Automatic data fitting method and system combined with equipment position calibration

Abstract

The invention provides an automatic data fitting method and system combined with equipment position calibration, and the method comprises the steps: firstly collecting a multi-source sensor data set of target equipment in a three-dimensional space, including equipment motion track data, position coordinate offset and environment interference parameters, carrying out the feature extraction of the multi-source sensor data set, and carrying out the feature extraction of the target equipment; generating a target feature set containing spatial position correlation features, dynamic trajectory fluctuation features and environment coupling features, then calling a pre-trained adaptive fitting model, inputting the target feature set to perform dynamic parameter matching, and generating an initial calibration parameter set; and generating a dynamic path sequence of equipment position calibration based on the initial calibration parameter set, triggering the equipment to execute automatic position calibration according to the sequence, and feeding back calibrated verification data to the model update parameters, thereby improving the precision, real-time performance and adaptivity of equipment position calibration.

Description

Technical Field

[0001] The present invention relates to the technical field of sensors, and in particular, to an automated data fitting method and system combined with device position calibration. Background Art

[0002] In many fields such as industrial automation, robot navigation, and intelligent warehousing, there are extremely high requirements for the accurate position calibration of target devices in three-dimensional space. Currently, traditional device position calibration methods mainly rely on manual measurement and fixed parameter settings. For example, in some simple industrial scenarios, workers will use measurement tools to manually record the initial position information of the device and set fixed calibration parameters based on this information.

[0003] However, these traditional methods have many limitations. On the one hand, manual measurement is easily affected by human factors such as measurement errors and irregular operations, resulting in low calibration accuracy. On the other hand, during actual operation, the device is affected by various factors such as environmental temperature changes, mechanical vibrations, and uneven ground. These factors will cause the position of the device to shift, and the fixed calibration parameters cannot adapt to these changes in real time, thus affecting the normal operation and working accuracy of the device. In addition, as the complexity of the device working environment and tasks continues to increase, traditional calibration methods are difficult to meet the requirements of high precision, real-time performance, and self-adaptability for device position calibration. There is an urgent need for a method that can combine device position calibration and has the ability of automated data fitting to solve these problems. Summary of the Invention

[0004] In view of the above-mentioned problems, in combination with the first aspect of the present invention, embodiments of the present invention provide an automated data fitting method combined with device position calibration, and the method includes: Collect a multi-source sensor data set of a target device in three-dimensional space, where the multi-source sensor data set includes device movement trajectory data, position coordinate offset, and environmental interference parameters; Extract features from the multi-source sensor data set to generate a target feature set, where the target feature set includes spatial position association features, dynamic trajectory fluctuation features, and environmental coupling features; Call a pre-trained adaptive fitting model, input the target feature set into the adaptive fitting model for dynamic parameter matching processing, and generate an initial calibration parameter set of the target device; Generate a dynamic path sequence for device position calibration based on the initial calibration parameter set, where the dynamic path sequence includes spatial coordinate adjustment parameters of multiple consecutive calibration points; Trigger the target device to perform an automated position calibration operation according to the dynamic path sequence, and collect the calibrated verification data set and feedback it to the adaptive fitting model to update the model parameters.

[0005] In another aspect, an embodiment of the present invention further provides an automated data fitting system combined with device position calibration, including a processor and a machine-readable storage medium. The machine-readable storage medium is connected to the processor. The machine-readable storage medium is used to store programs, instructions, or codes, and the processor is used to execute the programs, instructions, or codes in the machine-readable storage medium to implement the above method.

[0006] Based on the above aspects, in the embodiment of the present invention, by collecting a multi-source sensor data set of the target device in a three-dimensional space, covering various information such as device movement trajectory data, position coordinate offset, and environmental interference parameters, feature extraction is performed on the multi-source sensor data set to generate a target feature set, including spatial position association features, dynamic trajectory fluctuation features, and environmental coupling features, which can accurately describe the state of the device in the three-dimensional space and the interaction relationship with the environment. Call the pre-trained adaptive fitting model to perform dynamic parameter matching processing on the target feature set to generate an initial calibration parameter set. The adaptive ability of the adaptive fitting model is utilized, and the parameters can be dynamically adjusted according to different feature information, improving the accuracy and adaptability of the calibration parameters. Generate a dynamic path sequence for device position calibration based on the initial calibration parameter set, enabling the device to perform an automated position calibration operation according to this sequence, realizing real-time dynamic adjustment of the device position, and further improving the accuracy and real-time performance of position calibration. Feed the collected and calibrated verification data set back to the adaptive fitting model to update the model parameters, enabling the adaptive fitting model to continuously learn and improve, better adapting to the position calibration requirements of the device in different environments and working states, thereby significantly improving the accuracy, real-time performance, and self-adaptability of device position calibration. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 is a schematic flowchart of the execution process of the automated data fitting method combined with device position calibration provided by an embodiment of the present invention.

[0008] Figure 2 is a schematic diagram of exemplary hardware and software components of the automated data fitting system combined with device position calibration provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0009] The present invention will be specifically described below with reference to the accompanying drawings of the specification. Figure 1 is a flowchart of the automated data fitting method combined with device position calibration provided by an embodiment of the present invention. The automated data fitting method combined with device position calibration will be introduced in detail below.

[0010] Step S110: Collect a multi-source sensor data set of the target device in a three-dimensional space, where the multi-source sensor data set includes device motion trajectory data, position coordinate offsets, and environmental interference parameters.

[0011] Taking an industrial automation production scenario as an example, specifically, taking an AGV (Automated Guided Vehicle) that performs a goods handling task in an intelligent warehousing system as an example, in order to achieve subsequent accurate position calibration and data fitting, it is necessary to collect its multi-source sensor data set.

[0012] Step S111: Real-time collect the displacement increment data of the target device on the X-axis, Y-axis, and Z-axis through a laser ranging sensor array deployed on the motion plane of the target device.

[0013] On the motion plane of the AGV, a laser ranging sensor array is deployed. The sensors in the laser ranging sensor array are distributed according to a set layout to ensure that the displacement information of the AGV in the three-dimensional space can be obtained comprehensively and accurately. Each laser ranging sensor emits a laser beam. When the laser beam encounters the AGV, it will be reflected back. The sensor can calculate the distance between the sensor and the AGV by measuring the time difference between the laser emission time and the reception time and combining the propagation speed of the laser in the air. As the AGV moves, the sensor will continuously repeat this process to monitor the displacement changes of the AGV in the X-axis, Y-axis, and Z-axis directions in real time.

