Warehouse automatic material taking oriented AGV positioning method and system

Abstract

The application relates to the technical field of positioning, in particular to an AGV trolley positioning method and system for automatic material taking in a warehouse, which comprises the following steps: acquiring environment sensing data and inertial measurement data of an AGV trolley during operation; performing abnormality detection on the inertial measurement data to determine the motion state credibility of the inertial measurement data; determining calibration parameters according to the cumulative error of predicting the environment sensing data by using the inertial measurement data in a target time period; adjusting the motion state credibility based on the calibration parameters to obtain adjusted motion state credibility; dynamically adjusting the noise covariance parameters of an extended Kalman filter used for data fusion according to the adjusted motion state credibility; and performing fusion processing on the environment sensing data and the inertial measurement data by using the adjusted extended Kalman filter to output the positioning information of the AGV trolley. The application improves the positioning precision of the AGV trolley in a complex warehouse environment.

Description

Technical Field

[0001] This invention relates to the field of positioning technology, specifically to a positioning method and system for AGV (Automated Guided Vehicle) carts used for automated material handling in warehouses. Background Technology

[0002] In intelligent warehousing and logistics systems, precise positioning of Automated Guided Vehicles (AGVs) is fundamental for achieving automated goods handling, route planning, and multi-vehicle scheduling. These systems typically rely on sensors such as LiDAR to acquire environmental perception data, combined with high-frequency motion data provided by Inertial Measurement Units (IMUs), to achieve continuous positioning through data fusion algorithms. However, warehouse environments are characterized by narrow aisles, dense shelving, and numerous dynamic obstacles, placing high demands on the real-time performance, robustness, and accuracy of the AGV positioning system.

[0003] Currently, the industry commonly adopts a fusion framework based on the Extended Kalman Filter (EKF) to combine the advantages of low-frequency environmental perception data and high-frequency inertial measurement data. However, in practical applications, this filter relies on preset fixed noise parameters, which are often set based on ideal working conditions. These parameters cannot adapt to the dynamic changes in the motion state of AGVs in complex and ever-changing warehouse scenarios, leading to the accumulation of positioning errors and insufficient overall positioning accuracy. Summary of the Invention

[0004] To address the technical problem of insufficient positioning accuracy of AGVs in complex warehouse environments, the present invention aims to provide an AGV positioning method and system for automated material handling in warehouses. The specific technical solution adopted is as follows:

[0005] Firstly, a method for AGV (Automated Guided Vehicle) positioning in warehouse automated material handling is provided. This method includes: acquiring environmental perception data and inertial measurement data of the AGV during its operation, wherein the sampling frequency of the inertial measurement data is higher than that of the environmental perception data; performing anomaly detection on the inertial measurement data to determine the motion state reliability of the inertial measurement data, whereby the motion state reliability characterizes the degree to which the inertial measurement data reflects the true motion state of the AGV; determining calibration parameters based on the inertial measurement data within a target time period, wherein the start and end times of the target time period are the previous sampling time and the current time of the environmental perception data; adjusting the motion state reliability based on the calibration parameters to obtain the adjusted motion state reliability; dynamically adjusting the noise covariance parameter of the extended Kalman filter used for data fusion based on the adjusted motion state reliability; and fusing the environmental perception data and inertial measurement data using the adjusted extended Kalman filter to output the positioning information of the AGV.

[0006] In one possible design, anomaly detection is performed on the inertial measurement data to determine the reliability of the motion state. This includes: performing anomaly detection on the data in each dimension of the inertial measurement data to identify anomalous data points in each dimension. The inertial measurement data includes position, acceleration, and angle dimensions. Based on the timestamp information of the anomalous data points in each dimension, the anomaly frequency and synchronization frequency are determined. The anomaly frequency is the frequency of anomalous data points occurring within a first historical time period, and the synchronization frequency is the frequency of anomalous data points occurring synchronously in each dimension within the first historical time period. Based on the anomaly frequency and synchronization frequency, the reliability of the motion state is determined.

[0007] In one possible design, the abnormal frequency and synchronization frequency are determined based on the timestamp information of the abnormal data points in each dimension, including: determining the number of abnormal times in the union of all sets of time sets corresponding to the abnormal data points in each dimension, and determining the number of synchronization abnormal times in the intersection of all sets; determining the abnormal frequency based on the number of abnormal times and the number of sampling times included in the first historical duration, and determining the synchronization frequency based on the number of synchronization abnormal times and the number of sampling times included in the first historical duration.

[0008] In one possible design, calibration parameters are determined based on the cumulative error of inertial measurement data within a target time period. This includes: acquiring the abnormal deviation values ​​and matching identifiers of all abnormal data points within the target time period. The abnormal deviation value is a quantification of the degree of deviation between the abnormal data point and the corresponding dimension's data change trend curve after time-series fitting and smoothing. The matching identifier indicates the number of abnormal data points in other dimensions at the time of occurrence of the abnormal data point. Based on the abnormal deviation values ​​and matching identifiers of all abnormal data points, the cumulative error is determined. An error relationship curve is constructed based on historical data, reflecting the correlation between the cumulative error and the observation error. Based on the cumulative error and the error relationship curve, the predicted observation error is determined. The observation error corresponding to the previous sampling time of the environmental perception data is determined as the first observation error. Based on the first observation error, the predicted observation error, and the duration of the target time period, the second observation error corresponding to the current time is determined. Based on the second observation error, the predicted observation error, and the mean square error of the error relationship curve, the calibration parameters are determined.

[0009] In one possible design, the calibration parameter is determined based on the second observation error, the predicted observation error, and the mean square error of the error relationship curve. This includes: determining an intermediate parameter based on the second observation error, the predicted observation error, and the mean square error; determining the intermediate parameter as the calibration parameter if the intermediate parameter is greater than or equal to zero; and determining zero as the calibration parameter if the intermediate parameter is less than zero.

[0010] In one possible design, the noise covariance parameter of the extended Kalman filter used for data fusion is dynamically adjusted based on the adjusted motion state reliability. This includes: obtaining the initial process noise covariance matrix corresponding to the extended Kalman filter; adjusting the initial process noise covariance matrix according to the adjusted motion state reliability to obtain the process noise covariance matrix at the current time, wherein the value of the adjusted process noise covariance matrix is ​​negatively correlated with the adjusted motion state reliability; obtaining the initial observation noise covariance matrix corresponding to the extended Kalman filter; determining the matrix amplification factor based on the calibration parameters corresponding to each sampling time within a second historical period, wherein the second historical period is the time interval between two environmental perception data sampling times prior to the current time; and adjusting the initial observation noise covariance matrix according to the matrix amplification factor to obtain the observation noise covariance matrix at the current time, wherein the adjusted observation noise covariance matrix is ​​positively correlated with the matrix amplification factor.

