from the historical data of non-fault sources. Then, the current global pose, absolute pose, and pre-loaded pose from the previous moment are fused according to the adjusted weights, for example, using a weighted summation of 0.49 × global pose + 0.3 × absolute pose + 0.21 × previous pose. This process, by gradually shifting weights, avoids abrupt changes in positioning data, ensuring the AGV maintains motion continuity in high-risk scenarios such as metal interference areas.

[0076] In some specific implementations, the trigger threshold for predictive switching can be set to a confidence rate change exceeding 5% per second, and the preloaded historical data time window can be set to the effective poses within the most recent three control cycles. The preset ratio during the weight adjustment process can be dynamically configured according to the sensor type; for example, a linear decay strategy is used when a visual sensor fails, while an exponential decay strategy is used when a UWB sensor fails.

[0077] This invention uses a predictive switching mechanism to initiate transition measures in the early stages of sensor performance degradation. Combined with the smooth introduction of historical pose data, this allows the AGV to complete a significant weight shift before entering the interference area, effectively improving the continuity of the positioning output trajectory.

[0078] Through the above technical solution, this invention effectively solves the problem of abrupt positioning changes caused by gradual sensor performance failure. In environments with metallic interference, it can control the fluctuation range of AGV pose output within a specified value, while ensuring the continuity of control commands and avoiding the risk of cargo tipping over due to sudden stops. This solution is particularly suitable for warehousing environments with intermittent signal interference, enabling autonomous fault tolerance of the positioning system without relying on additional hardware.

[0079] In some embodiments, the present invention further proposes an intelligent arbitration hub 13 including a spatiotemporal verification module. The spatiotemporal verification module is used to perform spatiotemporal verification on the first fused pose, the second fused pose, the third fused pose, and the fourth fused pose; output the corresponding fused pose when the respective verification constraints are met; and perform at least one of the following operations when the respective verification constraints are not met: reducing the confidence of the fault source, initiating cross-validation, or forcibly entering a safe mode. The verification constraints of the first fused pose include that the positional deviation between the absolute pose and the global pose is less than a first preset deviation, the velocity deviation between the global velocity and the absolute velocity is less than a second preset deviation, the fused velocity corresponding to the first fused pose is less than the maximum allowable velocity, and the displacement difference between adjacent frame fused poses in the first fused pose is less than a third preset deviation. The verification constraints of the second fused pose include that the corresponding fused velocity is less than the maximum allowable velocity, and the displacement difference between adjacent frame fused poses is less than a third preset deviation. The verification constraints of the third fused pose include that the corresponding fused velocity is less than the maximum allowable velocity, and the displacement difference between adjacent frame fused poses is less than a third preset deviation. The verification constraints of the fourth fused pose include that the corresponding fused velocity is less than the maximum allowable velocity, and the displacement difference between adjacent frame fused poses is less than a third preset deviation.

[0080] Among these, positional deviation refers to the magnitude of the difference in coordinates output by two different positioning methods, specifically calculated using Euclidean distance, used to detect significant spatial discrepancies between visual positioning and UWB positioning. Velocity deviation refers to the magnitude of the difference in motion vectors output by different sensors, specifically calculated using vector subtraction, used to determine the consistency of data between the inertial measurement unit and the wheeled odometer. Displacement difference refers to the magnitude of pose change between adjacent time points, specifically calculated using time series differencing, used to detect abrupt changes in positioning results. Maximum permissible speed refers to the highest allowable movement rate of the AGV drive system, specifically set according to motor performance parameters, serving as a physical limit constraint. Cross-validation refers to simultaneously calling backup sensor data for secondary calculations, specifically implemented using redundant positioning modules running in parallel, used to quickly verify data reliability under abnormal conditions.

[0081] Specifically, after receiving four fused poses, the spatiotemporal verification module first performs a four-fold constraint check on the first fused pose: calculating whether the coordinate difference between visual positioning and UWB positioning exceeds a preset threshold; if it does, a sensor failure risk is identified; comparing the vector difference between global velocity and absolute velocity to see if it is within the allowable range; if it exceeds the range, it indicates a contradiction in the motion state; checking whether the fused velocity exceeds the maximum allowable value of the AGV's mechanical structure to prevent control commands from exceeding limits; and analyzing the smoothness of pose changes between adjacent time moments to detect any data jumps. When all constraints are met, the first fused pose is deemed valid and output to the control system. If any constraint is not met, a confidence adjustment mechanism is triggered, such as automatically reducing the global positioning weight when metal interference causes UWB data anomalies, and simultaneously activating the visual positioning module for cross-validation. For the second, third, and fourth fused poses, the verification process focuses on velocity limits and displacement continuity, preventing sudden changes in control commands by monitoring the motion state in real time. When verification fails, a safety mode is activated to prevent AGV collisions.