[0014] To ensure the accuracy and reliability of the data, it is necessary to calibrate and debug the laser ranging sensors. When installing the sensors, it is necessary to ensure the accuracy and stability of their installation positions to avoid measurement errors caused by improper installation. At the same time, it is also necessary to set and adjust the parameters of the sensors to adapt to different working environments and measurement requirements. For example, under different lighting conditions, the measurement accuracy of the sensors may be affected, so it is necessary to adjust parameters such as the sensitivity and threshold of the sensors according to the actual situation.

[0015] During the data collection process, the sensors collect data at a set sampling frequency. The selection of the sampling frequency needs to comprehensively consider the motion speed of the AGV and the requirements of data processing. If the sampling frequency is too low, it may lead to data loss or incompleteness and cannot accurately reflect the motion trajectory of the AGV; if the sampling frequency is too high, a large amount of data will be generated, increasing the burden of data processing. Therefore, it is necessary to reasonably select the sampling frequency according to the actual motion situation of the AGV to ensure that the collected data can accurately reflect the motion state of the AGV without bringing too much pressure to subsequent data processing.

[0016] The collected displacement increment data is a sequence that varies with time, which records the displacement changes of the AGV in the X-axis, Y-axis, and Z-axis directions at different times. These data will be stored in a data storage device for subsequent analysis and processing.

[0017] Step S112: Invoke the inertial measurement unit to obtain the angular velocity offset and acceleration offset of the target device, and record the temperature interference parameter and electromagnetic interference intensity at each sampling moment.

[0018] An inertial measurement unit (IMU) is installed on the AGV, which is a sensor capable of measuring the angular velocity and acceleration of an object. Through the IMU, the angular velocity offset and acceleration offset of the AGV can be obtained. The angular velocity offset reflects the speed change of the AGV during the rotation process, while the acceleration offset reflects the speed change of the AGV during the linear motion process.

[0019] The working principle of the IMU is based on the measurement of inertial forces. It contains sensitive components such as accelerometers and gyroscopes inside. The accelerometer can measure the acceleration of the AGV in various directions, and the gyroscope can measure the angular velocity of the AGV. During the movement of the AGV, the IMU continuously collects these data and converts them into digital signals for output.

[0020] When collecting the angular velocity offset and acceleration offset, it is also necessary to record the temperature interference parameter and electromagnetic interference intensity at each sampling moment. In an intelligent warehousing system, temperature changes may affect the performance of the IMU, resulting in measurement errors. For example, an increase in temperature may cause thermal expansion of the electronic components inside the IMU, thereby affecting its measurement accuracy. Therefore, it is necessary to use a temperature sensor to monitor the ambient temperature in real time and record it.

[0021] Electromagnetic interference is also an important factor affecting the measurement accuracy of the IMU. In a warehousing environment, there may be various electrical devices and communication signals, which will generate electromagnetic interference. To reduce the impact of electromagnetic interference on the IMU, shielding measures need to be taken around the IMU, and at the same time, an electromagnetic interference sensor is used to monitor the electromagnetic interference intensity in real time and record it.

[0022] The purpose of recording these temperature interference parameters and electromagnetic interference intensities is to compensate and correct them during subsequent data processing to improve the accuracy and reliability of the data.

[0023] Step S113: Perform timestamp alignment processing on the displacement increment data, the angular velocity offset, and the acceleration offset to obtain synchronized motion trajectory data.

[0024] Due to the possible differences in the sampling times of the laser ranging sensor and the inertial measurement unit, the timestamps of the collected displacement increment data, angular velocity offset, and acceleration offset may not be consistent. To effectively integrate and analyze the displacement increment data, angular velocity offset, and acceleration offset, timestamp alignment processing is required.

[0025] The specific method of timestamp alignment processing is to first determine a unified time reference. For example, the sampling time of the laser ranging sensor can be selected as the reference, or the sampling time of the inertial measurement unit can be selected as the reference. In this embodiment, the sampling time of the laser ranging sensor is selected as the reference.

[0026] Then, based on this time reference, interpolation processing is performed on the displacement increment data, angular velocity offset, and acceleration offset. The purpose of interpolation processing is to obtain the values of these data at a unified time point. For example, if the laser ranging sensor collects displacement increment data at a certain time point, and the inertial measurement unit does not collect the corresponding angular velocity offset and acceleration offset at this time point, then through interpolation, based on the angular velocity offset and acceleration offset data at the previous and subsequent time points, the angular velocity offset and acceleration offset at this time point can be estimated.

[0027] Through timestamp alignment processing, synchronized motion trajectory data is obtained. These data are consistent in time and can accurately reflect the motion trajectory of the AGV in three-dimensional space.

[0028] Step S114: Normalize the temperature interference parameter and the electromagnetic interference intensity to obtain a standardized environmental interference parameter.

[0029] The numerical ranges of the collected temperature interference parameter and electromagnetic interference intensity may be different. To facilitate subsequent data analysis and processing, they need to be normalized. The purpose of normalization is to map these data to a unified numerical range so that different types of environmental interference parameters are comparable.

[0030] The specific method of normalization can adopt the linear normalization method. First, determine the maximum and minimum values of the temperature interference parameter and the electromagnetic interference intensity. Then, for each temperature interference parameter and electromagnetic interference intensity data, perform normalization processing through the following formula: Normalized value = (Original value - Minimum value) / (Maximum value - Minimum value).

[0031] Through the above method, the temperature interference parameter and electromagnetic interference intensity data are normalized to the range of 0 to 1, and a standardized environmental interference parameter analysis is obtained.

[0032] Step S115: Perform spatial coordinate mapping on the synchronized motion trajectory data and the standardized environmental interference parameters to generate a multi-source sensor data set including device motion trajectory data, position coordinate offset, and environmental interference parameters.

[0033] After obtaining the synchronized motion trajectory data and the standardized environmental interference parameters, it is necessary to perform spatial coordinate mapping on them to generate a multi-source sensor data set including device motion trajectory data, position coordinate offset, and environmental interference parameters.

[0034] The specific method of spatial coordinate mapping can be to associate the displacement increment data and position coordinate offset in the synchronized motion trajectory data with the standardized environmental interference parameters. They can be corresponding through timestamps, so that the displacement increment data and position coordinate offset at each moment correspond to the corresponding standardized environmental interference parameters.

[0035] For example, at a certain time point, the synchronized motion trajectory data records the displacement increments and position coordinate offsets of the AGV in the X-axis, Y-axis, and Z-axis directions, and at the same time the standardized environmental interference parameters record the temperature interference parameter and electromagnetic interference intensity at this time point. After associating these data, a multi-source sensor data set including device motion trajectory data, position coordinate offset, and environmental interference parameters is obtained.

[0036] This multi-source sensor data set is a multi-dimensional data set, which contains the motion information and environmental interference information of the AGV in the three-dimensional space.