[0011] In one possible design, after acquiring the environmental perception data and inertial measurement data of the AGV during its operation, the AGV positioning method for automatic material retrieval in the warehouse also includes: preprocessing the environmental perception data and inertial measurement data. The preprocessing includes calibration, noise reduction, and timestamp alignment to ensure that each frame of environmental perception data has corresponding inertial measurement data.

[0012] In one possible design, an adjusted extended Kalman filter is used to fuse environmental perception data and inertial measurement data to output the positioning information of the AGV. This includes: when no new environmental perception data is obtained, predicting the current state vector of the AGV based on the current inertial measurement data and the previous state vector of the AGV using the adjusted extended Kalman filter; when new environmental perception data is obtained, updating the current state vector of the AGV based on the current environmental perception data and the inertial measurement data; and outputting the updated state vector of the AGV as the positioning information of the AGV.

[0013] In one possible design, an error relationship curve is constructed based on historical data, including: extracting multiple sample pairs corresponding to historical time periods from the historical data, each sample pair including the cumulative error within a historical time period and the observation error determined after the end of the historical time period; performing function fitting on all extracted sample pairs to obtain the mapping relationship between the cumulative error and the observation error, determining the fitted mapping relationship function as the error relationship curve, and recording the mean square error of the fitting process as the mean square error of the error relationship curve.

[0014] Secondly, a positioning system for AGVs (Automated Guided Vehicles) in warehouses is provided, comprising: a data acquisition unit for acquiring environmental perception data and inertial measurement data during AGV operation, wherein the sampling frequency of the inertial measurement data is higher than that of the environmental perception data; a reliability calculation unit for detecting anomalies in the inertial measurement data and determining the reliability of the motion state of the inertial measurement data, wherein the motion state reliability characterizes the degree to which the inertial measurement data reflects the true motion state of the AGV; a calibration parameter determination unit for determining calibration parameters based on the cumulative error of the inertial measurement data within a target time period, wherein the start and end times of the target time period are the previous sampling time and the current time of the environmental perception data; a reliability adjustment unit for adjusting the motion state reliability based on the calibration parameters to obtain the adjusted motion state reliability; and a fusion positioning unit for dynamically adjusting the noise covariance parameter of the extended Kalman filter used for data fusion based on the adjusted motion state reliability; and fusing the environmental perception data and inertial measurement data through the adjusted extended Kalman filter to output the positioning information of the AGV.

[0015] The present invention has the following beneficial effects:

[0016] In the AGV positioning method for automated warehouse material handling provided in this invention embodiment, high-frequency inertial measurement data and low-frequency environmental perception data are first acquired synchronously to provide a foundation for fusion. Then, through in-depth anomaly detection and analysis of the inertial measurement data, a "motion state reliability" that quantifies the reliability of the actual motion state is dynamically generated. Based on this, a mechanism for determining "calibration parameters" is further introduced. These parameters are obtained based on the cumulative error of inertial data within a target time period and its correlation with the prediction deviation of perception data, thereby linking the microscopic anomalies of the inertial measurement data with the potential for macroscopic positioning errors. Finally, the motion state reliability is corrected a second time using the calibration parameters, and the process and observation noise covariance parameters of the extended Kalman filter are adjusted in real time and dynamically. This series of interconnected steps enables the entire positioning system to sensitively perceive changes in the AGV's own motion state (such as slippage and vibration) and changes in the external environment (such as open areas versus densely packed shelving areas), and automatically adjust the filter's trust weights for internal inertial prediction and external environmental observation. It significantly improves the positioning accuracy and robustness of AGVs in warehouse scenarios with variable operating environments and dynamic interference, suppresses the unconventional accumulation of errors, ensures the continuity and reliability of automatic material handling operations, and fundamentally enhances the automation level and operating efficiency of intelligent warehousing systems. Attached Figure Description

[0017] To more clearly illustrate the technical solutions and advantages in the embodiments of the present 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 only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the structure of an AGV (Automated Guided Vehicle) positioning system for automated material handling in a warehouse, provided in one embodiment of the present invention.

[0019] Figure 2 This is a flowchart illustrating an AGV (Automated Guided Vehicle) positioning method for automated material handling in a warehouse, as provided in one embodiment of the present invention. Detailed Implementation

[0020] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of an AGV (Automated Guided Vehicle) positioning method and system for automated material handling in warehouses proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0021] In embodiments of the present invention, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" or "for example" in embodiments of the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0022] In the description of this invention, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. The term "and / or" in this document 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 alone, A and B simultaneously, and B alone. Furthermore, "at least one" and "more than one" refer to two or more. The terms "first," "second," etc., do not limit the quantity or order of execution, and "first," "second," etc., do not necessarily imply differences.

[0023] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0024] The following description, in conjunction with the accompanying drawings, details the specific scheme of the AGV (Automated Guided Vehicle) positioning method and system for automated material handling in warehouses provided by this invention.

[0025] Please see Figure 1 The diagram illustrates a structural schematic of an AGV (Automated Guided Vehicle) positioning system for automated material handling in a warehouse, according to an embodiment of the present invention. Figure 1 As shown, the AGV positioning system 10 for automated warehouse material handling includes a data acquisition unit 11, a reliability calculation unit 12, a calibration parameter determination unit 13, a reliability adjustment unit 14, and a fusion positioning unit 15.

[0026] The data acquisition unit 11 is used to acquire environmental perception data through a Simultaneous Localization and Mapping (SLAM) system (such as LiDAR) mounted on the AGV vehicle. The environmental perception data provides the vehicle's position estimate in the global environment at a relatively low frequency (such as 20Hz).

[0027] The data acquisition unit 11 is also used to acquire inertial measurement data through the inertial measurement unit mounted on the AGV. The sampling frequency of the inertial measurement data (e.g., 1000-2000Hz) is much higher than the sampling frequency of the environmental perception data. The inertial measurement data includes its own motion state data such as position dimension, acceleration dimension, and angle dimension.

[0028] The data acquisition unit 11 is also used to preprocess the acquired environmental perception data and inertial measurement data, and associate each frame of low-frequency environmental perception data with its timestamp-corresponding high-frequency inertial measurement data, laying the foundation for subsequent data synchronization and fusion.