[0082] This invention constructs a multi-dimensional spatiotemporal joint verification mechanism, which not only covers multiple factors such as spatial consistency, motion continuity, and physical limits, but also establishes a dynamic fault response strategy, enabling rapid switching to a backup solution when sensor data is abnormal. For example, when visual positioning fails due to occlusion, displacement difference detection can detect the anomaly in advance and initiate a smooth transition.

[0083] Through the above technical solution, this invention effectively solves the problem of sudden positioning changes caused by sensor failure, and can maintain the continuous and stable operation of AGVs even under complex working conditions such as visual occlusion and metal interference. For example, when a temporary cardboard box obstructs the camera, causing a sudden drop in visual positioning reliability, the system can not only quickly switch to UWB-dominated mode, but also prevent pose jumps caused by UWB multipath errors through displacement difference verification, ensuring that the AGV maintains a smooth movement trajectory in the shelf aisle. At the same time, the independent verification mechanism of the four-fold fused pose provides targeted anomaly detection capabilities for different working modes, improving the overall robustness of the system.

[0084] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0085] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An AGV global positioning controller based on dual-filter arbitration and spatiotemporal verification, characterized in that, include: The visual motion estimator is used to acquire depth camera images, inertial measurement unit data, and wheel odometer data, process them, and output the relative pose, relative speed, and visual positioning reliability of the AGV. A global absolute positioner is used to acquire ranging values ​​from at least three ultra-wideband base stations and data from the inertial measurement unit, process them, and output the absolute pose of the AGV, the absolute speed of the AGV, and the global positioning confidence level. Intelligent arbitration center, used for: The relative pose, relative velocity, visual positioning confidence, absolute pose, absolute velocity, and global positioning confidence are obtained. Data synchronization is achieved through hardware timestamps. After converting the relative pose and relative velocity into global pose and global velocity in the global coordinate system, data analysis is performed to obtain data analysis results. When the data analysis results do not meet the preset switching conditions, the global pose and the absolute pose are weighted and fused to obtain the first fused pose. When the data analysis results meet the preset switching conditions, if the preset switching condition is to switch to global positioning-dominated, a velocity-maintaining interpolation strategy is adopted, and the global velocity is used to perform motion integration on the absolute pose to obtain the second fused pose; if the preset switching condition is to switch to visual positioning-dominated, a drift compensation strategy is adopted, and the position deviation compensation amount between the absolute pose and the relative pose is superimposed on the global pose to obtain the third fused pose.

2. The AGV global positioning controller based on dual-filter arbitration and spatiotemporal verification according to claim 1, characterized in that, The intelligent arbitration center includes: The first fusion module is used for: When the data analysis results do not meet the preset switching conditions, calculate the visual positioning weight and the global positioning weight; The first product value is obtained by multiplying the visual localization weights with the global pose. The second product value is obtained by multiplying the global positioning weights and the absolute pose. The first fused pose is obtained by summing the first product value and the second product value.

3. The AGV global positioning controller based on dual-filter arbitration and spatiotemporal verification according to claim 2, characterized in that, The first fusion module includes: The weight calculation unit is used for: Calculate the magnitude of the velocity deviation between the global velocity and the absolute velocity, and calculate a first quotient of the magnitude of the velocity deviation and the maximum permissible velocity; calculate a first difference between 1 and the first quotient, and calculate a third product of the first difference and the visual positioning confidence; calculate a first sum of the visual positioning confidence and the global positioning confidence, and calculate a second quotient of the third product and the first sum as the visual positioning weight; Calculate the magnitude of the pose deviation between the global pose and the absolute pose, and calculate the third quotient of the magnitude of the pose deviation and the maximum allowable position deviation; calculate the second difference between 1 and the third quotient, and calculate the fourth product of the second difference and the global positioning confidence; calculate the fourth quotient of the fourth product and the first sum as the global positioning weight.

4. The AGV global positioning controller based on dual-filter arbitration and spatiotemporal verification according to claim 1, characterized in that, The intelligent arbitration center includes: The second fusion module is used for: The global velocity is decomposed into global linear velocity and global angular velocity, and motion integration is performed on the absolute pose with the intelligent arbitration center control cycle as the step size to generate a temporary fused pose. The temporary fused pose is smoothed by a first-order low-pass filter, and the second fused pose is output.