[0037] Step S120: Extract features from the multi-source sensor data set to generate a target feature set, where the target feature set includes spatial position association features, dynamic trajectory fluctuation features, and environmental coupling features.

[0038] After obtaining the multi-source sensor data set, in order to better analyze and process these data, it is necessary to extract features from it to generate a target feature set. The target feature set includes spatial position association features, dynamic trajectory fluctuation features, and environmental coupling features, and these features can more effectively reflect the motion state of the AGV and environmental impacts.

[0039] Step S121: Extract the trajectory change rate within a continuous time window from the device motion trajectory data, and construct spatial position association features based on the trajectory change rate. The spatial position association features include the coordinate transformation matrix between adjacent calibration points.

[0040] In the device motion trajectory data, the trajectory change rate reflects the motion speed and direction change of the AGV within a continuous time window. To extract the trajectory change rate, a suitable time window can be selected, such as a time window of 10 sampling moments. Within this time window, calculate the ratio of the displacement change amount of the AGV in the X-axis, Y-axis, and Z-axis directions to time to obtain the trajectory change rate.

[0041] Based on the trajectory change rate, spatial position correlation features can be constructed. The spatial position correlation features include the coordinate transformation matrix between adjacent calibration points. The calibration points refer to some representative points selected on the motion trajectory of the AGV. By calculating the coordinate transformation matrix between adjacent calibration points, the motion relationship of the AGV between different positions can be described.

[0042] The calculation method of the coordinate transformation matrix can adopt the method of rigid body transformation. Assume that between two adjacent calibration points, the AGV has translational and rotational motions. The translation vector and rotation matrix can be calculated and combined into the coordinate transformation matrix. The translation vector describes the translation distance of the AGV in the three coordinate axis directions, and the rotation matrix describes the rotation angle and direction of the AGV.

[0043] By extracting the trajectory change rate within a continuous time window and constructing spatial position correlation features based on this, the position change and motion relationship of the AGV in three-dimensional space can be described more accurately.

[0044] Step S122: Perform frequency domain transformation processing on the position coordinate offset, extract the low-frequency component as the reference position feature, and extract the high-frequency component as the dynamic trajectory fluctuation feature.

[0045] The position coordinate offset reflects the deviation of the AGV relative to the ideal position during the motion process. To better analyze these deviations, frequency domain transformation processing needs to be performed on the position coordinate offset.

[0046] The frequency domain transformation processing can adopt the method of Fourier transform. The Fourier transform can convert the time domain signal into a frequency domain signal, thereby decomposing the position coordinate offset into components of different frequencies.

[0047] In the frequency domain signal, the low-frequency component represents the slow-changing part of the position coordinate offset. This part of the change is usually caused by the long-term motion trend of the AGV or system errors. Extract the low-frequency component as the reference position feature, which can reflect the basic position state of the AGV.

[0048] The high-frequency components represent the rapidly changing part of the position coordinate offset, which is usually caused by the jitter, vibration, or external interference of the AGV during movement. Extracting the high-frequency components as the dynamic trajectory fluctuation features can reflect the dynamic changes of the AGV during movement.

[0049] By performing frequency-domain transformation on the position coordinate offset, extracting the low-frequency components as the reference position features, and extracting the high-frequency components as the dynamic trajectory fluctuation features, the motion state and position deviation of the AGV can be analyzed more deeply.

[0050] Step S123: Perform a coupling degree analysis on the environmental interference parameters and the dynamic trajectory fluctuation features, calculate the contribution weight of the environmental parameters to the trajectory fluctuation, and generate environmental coupling features based on the contribution weight.

[0051] There may be a certain coupling relationship between the environmental interference parameters and the dynamic trajectory fluctuation features. To analyze this coupling relationship, a coupling degree analysis is required.

[0052] The method of coupling degree analysis can adopt the method of correlation analysis. By calculating the correlation coefficient between the environmental interference parameters and the dynamic trajectory fluctuation features, the coupling degree between them can be evaluated. The closer the correlation coefficient is to 1, the higher the coupling degree between them; the closer the correlation coefficient is to 0, the lower the coupling degree between them.

[0053] Based on the results of the coupling degree analysis, calculate the contribution weight of the environmental parameters to the trajectory fluctuation. The contribution weight reflects the influence degree of the environmental interference parameters on the dynamic trajectory fluctuation features. The contribution weight can be obtained by normalizing the correlation coefficient.

[0054] According to the contribution weight, generate environmental coupling features. The environmental coupling features are a feature vector that comprehensively considers the environmental interference parameters and the dynamic trajectory fluctuation features. It combines the environmental interference parameters and the dynamic trajectory fluctuation features with weights, so that the influence of the environmental interference parameters on the trajectory fluctuation can be more accurately reflected in the feature vector.

[0055] By performing a coupling degree analysis on the environmental interference parameters and the dynamic trajectory fluctuation features, calculating the contribution weight of the environmental parameters to the trajectory fluctuation, and generating environmental coupling features based on this, the influence of environmental factors on the AGV motion trajectory can be considered more comprehensively.

[0056] Step S124: Perform dimension alignment processing on the spatial position association features, the dynamic trajectory fluctuation features, and the environmental coupling features to generate a target feature set with a unified timestamp and spatial coordinate reference system.

[0057] Since the spatial position correlation feature, dynamic trajectory fluctuation feature, and environmental coupling feature may have different dimensions and timestamps, in order to combine the spatial position correlation feature, dynamic trajectory fluctuation feature, and environmental coupling feature into a unified target feature set, dimension alignment processing is required.

[0058] The first step of dimension alignment processing is to unify the timestamps. A unified time reference can be selected to adjust the timestamps of the spatial position correlation feature, dynamic trajectory fluctuation feature, and environmental coupling feature to this time reference, thereby ensuring their consistency in time.

[0059] The second step is to unify the spatial coordinate reference system. The spatial position correlation feature, dynamic trajectory fluctuation feature, and environmental coupling feature may be defined based on different spatial coordinate reference systems and need to be converted into the same spatial coordinate reference system. A fixed spatial coordinate reference system can be selected as the benchmark to convert the coordinates of other features into this benchmark coordinate system.

[0060] Through dimension alignment processing, the spatial position correlation feature, dynamic trajectory fluctuation feature, and environmental coupling feature are combined into a target feature set with a unified timestamp and spatial coordinate reference system. This target feature set is a multi-dimensional feature vector that contains the position, movement, and environmental information of the AGV in three-dimensional space.

[0061] Step S130: Call the pre-trained adaptive fitting model, input the target feature set into the adaptive fitting model for dynamic parameter matching processing, and generate the initial calibration parameter set of the target device.