[0029] The credibility calculation unit 12 is used to evaluate the credibility of the inertial measurement data at the current moment in reflecting the true motion state of the AGV, that is, the credibility of the motion state.

[0030] The credibility calculation unit 12 includes an anomaly detection module and a frequency analysis module.

[0031] The anomaly detection module is specifically used to perform time-series smoothing and anomaly detection on multiple dimensions of inertial measurement data, identify abnormal data points that deviate from the normal trend, and record the time of occurrence and the abnormal deviation value.

[0032] The frequency analysis module is specifically used to calculate the frequency of anomalies in all dimensions and the synchronization frequency of anomalies in multiple dimensions within a set first historical period, based on the timestamp information of abnormal data points.

[0033] Finally, the credibility calculation unit 12 calculates the initial credibility of the motion state based on the abnormal frequency and the synchronization frequency. This credibility initially quantifies the possibility that the inertial measurement data is affected by systematic interference (such as slippage and ground vibration) rather than random noise.

[0034] The calibration parameter determination unit 13 is used to determine a calibration parameter based on all abnormal data points and their matching identifiers and abnormal deviation values ​​within the target time period (from the sampling time of the previous environmental perception data to the current time). This calibration parameter is used to calibrate the motion state credibility calculated by the credibility calculation unit.

[0035] The calibration parameter determination unit 13 includes a cumulative error calculation module, an observation error acquisition module, and a calibration parameter generation module.

[0036] The cumulative error calculation module is specifically used to combine the abnormal deviation value and matching identifier of each abnormal data point to calculate the cumulative error of the inertial measurement data within the target time period.

[0037] The observation error acquisition module is specifically used to construct an error relationship curve based on historical data. This curve is obtained by analyzing a large amount of historical normal operation data and fitting a function between the cumulative error and the observation error, reflecting the statistical correlation between the two. Furthermore, after calculating the cumulative error of the inertial measurement data within the target time period, the predicted observation error can be obtained based on the error relationship curve. The predicted observation error represents the expected level of observation error based on historical experience and the current degree of anomaly accumulation.

[0038] The observation error acquisition module is also used to acquire the actual observation error calculated at the last environmental perception data sampling time and determine it as the first observation error. Then, based on the first observation error, the predicted observation error, and the duration of the current target time period, the second observation error at the current time is generated through linear interpolation.

[0039] The calibration parameter generation module is specifically used to calculate the final calibration parameters by comparing the second observation error and the predicted observation error, combined with the mean square error of the error relationship curve (used to characterize the normal fluctuation range of the function obtained by fitting historical data).

[0040] The credibility adjustment unit 14 is used to obtain the adjusted credibility of the motion state based on the credibility of the motion state determined by the credibility calculation unit 12 and the calibration parameters generated by the calibration parameter determination unit 13.

[0041] Understandably, a large calibration parameter indicates abnormal inertial measurement data and may lead to overly optimistic positioning predictions, significantly lowering the initial motion state confidence. Conversely, a zero or very small calibration parameter results in minimal or no adjustment to the initial confidence level. This step effectively overcomes the problem of inaccurate initial confidence assessment caused by abnormal "spurious synchronization" of various dimensions of inertial measurement data when the AGV experiences slippage or other loss of control. Ultimately, this unit outputs a more reliable adjusted motion state confidence level that better reflects the actual contribution of inertial measurement data to positioning prediction.

[0042] The fusion positioning unit 15 is used to dynamically adjust the noise covariance parameter of the extended Kalman filter according to the adjusted motion state reliability, and then use the adjusted extended Kalman filter to fuse environmental perception data and inertial measurement data to output the positioning information of the AGV vehicle.

[0043] The fusion positioning unit 15 includes a parameter adjustment module and a filtering fusion positioning module.

[0044] The parameter adjustment module is specifically used to dynamically adjust the process noise covariance matrix of the extended Kalman filter based on the reliability of the adjusted motion state. Higher reliability indicates more reliable inertial measurement data, so the value of the process noise covariance matrix is ​​appropriately reduced to increase the filter's confidence in the prediction model of the inertial measurement data. Simultaneously, the parameter adjustment module is also used to dynamically adjust the observation noise covariance matrix based on the statistical characteristics (such as the average value) of all calibration parameters within the sampling interval of two adjacent environmental sensing data sets. Higher overall levels of calibration parameters indicate a greater risk of interference to the carrier or environmental sensing data observations under the current environment, so the value of the observation noise covariance matrix is ​​appropriately increased to reduce the filter's absolute confidence in a single observation, making it more robust during updates.

[0045] The filtering and fusion positioning module is specifically used to make predictions using state equations when only high-frequency inertial measurement data arrives, providing continuous position and attitude estimations; and to update the system using observation equations when low-frequency environmental perception data arrives, using actual environmental observations to correct the drift caused by inertial predictions.

[0046] Please see Figure 2 The diagram illustrates a flowchart of an AGV (Automated Guided Vehicle) positioning method for automated material handling in a warehouse, according to an embodiment of the present invention, including the following steps S201-S205.

[0047] S201. Acquire environmental perception data and inertial measurement data of the AGV during operation.

[0048] The sampling frequency of inertial measurement data is higher than that of environmental perception data.

[0049] As one possible implementation, a laser positioning module is used as the environmental perception data acquisition device. The laser positioning module emits a laser beam to the environment around the AGV and receives the reflected signal to collect point cloud data of the external environment. The point cloud data is the environmental perception data, which is used to characterize the spatial position association information of the AGV in the warehouse environment.

[0050] Optionally, the laser positioning module can be installed at the top center of the AGV, with its scanning angle covering a 360° unobstructed area around the vehicle. The scanning distance can be adapted to the maximum spacing requirements of the warehouse shelves. The sampling frequency of the environmental perception data is set to 20Hz to ensure stable acquisition of environmental position reference information in complex warehouse environments.

[0051] Simultaneously, an inertial measurement unit (IMU) is used as the inertial measurement data acquisition device to collect data related to the motion state of the AGV itself. The inertial measurement data specifically includes position dimension data, acceleration dimension data, and angular dimension data. Position dimension data characterizes the real-time displacement information of the AGV, acceleration dimension data characterizes the linear acceleration changes of the AGV, and angular dimension data characterizes the angular velocity and attitude angle changes of the AGV. These three types of data together comprehensively reflect the motion state of the AGV.