5. The AGV global positioning controller based on dual-filter arbitration and spatiotemporal verification according to claim 1, characterized in that, The intelligent arbitration center includes: The third fusion module is used for: Calculate the current position deviation between the absolute pose and the relative pose; Calculate the fifth product of the current position deviation and the preset compensation coefficient; The third fused pose is obtained by calculating the second sum of the global pose and the fifth product value.

6. The AGV global positioning controller based on dual-filter arbitration and spatiotemporal verification according to claim 5, characterized in that, The third fusion module is also used for: The rate of change of position deviation is determined based on the current position deviation and the previous position deviation; Based on the correlation between position deviation and compensation coefficient, the initial compensation coefficient is determined; Based on the correlation between the rate of change of position deviation and the compensation adjustment ratio, the compensation adjustment ratio corresponding to the rate of change of position deviation is determined, so as to adjust the initial compensation coefficient and obtain the preset compensation coefficient.

7. The AGV global positioning controller based on dual-filter arbitration and spatiotemporal verification according to claim 1, characterized in that, The visual motion estimator is specifically used for: After median filtering preprocessing of the depth camera image, a 3D point cloud is generated. The iterative nearest point algorithm is used to register two consecutive frames of point cloud, and the optimal rotation matrix and translation vector are solved to output the relative pose of the AGV, the local relative velocity of the AGV, and the visual positioning confidence. The process of determining the visual location confidence includes: Determine a fifth quotient between the desired accuracy and the accuracy of the relative pose; Determine the sixth quotient between the number of valid matching points in the 3D point cloud and the total number of points in the source point cloud that participated in the registration; A two-dimensional evaluation logic combining image features and motion state assistance is adopted to calculate the image feature score and the motion state correction score respectively, and then the image feature score and the motion state correction score are fused to obtain the motion blur score; The visual location reliability is obtained by weighted summation of the fifth quotient, the sixth quotient, and the motion blur score.

8. The AGV global positioning controller based on dual-filter arbitration and spatiotemporal verification according to claim 1, characterized in that, The global absolute locator is specifically used for: Anomalies are detected by historical rate of change. An overdetermined system of equations is constructed based on base station coordinates. The overdetermined system of equations is solved by the least squares method to obtain the absolute coordinates of the AGV, so as to output the absolute pose, absolute speed and global positioning confidence of the AGV. The process of determining the global location confidence includes: Determine the seventh quotient of the number of base stations that have successfully measured distances and 3, and determine the reciprocal of the geometric precision factor; The global location confidence is obtained by weighted summation of the seventh quotient and the reciprocal.

9. The AGV global positioning controller based on dual-filter arbitration and spatiotemporal verification according to any one of claims 1-8, characterized in that, The intelligent arbitration center is also used for: When the data analysis results meet the preset switching conditions, if the preset switching conditions are predictive switching, the weight of the fault source is reduced to a preset ratio, and the pose of the non-fault source at the previous moment is preloaded; wherein, when the fault source is a visual motion estimator, the non-fault source is a global absolute locator; when the fault source is a global absolute locator, the non-fault source is a visual motion estimator. The global pose, the absolute pose, and the previous pose are weighted and fused to obtain a fourth fused pose; wherein the weight of the previous pose is the weight by which the fault source weight is reduced.

10. The AGV global positioning controller based on dual-filter arbitration and spatiotemporal verification according to claim 9, characterized in that, The intelligent arbitration center includes: The spatiotemporal verification module is used for: Spatiotemporal verification is performed on the first fused pose, the second fused pose, the third fused pose, and the fourth fused pose; When the respective verification constraints are met, the first fused pose, the second fused pose, the third fused pose, or the fourth fused pose is output; If the respective verification constraints are not met, perform at least one of the following operations: reduce the confidence of the fault source, start cross-validation, or force entry into safe mode. The verification constraints of the first fused pose include: the positional deviation between the absolute pose and the global pose is less than a first preset deviation; the velocity deviation between the global velocity and the absolute velocity is less than a second preset deviation; the fusion velocity corresponding to the first fused pose is less than the maximum allowed velocity; and the displacement difference between the fused poses of adjacent frames in the first fused pose is less than a third preset deviation. The verification constraints for the second fused pose include: the fusion speed corresponding to the second fused pose is less than the maximum allowable speed; the displacement difference between the fused poses of adjacent frames in the second fused pose is less than the third preset deviation; The verification constraints of the third fused pose include: the fusion speed corresponding to the third fused pose is less than the maximum allowed speed; the displacement difference between the fused poses of adjacent frames in the third fused pose is less than the third preset deviation. The verification constraints of the fourth fused pose include: the fusion speed corresponding to the fourth fused pose is less than the maximum allowed speed; and the displacement difference between the fused poses of adjacent frames in the fourth fused pose is less than the third preset deviation.

Images