[0062] After obtaining the target feature set, it is necessary to call the pre-trained adaptive fitting model for dynamic parameter matching processing to generate the initial calibration parameter set of the target device. The adaptive fitting model is an artificial intelligence model trained with a large amount of data, which can automatically adjust the parameters of the model according to the input target feature set to achieve the best fit for the target device.

[0063] Step S131: Input the spatial position correlation feature into the first feature encoding layer of the adaptive fitting model to generate a spatial position encoding vector.

[0064] The first feature encoding layer of the adaptive fitting model is specifically designed to process the spatial position correlation feature. After inputting the spatial position correlation feature into the first feature encoding layer, the first feature encoding layer will perform encoding processing on it and convert it into a spatial position encoding vector.

[0065] The first feature encoding layer usually contains multiple neurons and weight matrices. During the encoding process, the spatial position correlation features are multiplied and summed with the weight matrix, and then undergo a non-linear transformation through an activation function to finally obtain the spatial position encoding vector.

[0066] The spatial position encoding vector is a multi-dimensional vector that effectively compresses and represents the spatial position correlation features. This spatial position encoding vector contains the position relationship and motion information of the AGV in the three-dimensional space and will be used for subsequent feature fusion and parameter decoding.

[0067] Step S132: Input the dynamic trajectory fluctuation features into the second feature encoding layer of the adaptive fitting model to generate a trajectory fluctuation encoding vector.

[0068] The second feature encoding layer of the adaptive fitting model is used to process the dynamic trajectory fluctuation features. After inputting the dynamic trajectory fluctuation features into the second feature encoding layer, this second feature encoding layer will perform encoding processing on them to generate a trajectory fluctuation encoding vector.

[0069] The working principle of the second feature encoding layer is similar to that of the first feature encoding layer. It also contains multiple neurons and weight matrices, and through operations and non-linear transformations with the dynamic trajectory fluctuation features, it converts them into a trajectory fluctuation encoding vector.

[0070] The trajectory fluctuation encoding vector is a multi-dimensional vector that can more effectively represent the dynamic changes of the AGV during movement. This trajectory fluctuation encoding vector will be fused with the spatial position encoding vector and the environmental coupling encoding vector to achieve more accurate parameter matching.

[0071] Step S133: Input the environmental coupling features into the third feature encoding layer of the adaptive fitting model to generate an environmental coupling encoding vector.

[0072] The third feature encoding layer of the adaptive fitting model is used to process the environmental coupling features. After inputting the environmental coupling features into the third feature encoding layer, this third feature encoding layer can perform encoding processing on them to generate an environmental coupling encoding vector.

[0073] The third feature encoding layer also contains multiple neurons and weight matrices, and through operations and non-linear transformations on the environmental coupling features, it converts them into an environmental coupling encoding vector.

[0074] The environmental coupling encoding vector is a multi-dimensional vector that synthesizes the information of environmental interference parameters and dynamic trajectory fluctuation features, reflecting the influence of environmental factors on the AGV movement trajectory. This environmental coupling encoding vector will participate in the subsequent attention weight allocation processing together with the spatial position encoding vector and the trajectory fluctuation encoding vector.

[0075] Step S134: Perform attention weight assignment processing on the spatial position encoding vector, trajectory fluctuation encoding vector, and environmental coupling encoding vector to obtain a weighted fusion feature vector.

[0076] After obtaining the spatial position encoding vector, trajectory fluctuation encoding vector, and environmental coupling encoding vector, it is necessary to perform attention weight assignment processing on them to obtain a weighted fusion feature vector. The purpose of attention weight assignment processing is to perform weighted fusion on them according to the importance of different features and improve the performance of the model.

[0077] Step S1341: Calculate the first similarity score between the spatial position encoding vector and the trajectory fluctuation encoding vector.

[0078] In this embodiment, in order to calculate the first similarity score between the spatial position encoding vector and the trajectory fluctuation encoding vector, for example, the cosine similarity method can be used. Cosine similarity measures the similarity between two vectors by calculating the cosine value of the angle between them.

[0079] Regarding the spatial position encoding vector and the trajectory fluctuation encoding vector as two vectors in a high-dimensional space, after aligning their dimensions, calculate their dot product and divide it by the product of their magnitudes to obtain the first similarity score. The closer the first similarity score is to 1, the more similar the two vectors are; the closer the score is to 0, the less similar the two vectors are.

[0080] Step S1342: Calculate the second similarity score between the trajectory fluctuation encoding vector and the environmental coupling encoding vector.

[0081] In this embodiment, the cosine similarity method is also used to calculate the second similarity score between the trajectory fluctuation encoding vector and the environmental coupling encoding vector. By calculating the dot product of these two vectors and dividing it by the product of their magnitudes, the second similarity score is obtained.

[0082] The second similarity score reflects the degree of similarity between the trajectory fluctuation feature and the environmental coupling feature. If the second similarity score is high, it indicates that the environmental factor has a greater impact on the trajectory fluctuation; if the second similarity score is low, it indicates that the environmental factor has a smaller impact on the trajectory fluctuation.

[0083] Step S1343: Calculate the third similarity score between the environmental coupling encoding vector and the spatial position encoding vector.

[0084] In this embodiment, the cosine similarity method is still used to calculate the third similarity score between the environmental coupling encoding vector and the spatial position encoding vector. By calculating the dot product of these two vectors and dividing it by the product of their magnitudes, the third similarity score is obtained.

[0085] The third similarity score reflects the similarity between environmental factors and the relationship of spatial positions, which can help the model better understand the impact of environmental factors on the position change of the AGV.

[0086] Step S1344: Perform softmax normalization on the first similarity score, the second similarity score, and the third similarity score to obtain the spatial position attention weight, the trajectory fluctuation attention weight, and the environmental coupling attention weight.

[0087] After obtaining the first similarity score, the second similarity score, and the third similarity score, in order to convert these scores into appropriate attention weights, softmax normalization needs to be performed on them. Softmax normalization can map different similarity scores to a probability distribution, making the sum of all attention weights equal to 1.

[0088] The specific processing process is to first perform an exponential operation on each similarity score to highlight the differences between the scores. Then, divide each score after the exponential operation by the sum of all scores after the exponential operation to obtain the corresponding attention weight. For example, for the first similarity score, after the exponential operation, divide it by the sum of the three scores after the exponential operation to obtain the spatial position attention weight. Similarly, the trajectory fluctuation attention weight and the environmental coupling attention weight can be obtained, and these attention weights reflect the importance of different features in the subsequent fusion process.