[0052] Optionally, the inertial measurement unit can be fixedly installed at the center of the chassis of the AGV, in line with the direction of movement of the vehicle. The sampling frequency of the inertial measurement data is set to 1000-2000Hz, for example, 1500Hz, which is much higher than the sampling frequency of the environmental perception data, and can quickly capture the instantaneous state changes during the movement of the vehicle.

[0053] In some embodiments, to ensure the acquisition quality of environmental perception data and inertial measurement data, the laser positioning module records the timestamp corresponding to each frame of data synchronously when acquiring environmental perception data, with the timestamp accuracy down to the millisecond level. Similarly, the inertial measurement unit records the timestamp of each sampling point when acquiring inertial measurement data, and uses the same timing reference as the laser positioning module's timestamp, providing a basis for timestamp alignment in subsequent data preprocessing. During the acquisition process, environmental perception data is stored in real-time in the vehicle's local cache module. Due to the high sampling frequency of inertial measurement data, a "real-time transmission + segmented caching" approach is adopted to ensure both timely data transmission and avoid data loss.

[0054] In some embodiments, after acquiring environmental perception data and inertial measurement data, the two types of data are preprocessed to ensure they can be used for subsequent fusion and computation. The preprocessing includes the following steps:

[0055] First, the raw environmental perception data and inertial measurement data are calibrated and denoised to eliminate the effects of sensor zero bias, temperature drift and random noise.

[0056] Secondly, timestamp alignment is performed. Each frame of environmental perception data and each inertial measurement data sample is marked with its precise acquisition timestamp. Through a time synchronization mechanism, each frame of environmental perception data is associated with its corresponding inertial measurement data sample, thus establishing a strict correspondence between each low-frequency environmental perception data frame and its acquisition time, as well as the acquisition time of the previous frame of environmental perception data, across the entire time period. This correspondence forms the basis for subsequent data fusion and time interval analysis.

[0057] S202. Perform anomaly detection on the inertial measurement data to determine the reliability of the motion state of the inertial measurement data.

[0058] Among them, motion state reliability is used to characterize the reliability of inertial measurement data in reflecting the true motion state of the AGV.

[0059] As one possible approach, the acquired inertial measurement data is first divided into three independent motion state dimensions: position, acceleration, and angle. Then, the original data sequence for each dimension is subjected to time-series fitting and smoothing to filter out high-frequency noise and highlight its changing trends, resulting in a smoothed data curve.

[0060] Furthermore, for the smoothed data sequences of each dimension, a preset anomaly detection algorithm is used to identify abnormal data points.

[0061] Optionally, the preset anomaly detection algorithm can be a dimensionality reduction anomaly detection method based on principal component analysis: projecting the feature sequence of a single dimension within a short time window onto a low-dimensional principal component space, and calculating the projection distance of each data point on the principal component hyperplane. This projection distance is defined as the anomaly deviation value of the data point. If the anomaly deviation value of a data point exceeds a preset threshold, it is marked as an anomaly data point in that dimension, and its occurrence timestamp and corresponding anomaly deviation value are recorded.

[0062] Furthermore, based on the timestamp information of the abnormal data points in each dimension, the abnormal frequency and synchronization frequency are determined.

[0063] Among them, the anomaly frequency is the frequency of abnormal data points appearing within the first historical period, and the synchronization frequency is the frequency of abnormal data points appearing synchronously in each dimension within the first historical period.

[0064] In some embodiments, to determine the anomaly frequency and synchronization frequency, based on the time sets corresponding to the anomaly data points in each dimension, the number of times included in the union of all sets is determined as the number of anomaly times, and the number of times included in the intersection of all sets is determined as the number of synchronization anomalies. Then, based on the number of anomaly times and the number of sampling times included within the first historical duration, the anomaly frequency is determined, for example, the ratio of the number of anomaly times to the number of sampling times included within the first historical duration is determined as the anomaly frequency, which reflects the frequency of abnormal fluctuations exhibited by the inertial measurement data within the first historical duration. Based on the number of synchronization anomalies and the number of sampling times included within the first historical duration, the synchronization frequency is determined, for example, the ratio of the number of synchronization anomalies to the number of sampling times included within the first historical duration is determined as the synchronization frequency, which reflects the proportion of anomalies within the first historical duration that may be caused by systematic events (such as vehicle slippage or ground vibration).

[0065] It should be noted that the first historical duration is the sampling time of a preset number of inertial measurement data before the current time. The preset number can be empirically valued as 30, 40, 50, etc., and the embodiments of the present invention do not make specific limitations on this.

[0066] In some embodiments, cross-dimensional anomaly matching analysis is performed. For any dimension of the inertial measurement data, the time t corresponding to any anomalous data point is determined. At that time t, it is determined whether the data of the other two dimensions are also identified as anomalous data points. If neither of the other two dimensions is identified as an anomalous data point, the matching identifier of the anomalous data point is set to 0. If both of the other two dimensions are identified as anomalous data points, the matching identifier of the anomalous data point is set to 1. If one of the other two dimensions is identified as an anomalous data point, the matching identifier of the anomalous data point is set to 1 / 2.

[0067] Finally, the reliability of the motion state is determined based on the abnormal frequency and the synchronization frequency.

[0068] In some embodiments, the formula for calculating the reliability of the motion state is as follows:

[0069]

[0070] In the formula, To assess the reliability of the motion state, This is an abnormal frequency. This is the synchronization frequency.

[0071] Among them, when the abnormal frequency is close to the synchronization frequency (and A value close to 0 indicates that the inertial measurement data within the first historical period is genuinely abnormal data affected by environmental interference. This is due to sudden abnormal phenomena in the AGV's state (such as slippage or vibration). Although the AGV's inertial measurement data shows abnormalities at this time, it still represents the true motion state characteristics of the AGV, indicating that the inertial measurement data reliably reflects the true motion state, thus increasing the reliability of the motion state. Approaching 1. When outlier data points are mainly composed of mismatched random noise (and...) (relatively large), reliability of motion state reduce.

[0072] Understandably, this embodiment of the invention achieves multi-angle, refined perception of the AGV's motion state by dividing inertial measurement data into multiple independent motion state dimensions (such as position, acceleration, and angle) and performing anomaly detection on each. Compared to single-dimensional anomaly judgment, this method can effectively capture anomaly correlation patterns between different physical quantities. By calculating the anomaly frequency and synchronization frequency, the system can intelligently distinguish between isolated anomalies caused by random noise and real anomaly states caused by systematic interference such as slippage and collisions, which occur synchronously across multiple dimensions. This design significantly improves the system's ability to evaluate the quality of inertial measurement data, ensuring that the reliability of the final determined motion state not only reflects the volatility of the data itself but, more importantly, reveals the physical causes behind the anomalies. This provides a more accurate and reliable basis for subsequent adaptive parameter adjustments, enhancing the overall positioning system's judgment and robustness in complex, dynamic warehouse environments.

[0073] S203. Determine the calibration parameters based on the cumulative error of the inertial measurement data within the target time period.

[0074] The start and end times of the target time period are the previous sampling time and the current time of the environmental perception data.

[0075] As one possible approach, the first step is to obtain the anomaly deviation value and matching identifier for each anomalous data point within the target time period. The anomaly deviation value is a quantification of the degree of deviation between the anomalous data point and the corresponding dimension's data change trend curve after time-series fitting and smoothing. The matching identifier indicates the number of anomalous data points in other dimensions at the time of occurrence of the anomalous data point.

[0076] In some embodiments, the determination of the abnormal deviation value and matching identifier of each abnormal data point can be referred to the description in step S202 of the present invention, which will not be repeated here.

[0077] Furthermore, the cumulative error is determined based on the abnormal deviation values ​​and matching identifiers of all abnormal data points.

[0078] In some embodiments, the formula for calculating the cumulative error is as follows:

[0079]

[0080] In the formula, For cumulative error, The number of outlier data points within the target time period. For the first Abnormal deviation values ​​of anomalous data points For the first Matching identifiers for anomalous data points.

[0081] in, Indicates passing through the first Each outlier data point is matched with an identifier, and its outlier deviation is weighted. If the outlier is caused by random noise, then... If the value is 0, this item is 0 and contributes nothing to the cumulative error; if the data anomaly is not caused by random noise, then... The value is 1 / 2 or 1. This item contributes to the cumulative error. The contribution of this abnormal data point to the cumulative error is summed with the contributions of other abnormal data to obtain the cumulative error accumulated by all non-random noise within the target time period.

[0082] Furthermore, an error relationship curve is constructed based on historical data, which is used to reflect the correlation between cumulative error and observation error.

[0083] In some embodiments, multiple sample pairs corresponding to historical time periods are extracted from historical data. Each sample pair contains the cumulative error within a historical time period and the observation error determined after the end of that historical time period. A function is fitted to all extracted sample pairs to obtain an error relationship curve reflecting the correlation between the cumulative error and the observation error. This curve quantifies the expected observation error range corresponding to different levels of cumulative error. Simultaneously, the mean square error of the function fitting process is recorded. As a threshold for subsequent judgment of the reasonableness of error deviation, mean square error is used to characterize the fitting accuracy of the error relationship curve and reflect the degree of dispersion of the correlation between cumulative error and observation error in historical data.

[0084] The historical time period refers to the time interval between every two adjacent environmental perception data acquisition times when the AGV is operating normally (under control). Based on the inertial measurement data within this historical time period, the cumulative error and the predicted environmental perception data can be obtained. The actual environmental perception data collected at the end of this historical period is Thus, the observation error for that historical period can be obtained. ,in, Indicates calculation and The difference norm between them.

[0085] Furthermore, the prediction observation error is determined based on the cumulative error and the error relationship curve.

[0086] In some embodiments, the cumulative error calculated based on inertial measurement data within the target time period is used. Substitute into the error relationship curve In this context, the output value represents the prediction observation error. .

[0087] Furthermore, the observation error corresponding to the previous sampling time of the environmental perception data is determined as the first observation error; and based on the first observation error, the predicted observation error, and the duration of the target time period, the second observation error corresponding to the current time is determined.

[0088] In some embodiments, the environmental perception data collected at the previous sampling time is compared with the environmental perception data predicted based on inertial measurement data between the previous and the time before that sampling time. The difference norm between the two is calculated as the observation error, i.e., the first observation error. Then, based on the first observation error, the predicted observation error determined based on the cumulative error at the current time, and the duration of the target time period, an estimated value of the observation error at the current time is constructed using a linear interpolation method, i.e., the second observation error. The calculation formula is as follows:

[0089]

[0090] In the formula, For the current moment The estimated value of the observation error, also known as the second observation error, This is the first observation error (i.e., the observation error corresponding to the previous sampling time of the environmental perception data). This refers to the previous sampling time of the environmental sensing data. The duration of the target time period. The sampling interval for environmental perception data. , This refers to the predicted observation error determined based on the cumulative error and the error relationship curve.

[0091] This formula is a linear interpolation model used to estimate the sampling time of the current inertial measurement data. The observation error. The first observation error, and the error at the previous environmental perception data sampling time. The calculated true observation error is used as the known starting point for linear interpolation; To predict observation error, it represents the expected level of observation error caused by the cumulative degree of anomalies in current inertial measurement data based on historical statistical patterns, and serves as the estimated endpoint for linear interpolation. It is a linear scaling factor. This represents the actual time elapsed from the current moment to the previous environmental perception data sampling moment. This represents the sampling interval for environmental sensing data. The value of this coefficient ranges from 0 to 1, accurately representing the current moment. The relative time position within the sampling interval of environmental perception data.

[0092] It should be noted that within the sampling interval of a single environmental perception data point, the positioning prediction error caused by anomalies in inertial measurement data increases approximately linearly with time. Within such a short period, barring sudden and drastic disturbances, the motion state of the AGV and the sensor error characteristics typically change continuously, and their cumulative effect can be approximated as a uniform increase. The linear interpolation model is the simplest and most effective approximation of this short-term physical process. It avoids the enormous computational overhead of complex nonlinear models and is well-suited for real-time computation in AGV-mounted embedded systems.

[0093] Finally, the calibration parameters are determined based on the second observation error, the predicted observation error, and the mean square error of the error relationship curve.

[0094] In some embodiments, an intermediate parameter is determined based on the second observation error, the predicted observation error, and the mean square error; this intermediate parameter can be the difference between the predicted observation error and the second observation error and the mean square error. Then, if the intermediate parameter is greater than or equal to zero, it is determined as the calibration parameter; if the intermediate parameter is less than zero, zero is determined as the calibration parameter.

[0095] For example, the formula for calculating the calibration parameters is as follows:

[0096]

[0097] In the formula, For calibration parameters, To predict observation errors, This is the second observation error. Mean square error, For a linear rectified function, in When the value is greater than 0, the output is ,exist When the value is less than 0, the output is 0.