[0089] Step S1345: Scale the spatial position encoding vector according to the spatial position attention weight, scale the trajectory fluctuation encoding vector according to the trajectory fluctuation attention weight, scale the environmental coupling encoding vector according to the environmental coupling attention weight, and perform element-wise fusion on the scaled spatial position encoding vector, the trajectory fluctuation encoding vector, and the environmental coupling encoding vector to generate a weighted fusion feature vector.

[0090] In this embodiment, after obtaining the spatial position attention weight, the trajectory fluctuation attention weight, and the environmental coupling attention weight, the corresponding encoding vectors are scaled by the spatial position attention weight, the trajectory fluctuation attention weight, and the environmental coupling attention weight respectively. For the spatial position encoding vector, each of its elements is multiplied by the spatial position attention weight; for the trajectory fluctuation encoding vector, each of its elements is multiplied by the trajectory fluctuation attention weight; for the environmental coupling encoding vector, each of its elements is multiplied by the environmental coupling attention weight.

[0091] The three scaled encoding vectors are dimensionally matched, and then element-wise fusion is performed. The process of element-wise fusion is to combine the elements at the corresponding positions of the three scaled encoding vectors to form a weighted fusion feature vector. The weighted fusion feature vector synthesizes information from multiple aspects such as spatial position, dynamic trajectory fluctuations, and environmental coupling, and more comprehensively reflects the motion state of the AGV and the environmental impact.

[0092] Step S135: Invoke the parameter decoder of the adaptive fitting model to perform inverse mapping processing on the weighted fusion feature vector, and generate an initial calibration parameter set including position compensation parameters, motion acceleration correction coefficients, and environmental interference suppression coefficients.

[0093] After receiving the weighted fusion feature vector, the parameter decoder of the adaptive fitting model performs inverse mapping processing on it. The parameter decoder is an important module in the adaptive fitting model, and its role is to convert the weighted fusion feature vector into the initial calibration parameter set required in practice.

[0094] During the inverse mapping process, the neurons and weight matrices inside the parameter decoder perform a series of operations and transformations on the weighted fusion feature vector. In the above way, the information contained in the weighted fusion feature vector is decoded into position compensation parameters, motion acceleration correction coefficients, and environmental interference suppression coefficients. The position compensation parameter is used to correct the actual position of the AGV to make it closer to the ideal position; the motion acceleration correction coefficient is used to adjust the motion acceleration of the AGV to improve its motion stability; the environmental interference suppression coefficient is used to suppress the influence of environmental interference on the motion of the AGV.

[0095] Step S140: Generate a dynamic path sequence for device position calibration based on the initial calibration parameter set, and the dynamic path sequence includes spatial coordinate adjustment parameters of multiple consecutive calibration points.

[0096] After obtaining the initial calibration parameter set, it is necessary to generate a dynamic path sequence for device position calibration based on these parameters. The dynamic path sequence describes the motion path of the AGV from the current position to the target position and includes spatial coordinate adjustment parameters of multiple consecutive calibration points.

[0097] Step S141: Calculate the coordinate difference between the starting point and the ending point of the calibration path of the target device in three-dimensional space according to the position compensation parameter.

[0098] The position compensation parameters include the position correction information of the AGV in the X-axis, Y-axis, and Z-axis directions. According to this information, the coordinate differences in the three coordinate axis directions between the starting point and the ending point of the calibrated path can be calculated. For example, for the X-axis direction, subtract the X coordinate of the starting point from the X coordinate of the ending point, and then add the correction value in the X-axis direction in the position compensation parameters to obtain the coordinate difference in the X-axis direction. Similarly, the coordinate differences in the Y-axis and Z-axis directions can be calculated, and these coordinate differences reflect the distances that the AGV needs to move in the three coordinate axis directions during the calibration process.

[0099] Step S142: Perform piecewise linear interpolation on the coordinate differences based on the motion acceleration correction coefficient to generate the theoretical coordinate values of multiple intermediate calibration points.

[0100] In this embodiment, after obtaining the coordinate differences, use the motion acceleration correction coefficient to perform piecewise linear interpolation on them. The purpose of piecewise linear interpolation is to divide the calibrated path into multiple small segments, and each small segment corresponds to an intermediate calibration point. During the interpolation process, determine the length and slope of each small segment according to the motion acceleration correction coefficient, so that the acceleration of the AGV during the movement meets the requirements.

[0101] For example, the specific interpolation process is as follows: First, divide the coordinate differences according to the set rules based on the motion acceleration correction coefficient to obtain the lengths of multiple small segments. Then, according to the lengths of these small segments and the starting point coordinates, calculate the theoretical coordinate values of each intermediate calibration point in turn. For example, for the X-axis direction, start from the X coordinate of the starting point and accumulate in sequence according to the length of the small segment to obtain the theoretical X coordinate value of each intermediate calibration point. Similarly, the theoretical coordinate values in the Y-axis and Z-axis directions can be calculated.

[0102] Step S143: Dynamically adjust the theoretical coordinate values of each intermediate calibration point in combination with the environmental interference suppression coefficient to generate a dynamic path sequence including actual coordinate values, adjustment time intervals, and motion direction parameters.

[0103] Step S1431: Obtain the theoretical coordinate value of the current intermediate calibration point and the corresponding environmental interference suppression coefficient, and perform a scalar multiplication operation between the environmental interference suppression coefficient and each spatial dimension component of the theoretical coordinate value to generate a three-dimensional environmental compensation vector for each intermediate calibration point.

[0104] For each intermediate calibration point, first obtain its theoretical coordinate value and the corresponding environmental interference suppression coefficient. The environmental interference suppression coefficient is a coefficient that comprehensively considers various environmental interference factors, and it reflects the degree of influence of environmental interference on the position of the AGV.

[0105] Perform scalar multiplication operations on the environmental interference suppression coefficients with the components of the theoretical coordinate values in the X-axis, Y-axis, and Z-axis directions respectively. For example, for the X-axis direction, multiply the environmental interference suppression coefficient by the X-axis component of the theoretical coordinate value to obtain the environmental compensation component in the X-axis direction; similarly, the environmental compensation components in the Y-axis and Z-axis directions can be obtained. These three environmental compensation components form a three-dimensional environmental compensation vector. The three-dimensional environmental compensation vector reflects the correction information of environmental interference on the position of the intermediate calibration point.

[0106] Step S1432: Determine the dynamic adjustment priorities in each axial direction according to the component amplitudes of the three-dimensional environmental compensation vector in the X-axis, Y-axis, and Z-axis, and superimpose the three-dimensional environmental compensation vector on the corresponding axial components of the theoretical coordinate value in the order from high to low priority to generate the actual coordinate value.