[0098] Among them, only when the predicted observation error is significantly greater than (exceeds the historical normal fluctuation range) Current error estimated based on the linear assumption At that time, calibration parameters Only positive values ​​are taken, and the calibration parameters increase with the larger the prediction and observation errors. The larger the value, the more likely the current inertial measurement data is to be abnormal, potentially leading to a more severe deviation in positioning prediction than expected, indicating a risk of "accelerated error accumulation" that could trigger subsequent correction actions. If the difference is within the normal fluctuation range, then... This indicates that the current state does not require special correction.

[0099] Understandably, this embodiment of the invention establishes an empirical model between the abnormal accumulation of inertial measurement data and the final positioning observation error by utilizing the error relationship curve obtained from fitting historical normal operation data. This allows the system to proactively predict the impact on positioning accuracy based on the degree of anomalies currently detected. Furthermore, by introducing a linear interpolation method, continuous and smooth estimation of the error state is achieved. Finally, by comparing the prediction error based on historical experience with the current estimation error based on the linear model, and subtracting inherent fluctuations, the generated calibration parameters can sensitively identify the extraordinary risk of "accelerated error accumulation." This entire mechanism enables the system to no longer passively wait for positioning deviations to occur before correction, but rather to perceive risks in advance and dynamically adjust the trust weights of the filter, thereby significantly improving the adaptive capability of the extended Kalman filter in time-varying environments and the overall positioning accuracy and stability.

[0100] S204. Adjust the motion state reliability based on the calibration parameters to obtain the adjusted motion state reliability.

[0101] In some embodiments, the formula for calculating the reliability of the adjusted motion state is as follows:

[0102]

[0103] In the formula, To improve the reliability of the adjusted motion state, For calibration parameters, To assess the reliability of the motion state, Representing a system with the natural constant as the base and... It is an exponential function of an exponent.

[0104] Among them, when abnormal error risk is detected ( >0), Only then might the value be non-zero, thus initiating substantial adjustments to the parameters of the extended Kalman filter. If 0, then regardless Why is it worth it? All values ​​are 0, requiring no special correction. When the initial confidence level... When the value is too high (such as in the above-mentioned out-of-control but data-synchronized scenario). The value will become very small. This will greatly suppress the calibration parameters. For the final The impact of the value. Its physical meaning is: even if the system detects a potential error risk ( >0), but if the current inertial measurement data itself exhibits a high degree of "consistency" (which may be a false reliability), the system will also adopt a relatively cautious approach and will not excessively adjust the parameters. Conversely, when the initial confidence level is low... At lower levels, The value is relatively large. The impact will be more fully reflected, enabling the system to respond quickly to unreliable data.

[0105] S205. Based on the adjusted motion state reliability, dynamically adjust the noise covariance parameter of the extended Kalman filter used for data fusion; and through the adjusted extended Kalman filter, fuse the environmental perception data and inertial measurement data to output the positioning information of the AGV vehicle.

[0106] It should be noted that the performance of the extended Kalman filter (EKF) is highly dependent on its process noise covariance matrix. and observation noise covariance matrix The setting of the process noise covariance matrix. Characterizing the degree of confidence in the uncertainty of the system's motion model (inertial measurement data), the observation noise covariance matrix. It characterizes the degree of confidence in the uncertainty of sensor observations (i.e., environmental perception data).

[0107] As one possible implementation, the initial process noise covariance matrix corresponding to the extended Kalman filter is first obtained. and the initial observation noise covariance matrix .

[0108] Furthermore, based on the adjusted motion state reliability, the initial process noise covariance matrix is ​​adjusted to obtain the process noise covariance matrix at the current moment. The value of the adjusted process noise covariance matrix is ​​negatively correlated with the adjusted motion state reliability.

[0109] In some embodiments, the formula for calculating the process noise covariance matrix at the current moment is as follows:

[0110]

[0111] In the formula, Let be the process noise covariance matrix at the current moment. The initial process noise covariance matrix, To improve the reliability of the adjusted motion state, For example, the maximum and minimum value normalization method can be used to normalize the adjusted motion state confidence based on the historical maximum and minimum values ​​of the adjusted motion state confidence.

[0112] in, The larger the value, the better. The smaller the value, the better. The smaller the adjustment, the smaller the process noise covariance matrix in the extended Kalman filter. This means that EKF tends to trust the state evolution derived from the system's motion model rather than process noise in the prediction step. Therefore, when When the initial process noise covariance matrix is ​​large, by reducing the initial process noise covariance matrix, the excessive influence of abnormal fluctuations in inertial measurement data on the prediction results is effectively constrained, making the state prediction more dependent on the model itself, thereby suppressing prediction divergence caused by local unreliability of data.

[0113] Furthermore, based on the calibration parameters corresponding to each sampling moment within the second historical duration, the matrix amplification factor is determined, and the initial observation noise covariance matrix is ​​adjusted according to the matrix amplification factor to obtain the observation noise covariance matrix at the current moment. The adjusted observation noise covariance matrix is ​​positively correlated with the matrix amplification factor. The second historical duration is the time period between two environmental perception data sampling moments prior to the current moment.

[0114] In some embodiments, the formula for calculating the observation noise covariance matrix at the current time is as follows:

[0115]

[0116] In the formula, Let be the observation noise covariance matrix at the current moment. The initial observation noise covariance matrix, This represents the average of the calibration parameters corresponding to each sampling time point within the second historical period. The normalization function can be used, for example, to normalize the mean using the historical maximum value. If the mean is greater than the historical maximum value, the normalization value is 1; if the mean is less than the historical maximum value, the normalized value is the ratio of the mean to the historical maximum value. The matrix scaling factor is... .

[0117] in, This value is used to characterize the overall risk level of positioning prediction errors exceeding a linear, normal growth pattern due to anomalies in inertial measurement data within the most recent environmental sensing data sampling interval. A higher value indicates a higher risk of environmental disturbances or model mismatch, and the observations in the expected environmental sensing data may contain greater uncertainty or anomalies. Therefore, [the value is used to characterize the overall risk level of positioning prediction errors exceeding a linear, normal growth pattern due to anomalies in inertial measurement data within the most recent environmental sensing data sampling interval]. The corresponding amplification means that EKF takes a relatively "conservative" or "tolerant" attitude towards newly arrived environmental perception data in the update step, reducing the impact of a single abnormal observation on the overall state estimation and enhancing the robustness of the system.

[0118] Furthermore, the environmental perception data and inertial measurement data are fused using an adjusted extended Kalman filter to output the positioning information of the AGV vehicle.

[0119] In some embodiments, regarding the timing control of the fusion processing, when no new environmental sensing data is obtained, the EKF continuously executes the state prediction process. For each frame of inertial measurement data acquired, the predicted state vector is updated once through the state prediction equation to ensure that high-frequency continuous positioning information can still be output within the sampling interval of environmental sensing data (such as the 50ms interval corresponding to 20Hz sampling). When new environmental sensing data is acquired, the observation update process is immediately triggered, and the prediction deviation is corrected by utilizing the stable reference characteristics of the environmental sensing data to ensure the long-term accuracy of the positioning information.

[0120] In the absence of new environmental perception data, the predicted state vector for the current moment is recursively calculated using the current high-frequency inertial measurement data and the optimal state estimate from the previous moment, through the EKF state prediction equation. This state vector typically includes information such as the AGV's position, velocity, and attitude angles.

[0121] When new environmental perception data is obtained, this data is used as the observation input to the EKF. Using the observation update equation of the EKF, the predicted state vector is fused with the current actual observation, the Kalman gain is calculated, and the predicted state is optimally corrected to obtain a more accurate state vector estimate updated at the current time.

[0122] Finally, the positioning information output by the fusion processing consists of the position and attitude parameters contained in the updated state vector. This positioning information is output to the control system of the AGV in a standardized data format. The data update frequency is consistent with the sampling frequency of the inertial measurement data, ensuring that the AGV can obtain accurate position and attitude references in real time during the automatic material picking operation in the warehouse, meeting the operational needs in complex scenarios such as narrow passages and dynamic obstacles.

[0123] Understandably, in the AGV positioning method for automated warehouse material handling provided in this embodiment of the invention, high-frequency inertial measurement data and low-frequency environmental perception data are first acquired synchronously to provide a foundation for fusion. Then, through in-depth anomaly detection and analysis of the inertial measurement data, a "motion state reliability" that quantifies the reliability of the actual motion state is dynamically generated. Based on this, a mechanism for determining "calibration parameters" is further introduced. These parameters are obtained based on the cumulative error of inertial data within a target time period and its correlation with the prediction deviation of the perception data, thereby linking the microscopic anomalies of the inertial measurement data with the potential for macroscopic positioning errors. Finally, the motion state reliability is corrected a second time using the calibration parameters, and the process and observation noise covariance parameters of the extended Kalman filter are adjusted in real time and dynamically. This series of steps is interconnected, enabling the entire positioning system to sensitively perceive changes in the AGV's own motion state (such as slippage and vibration) and changes in the external environment (such as open areas and densely packed shelving areas), and automatically adjust the filter's trust weights for internal inertial prediction and external environmental observation. It significantly improves the positioning accuracy and robustness of AGVs in warehouse scenarios with variable operating environments and dynamic interference, suppresses the unconventional accumulation of errors, ensures the continuity and reliability of automatic material handling operations, and fundamentally enhances the automation level and operating efficiency of intelligent warehousing systems.

[0124] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0125] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for positioning AGVs (Automated Guided Vehicles) for automated material handling in warehouses, characterized in that, The method includes: The AGV acquires environmental perception data and inertial measurement data during its operation, wherein the sampling frequency of the inertial measurement data is higher than that of the environmental perception data. Anomaly detection is performed on the data in each dimension of the inertial measurement data to identify abnormal data points in each dimension. The inertial measurement data includes position dimension, acceleration dimension, and angle dimension. Based on the timestamp information of the abnormal data points in each dimension, the abnormal frequency and synchronization frequency are determined. The abnormal frequency is the frequency of abnormal data points occurring within the first historical duration, and the synchronization frequency is the frequency of abnormal data points occurring synchronously in each dimension within the first historical duration. The first historical duration is the sampling time of a preset number of inertial measurement data points before the current moment. Based on the abnormal frequency and the synchronization frequency, the motion state reliability is determined. The motion state reliability characterizes the extent to which the inertial measurement data reflects the true motion state of the AGV. The motion state reliability is determined using the formula... Sure, The reliability of the stated motion state. The abnormal frequency, The synchronization frequency; The calibration parameters are determined based on the cumulative error of the inertial measurement data within the target time period, wherein the start and end times of the target time period are the previous sampling time and the current time of the environmental perception data. The motion state reliability is adjusted based on the calibration parameters to obtain the adjusted motion state reliability. Based on the adjusted motion state reliability, the noise covariance parameter of the extended Kalman filter used for data fusion is dynamically adjusted; and the environmental perception data and the inertial measurement data are fused through the adjusted extended Kalman filter to output the positioning information of the AGV vehicle. The step of dynamically adjusting the noise covariance parameter of the extended Kalman filter used for data fusion based on the adjusted motion state reliability includes: Obtain the initial process noise covariance matrix corresponding to the extended Kalman filter; Based on the adjusted motion state reliability, the initial process noise covariance matrix is ​​adjusted to obtain the process noise covariance matrix at the current moment. The process noise covariance matrix at the current moment is obtained through the formula... Sure, Let be the process noise covariance matrix at the current moment. The initial process noise covariance matrix is... The reliability of the adjusted motion state, As a normalization function, the value of the adjusted process noise covariance matrix is ​​negatively correlated with the reliability of the adjusted motion state; Obtain the initial observation noise covariance matrix corresponding to the extended Kalman filter; The matrix amplification factor is determined based on the calibration parameters corresponding to each sampling moment within the second historical duration, where the second historical duration is the time period between two environmental perception data sampling moments prior to the current moment. The initial observation noise covariance matrix is ​​adjusted according to the matrix magnification factor to obtain the observation noise covariance matrix at the current time. The observation noise covariance matrix at the current time is obtained through the formula... Sure, Let be the observation noise covariance matrix at the current moment. The initial observation noise covariance matrix, This represents the average of the calibration parameters corresponding to each sampling time point within the second historical period. The normalization function is used, and the adjusted observation noise covariance matrix is ​​positively correlated with the matrix amplification factor.