[0107] Analyze the component amplitudes of the three-dimensional environmental compensation vector in the X-axis, Y-axis, and Z-axis directions. The larger the amplitude, the more serious the impact of environmental interference in that axial direction, and the higher the priority. In the order from high to low priority, sequentially superimpose the corresponding components of the three-dimensional environmental compensation vector on the corresponding axial components of the theoretical coordinate value.

[0108] For example, if the component amplitude in the X-axis direction is the largest, followed by the Y-axis direction, and the smallest in the Z-axis direction, then first add the X-axis component of the three-dimensional environmental compensation vector to the X-axis component of the theoretical coordinate value to obtain the adjusted actual coordinate value of the X-axis; then add the Y-axis component of the three-dimensional environmental compensation vector to the Y-axis component of the theoretical coordinate value to obtain the adjusted actual coordinate value of the Y-axis; finally, add the Z-axis component of the three-dimensional environmental compensation vector to the Z-axis component of the theoretical coordinate value to obtain the adjusted actual coordinate value of the Z-axis. In this way, the actual coordinate value of each intermediate calibration point is obtained.

[0109] Step S1433: Calculate the adjustment time interval for each intermediate calibration point based on the ratio of the total amplitude of the three-dimensional environmental compensation vector to the preset reference adjustment speed, where the total amplitude of the three-dimensional environmental compensation vector is the result of taking the square root of the sum of the squares of the axial components.

[0110] Calculate the total amplitude of the three-dimensional environmental compensation vector, which reflects the comprehensive degree of influence of environmental interference on the position of the intermediate calibration point. The preset reference adjustment speed is a preset constant, which represents the basic speed of the AGV when adjusting its position.

[0111] Divide the total amplitude of the three-dimensional environmental compensation vector by the preset reference adjustment speed to obtain the adjustment time interval for each intermediate calibration point. The adjustment time interval represents the time required for the AGV to adjust from the theoretical coordinate value to the actual coordinate value. The calculation of the adjustment time interval takes into account the degree of environmental interference. The greater the environmental interference, the longer the adjustment time interval.

[0112] Step S1434: Determine the motion direction parameters according to the positive and negative signs of the axial components of the three-dimensional environment compensation vector, where the motion direction parameters include the positive or negative movement flag for the X-axis, the positive or negative movement flag for the Y-axis, and the positive or negative movement flag for the Z-axis.

[0113] Observe the positive and negative signs of the components of the three-dimensional environment compensation vector in the X-axis, Y-axis, and Z-axis directions. If the X-axis component is positive, then the AGV needs to move in the positive X-axis direction, and the corresponding X-axis movement flag is the positive movement flag; if the X-axis component is negative, then the AGV needs to move in the negative X-axis direction, and the corresponding X-axis movement flag is the negative movement flag. Similarly, the movement flags for the Y-axis and Z-axis directions can be determined. The motion direction parameters clarify the direction in which the AGV needs to move at each intermediate calibration point.

[0114] Step S1435: Arrange the actual coordinate values, the adjustment time interval, and the motion direction parameters in chronological order to generate an initial dynamic path sequence.

[0115] Arrange the actual coordinate values, the adjustment time interval, and the motion direction parameters of each intermediate calibration point in chronological order to form an initial dynamic path sequence. The initial dynamic path sequence records the position information, adjustment time, and motion direction of each intermediate calibration point during the calibration process of the AGV, providing detailed guidance for the actual movement of the AGV.

[0116] Step S1436: Perform kinematic feasibility verification on each actual coordinate value in the initial dynamic path sequence. When it is detected that the displacement amount between adjacent actual coordinate values exceeds the product of the maximum motion speed of the device and the adjustment time interval, perform geometric scaling on the adjustment time interval until the displacement constraint condition is satisfied.

[0117] Perform kinematic feasibility verification on each actual coordinate value in the initial dynamic path sequence, mainly to check whether the displacement amount between adjacent actual coordinate values conforms to the motion ability of the AGV. The AGV has a limit on the maximum motion speed. If the displacement amount between adjacent actual coordinate values exceeds the product of the maximum motion speed of the device and the adjustment time interval, it means that the AGV cannot complete the above displacement within this time interval.

[0118] When it is detected that the displacement constraint condition is not met, it is necessary to perform geometric scaling on the adjustment time interval. The process of geometric scaling is to increase the adjustment time interval according to a set ratio until the displacement amount between adjacent actual coordinate values satisfies the constraint condition of the product of the maximum motion speed of the device and the adjustment time interval.

[0119] Step S1437: Update the time parameters in the initial dynamic path sequence according to the scaled adjustment time interval to generate an intermediate dynamic path sequence.

[0120] After completing the geometric scaling of the adjustment time interval, update the time parameters in the initial dynamic path sequence according to the scaled adjustment time interval. Replace the adjustment time of each intermediate calibration point with the scaled adjustment time, thus obtaining the intermediate dynamic path sequence. Under the premise of meeting kinematic feasibility, the intermediate dynamic path sequence arranges the movement time of the AGV more reasonably.

[0121] Step S1438: Insert device attitude stability detection points in the intermediate dynamic path sequence. The insertion position of the attitude stability detection points is the median moment of the adjustment time interval between every two adjacent intermediate calibration points, and add attitude maintenance instructions at the attitude stability detection points to suppress inertial drift.

[0122] To ensure the attitude stability of the AGV during movement, insert device attitude stability detection points in the intermediate dynamic path sequence. The insertion position of the attitude stability detection points is the median moment of the adjustment time interval between every two adjacent intermediate calibration points.

[0123] Add attitude maintenance instructions at each attitude stability detection point. The role of the attitude maintenance instructions is to suppress the inertial drift of the AGV. During the movement of the AGV, due to the effect of inertia, the attitude may deviate. The attitude maintenance instructions can adjust the attitude control parameters of the AGV to keep its attitude stable.

[0124] Step S1439: Reallocate the adjustment time intervals of each intermediate calibration point according to the insertion quantity of the attitude stability detection points, ensuring that the execution time of the attitude maintenance instructions is included within the total adjustment time interval.

[0125] After inserting the attitude stability detection points, it is necessary to reallocate the adjustment time intervals of each intermediate calibration point according to the insertion quantity. The purpose of the reallocation is to ensure that the execution time of the attitude maintenance instructions is included within the total adjustment time interval, while ensuring that the total time of the entire dynamic path sequence remains unchanged.

[0126] The specific reallocation process is to appropriately adjust the adjustment time intervals between adjacent intermediate calibration points according to the insertion position and quantity of the attitude stability detection points. So that each attitude stability detection point has sufficient time to execute the attitude maintenance instructions, without affecting the total time for the AGV to reach the target position.

[0127] Step S1440: Merge the updated intermediate dynamic path sequence with the attitude stability detection points to generate a dynamic path sequence including actual coordinate values, adjustment time intervals, and motion direction parameters.

[0128] Merge the updated intermediate dynamic path sequence with the attitude stability detection points, that is, incorporate the information of the attitude stability detection points into the intermediate dynamic path sequence. The merged sequence contains the actual coordinate values, adjusted time intervals, and motion direction parameters of all intermediate calibration points and attitude stability detection points, forming the final dynamic path sequence. This dynamic path sequence provides complete guiding information for the automated position calibration operation of the AGV.

[0129] Step S144: Perform collision detection and path smoothing on multiple consecutive calibration points in the dynamic path sequence to generate an optimized dynamic path sequence.

[0130] Performing collision detection on multiple consecutive calibration points in the dynamic path sequence mainly checks whether the AGV will collide with surrounding obstacles when moving according to the dynamic path sequence. This can be achieved by establishing geometric models of the AGV and surrounding obstacles, and then, based on the position information of each calibration point in the dynamic path sequence, determining whether the AGV will enter the spatial range of the obstacles during movement.

[0131] If a collision risk is detected, the dynamic path sequence needs to be adjusted to avoid obstacles. Path smoothing is to make the movement of the AGV smoother and reduce unnecessary rapid acceleration and deceleration. The method of curve fitting can be used to fit the calibration points in the dynamic path sequence to generate a smooth path.

[0132] After collision detection and path smoothing, an optimized dynamic path sequence is obtained. The optimized dynamic path sequence not only ensures the safety of the AGV movement but also improves the smoothness of the movement.

[0133] Step S150: Trigger the target device to perform an automated position calibration operation according to the dynamic path sequence, and feed back the collected verification data set after calibration to the adaptive fitting model to update the model parameters.

[0134] Step S151: Convert the dynamic path sequence into a device control instruction set, and the control instruction set contains the motor drive parameters and servo control parameters of each calibration point.

[0135] The dynamic path sequence contains detailed information of the AGV during the calibration process. To make the AGV move along this path, it is necessary to convert the dynamic path sequence into a device control instruction set. The device control instruction set contains the motor drive parameters and servo control parameters of each calibration point.

[0136] The motor drive parameters are used to control the motors of the AGV to rotate at a specified speed and direction. The servo control parameters are used to precisely control the position and attitude of the AGV to ensure that the AGV can accurately reach each calibration point.

[0137] The conversion process calculates the corresponding motor drive parameters and servo control parameters according to the actual coordinate values of each calibration point in the dynamic path sequence, the adjusted time interval, and the motion direction parameters, in combination with the kinematic model and dynamic model of the AGV.

[0138] Step S152: Real-time monitor the actual motion trajectory data of the target device when performing the calibration operation according to the control instruction set.

[0139] During the process of the AGV performing the calibration operation according to the control instruction set, real-time monitor its actual motion trajectory data. The real-time position and attitude information of the AGV during movement can be obtained through sensors installed on the AGV, such as laser range sensors, inertial measurement units, etc.

[0140] These actual motion trajectory data are sequences that change over time, recording the actual position and attitude of the AGV at different times. By real-time monitoring the actual motion trajectory data, it is possible to promptly detect whether there are deviations in the movement of the AGV.

[0141] Step S153: Collect the actual position coordinate data after calibration, and perform deviation calculation with the theoretical coordinate values in the dynamic path sequence to generate position calibration error data.

[0142] After the calibration operation is completed, collect the actual position coordinate data of the AGV. Perform deviation calculation on these actual position coordinate data and the theoretical coordinate values in the dynamic path sequence. The process of deviation calculation is to subtract the actual position coordinate data from the theoretical coordinate values in the X-axis, Y-axis, and Z-axis directions respectively to obtain the deviation values in each coordinate axis direction.

[0143] These deviation values constitute the position calibration error data, and the position calibration error data reflects the accuracy of the AGV during the position calibration process.

[0144] Step S154: Combine the actual motion trajectory data, position calibration error data, and real-time environmental interference parameters into a verification data set.

[0145] Combine the actual motion trajectory data obtained by real-time monitoring, the calculated position calibration error data, and the collected real-time environmental interference parameters together to form a verification data set. The verification data set contains various information of the AGV during the position calibration process.

[0146] Step S155: Input the verification data set into the error backpropagation link of the adaptive fitting model to update the weight parameters of the first feature encoding layer, the second feature encoding layer, and the parameter decoder.

[0147] Step S1551: Extract the root mean square error between the actual motion trajectory data and the theoretical trajectory data from the validation data set as the first loss function.

[0148] Extract the actual motion trajectory data and the theoretical trajectory data in the dynamic path sequence from the validation data set. Calculate the root mean square error between them. The root mean square error can measure the overall deviation degree between the actual motion trajectory and the theoretical trajectory. Take this root mean square error as the first loss function, and the first loss function reflects the error situation of the AGV on the motion trajectory.

[0149] Step S1552: Calculate the maximum offset of the position calibration error data in different spatial dimensions as the second loss function.

[0150] Find the maximum offset of the position calibration error data in the X-axis, Y-axis, and Z-axis directions respectively. These maximum offsets reflect the maximum deviation situation of the AGV in different spatial dimensions during the position calibration process. Take these maximum offsets as the second loss function, and the second loss function focuses on the maximum error of the position calibration.

[0151] Step S1553: Generate a third loss function based on the prediction error of the environmental coupling coding vector according to the real-time environmental interference parameters.

[0152] The real-time environmental interference parameters contain the latest information of the environmental interference. Compare it with the environmental coupling coding vector and calculate the prediction error. The prediction error reflects the prediction accuracy of the adaptive fitting model for the impact of environmental interference. Generate a third loss function based on this prediction error, and the third loss function reflects the performance of the model in dealing with environmental interference.

[0153] Step S1554: Perform weighted summation on the first loss function, the second loss function, and the third loss function to obtain a comprehensive loss value.

[0154] To comprehensively consider the motion trajectory error, the maximum position calibration error, and the environmental interference prediction error, perform weighted summation on the first loss function, the second loss function, and the third loss function. Assign a weight to each loss function, and the size of the weight is determined according to the importance of different loss functions.

[0155] Multiply the three loss functions by their corresponding weights and then add them up to obtain a comprehensive loss value. The comprehensive loss value reflects the overall error situation of the adaptive fitting model during the entire position calibration process.

[0156] Step S1555: Use the adaptive momentum optimization algorithm to update the weights of the fully connected layer and the attention allocation parameters of the adaptive fitting model according to the comprehensive loss value.