2. The AGV positioning method for automated material handling in warehouses according to claim 1, characterized in that, Based on the timestamp information of the outlier data points in each dimension, determine the outlier frequency and synchronization frequency, including: Based on the time set corresponding to the abnormal data points in each dimension, the number of time points included in the union of all sets is determined as the number of abnormal time points, and the number of time points included in the intersection of all sets is determined as the number of synchronization anomalies. The abnormal frequency is determined based on the number of abnormal moments and the number of sampling moments included in the first historical duration, and the synchronization frequency is determined based on the number of synchronization abnormalities and the number of sampling moments included in the first historical duration.

3. The AGV positioning method for automated warehouse material handling according to claim 1, characterized in that, Based on the cumulative error of the inertial measurement data within the target time period, calibration parameters are determined, including: Obtain the abnormal deviation value and matching identifier of all abnormal data points within the target time period. The abnormal deviation value is the quantitative result of the degree of deviation of the abnormal data point from the corresponding dimension data change trend curve after the abnormal data point has undergone time series fitting and smoothing processing. The matching identifier is used to indicate the number of abnormal data points in other dimensions at the time of occurrence of the abnormal data point. The cumulative error is determined based on the abnormal deviation values ​​and matching identifiers of all abnormal data points; An error relationship curve is constructed based on historical data, which is used to reflect the correlation between cumulative error and observation error; The predicted observation error is determined based on the cumulative error and the error relationship curve. The observation error corresponding to the previous sampling time of the environmental perception data is determined as the first observation error; Based on the first observation error, the predicted observation error, and the duration of the target time period, determine the second observation error corresponding to the current moment; The calibration parameters are determined based on the second observation error, the predicted observation error, and the mean square error of the error relationship curve.

4. The AGV positioning method for automated material handling in warehouses according to claim 3, characterized in that, The calibration parameters are determined based on the second observation error, the predicted observation error, and the mean square error of the error relationship curve, including: The intermediate parameters are determined based on the second observation error, the predicted observation error, and the mean square error. If the intermediate parameter is greater than or equal to zero, the intermediate parameter is determined as the calibration parameter; If the intermediate parameter is less than zero, zero is determined as the calibration parameter.

5. The AGV positioning method for automated material handling in warehouses according to claim 1, characterized in that, After acquiring environmental perception data and inertial measurement data of the AGV during operation, the method further includes: The environmental perception data and the inertial measurement data are preprocessed, including calibration, denoising, and timestamp alignment, so that each frame of environmental perception data has corresponding inertial measurement data.

6. The AGV positioning method for automated material handling in warehouses according to claim 1, characterized in that, The environmental perception data and the inertial measurement data are fused using an adjusted extended Kalman filter to output the positioning information of the AGV, including: Without obtaining new environmental perception data, the state vector of the AGV at the current moment is predicted using the adjusted extended Kalman filter based on the inertial measurement data at the current moment and the state vector of the AGV at the previous moment. Upon obtaining new environmental perception data, the state vector of the AGV is updated based on the current environmental perception data and the inertial measurement data. The updated state vector of the AGV is output as the positioning information of the AGV.

7. The AGV positioning method for automated material handling in warehouses according to claim 3, characterized in that, Error relationship curves are constructed based on historical data, including: Extract sample pairs corresponding to multiple historical time periods from historical data. Each sample pair includes the cumulative error within a historical time period and the observation error determined after the end of the historical time period. A function is fitted to all extracted sample pairs to obtain the mapping relationship between the cumulative error and the observation error. The fitted mapping relationship function is determined as the error relationship curve, and the mean square error of the fitting process is recorded as the mean square error of the error relationship curve.

8. A positioning system for AGV (Automated Guided Vehicle) carts for automated material handling in warehouses, characterized in that, include: The data acquisition unit is used to acquire environmental perception data and inertial measurement data of the AGV during operation, wherein the sampling frequency of the inertial measurement data is higher than the sampling frequency of the environmental perception data; A credibility calculation unit is used to perform anomaly detection on the data in each dimension of the inertial measurement data and determine the abnormal data points in each dimension. The inertial measurement data includes position dimension, acceleration dimension and angle dimension. Based on the timestamp information of the abnormal data points in each dimension, the abnormal frequency and synchronization frequency are determined. The abnormal frequency is the frequency of abnormal data points occurring within the first historical duration, and the synchronization frequency is the frequency of abnormal data points occurring synchronously in each dimension within the first historical duration. The first historical duration is the sampling time of a preset number of inertial measurement data points before the current moment. Based on the abnormal frequency and the synchronization frequency, the motion state reliability is determined. The motion state reliability characterizes the extent to which the inertial measurement data reflects the true motion state of the AGV. The motion state reliability is determined using the formula... Sure, The reliability of the stated motion state. The abnormal frequency, The synchronization frequency; The calibration parameter determination unit is used to determine calibration parameters based on the cumulative error of the inertial measurement data within a target time period, wherein the start and end times of the target time period are the previous sampling time and the current time of the environmental perception data. A credibility adjustment unit is used to adjust the credibility of the motion state based on the calibration parameters to obtain the adjusted credibility of the motion state. The fusion positioning unit is used to dynamically adjust the noise covariance parameter of the extended Kalman filter used for data fusion based on the adjusted motion state reliability; and to fuse the environmental perception data and the inertial measurement data through the adjusted extended Kalman filter to output the positioning information of the AGV vehicle. The fusion localization unit is specifically used to obtain the initial process noise covariance matrix corresponding to the extended Kalman filter; Based on the adjusted motion state reliability, the initial process noise covariance matrix is ​​adjusted to obtain the process noise covariance matrix at the current moment. The process noise covariance matrix at the current moment is obtained through the formula... Sure, Let be the process noise covariance matrix at the current moment. The initial process noise covariance matrix is... The reliability of the adjusted motion state, As a normalization function, the value of the adjusted process noise covariance matrix is ​​negatively correlated with the reliability of the adjusted motion state; Obtain the initial observation noise covariance matrix corresponding to the extended Kalman filter; The matrix amplification factor is determined based on the calibration parameters corresponding to each sampling moment within the second historical duration, where the second historical duration is the time period between two environmental perception data sampling moments prior to the current moment. The initial observation noise covariance matrix is ​​adjusted according to the matrix magnification factor to obtain the observation noise covariance matrix at the current time. The observation noise covariance matrix at the current time is obtained through the formula... Sure, Let be the observation noise covariance matrix at the current moment. The initial observation noise covariance matrix is... This represents the average value of the calibration parameters corresponding to each sampling moment within the second historical duration. The normalization function is used, and the adjusted observation noise covariance matrix is ​​positively correlated with the matrix amplification factor.

Images