[0157] The adaptive momentum optimization algorithm is an optimization algorithm used to adjust model parameters. It combines the ideas of momentum and adaptive learning rate, and can update the model parameters more efficiently. In this embodiment, this algorithm is used to update the weights of the fully connected layer and the attention allocation parameters of the adaptive fitting model according to the comprehensive loss value.

[0158] Step S1555-1: Calculate the first gradient component of the comprehensive loss value with respect to the spatial position encoding vector.

[0159] The comprehensive loss value is a function of the model parameters, and the spatial position encoding vector is the result after encoding the model input features, which will affect the magnitude of the comprehensive loss value. To update the model parameters, it is necessary to calculate the gradient of the comprehensive loss value with respect to the spatial position encoding vector. The gradient represents the rate of change of the comprehensive loss value with respect to the spatial position encoding vector. Through the backpropagation algorithm, starting from the comprehensive loss value, calculate backward along the calculation path of the model to obtain the first gradient component of the comprehensive loss value with respect to the spatial position encoding vector. This first gradient component reflects the direction and degree of the impact of the change in the spatial position encoding vector on the comprehensive loss value.

[0160] Step S1555-2: Calculate the second gradient component of the comprehensive loss value with respect to the trajectory fluctuation encoding vector.

[0161] Similarly, for the trajectory fluctuation encoding vector, it is also necessary to calculate the gradient of the comprehensive loss value with respect to it. Through the backpropagation algorithm, in the computational graph of the model, starting from the comprehensive loss value, calculate step by step backward to obtain the second gradient component of the comprehensive loss value with respect to the trajectory fluctuation encoding vector. The second gradient component reflects the impact of the change in the trajectory fluctuation encoding vector on the comprehensive loss value.

[0162] Step S1555-3: Calculate the third gradient component of the comprehensive loss value with respect to the environmental coupling encoding vector.

[0163] Similar to the previous ones, calculate the third gradient component of the comprehensive loss value with respect to the environmental coupling encoding vector. By means of backpropagation, along the computational link of the model, determine the rate of change of the comprehensive loss value with respect to the environmental coupling encoding vector to obtain the third gradient component. This third gradient component reflects the effect of the change in the environmental coupling encoding vector on the comprehensive loss value, which helps to optimize the parameters related to environmental coupling in the model.

[0164] Step S1555-4: Determine the momentum update direction of each parameter node according to the exponentially weighted sum result of the historical gradient moving average and the current gradient.

[0165] The adaptive momentum optimization algorithm introduces the concept of momentum, which utilizes the information of historical gradients to smooth the process of parameter update. First, for each parameter node, the moving average of its historical gradients is calculated. The moving average of historical gradients is the weighted average of gradients over a past period of time, and the weights decay exponentially over time. Then, the current calculated gradient is exponentially weighted and summed with the moving average of historical gradients. The summation result comprehensively considers the information of historical gradients and the current gradient, and based on this summation result, the momentum update direction for each parameter node can be determined. The momentum update direction can guide the parameters to update in a more optimal direction, avoiding violent oscillations during the parameter update process.

[0166] Step S1555-5: Iteratively update the weights of the fully connected layer of the adaptive fitting model according to the momentum update direction, and at the same time adjust the temperature coefficient in the attention allocation parameter to control the smoothness of softmax normalization.

[0167] After determining the momentum update direction for each parameter node, the weights of the fully connected layer of the adaptive fitting model are iteratively updated according to this momentum update direction. The iterative update process is to gradually adjust the magnitude of the weights of the fully connected layer, making the comprehensive loss value gradually decrease. At the same time, the temperature coefficient in the attention allocation parameter also needs to be adjusted. The temperature coefficient plays an important role in the softmax normalization process, and it can control the smoothness of the output of the softmax function. When the temperature coefficient is large, the output of the softmax function will be smoother, and the difference between different categories will decrease; when the temperature coefficient is small, the output of the softmax function will be sharper, highlighting the categories with larger differences. By adjusting the temperature coefficient, the attention allocation can be made more reasonable, improving the model's attention to different features.

[0168] Through the above series of steps, the update of the parameters of the adaptive fitting model is completed. The updated model can better adapt to the actual situation of the target device, improving the accuracy and reliability of device position calibration.

[0169] During the entire data collection and processing process, the protection of privacy-sensitive data is involved. For example, the device's movement trajectory data and environmental interference parameters may contain certain sensitive information. To prevent the leakage of this data, a variety of privacy protection and anti-leakage technical means are adopted. First, in the data collection stage, the data is encrypted. The symmetric encryption algorithm is used to encrypt the collected data to ensure the security of the data during transmission. Second, in terms of data storage, a secure storage system is adopted to perform access control on the data, and only authorized personnel can access the data. At the same time, the data is regularly backed up to prevent data loss. In addition, during the data processing process, the data is anonymized to remove the personal identity information that may be contained in the data, protecting the privacy of users.

[0170] In practical applications, the adaptive fitting model is closely integrated with the intelligent warehousing system. The input data of the adaptive fitting model is a set of multi-source sensor data after collection and processing, including equipment movement trajectory data, position coordinate offsets, and environmental interference parameters, etc. The output of the adaptive fitting model is a set of initial calibration parameters for equipment position calibration and a dynamic path sequence. These output data are directly used to control the movement of the AGV, enabling it to accurately complete the cargo handling task. By continuously collecting the calibrated verification data set and updating the model parameters, the adaptive fitting model can continuously learn and optimize, improving the prediction and control capabilities of the AGV movement state, thereby enhancing the operation efficiency and reliability of the entire intelligent warehousing system.

[0171] Figure 2 FIG. shows a schematic diagram of exemplary hardware and software components of an automated data fitting system 100 for combined device position calibration that can implement the inventive concept provided by some embodiments of the present invention. For example, a processor 120 can be used on the automated data fitting system 100 for combined device position calibration and is used to execute the functions in the present invention.

[0172] The automated data fitting system 100 for combined device position calibration can be a general-purpose server or a special-purpose server, both of which can be used to implement the automated data fitting method for combined device position calibration of the present invention. Although only one server is shown in the present invention, for convenience, the functions described in the present invention can be implemented in a distributed manner on multiple similar platforms to balance the processing load.

[0173] For example, the automated data fitting system 100 for combined device position calibration can include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and different forms of storage media 140, such as disks, ROM, or RAM, or any combination thereof. Exemplarily, the automated data fitting system 100 for combined device position calibration can also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The method of the present invention can be implemented according to these program instructions. The automated data fitting system 100 for combined device position calibration also includes an I / O interface 150 between the computer and other input / output devices.

[0174] For ease of explanation, only one processor is described in the automated data fitting system 100 for combining device position calibration. However, it should be noted that the automated data fitting system 100 for combining device position calibration in the present invention may also include multiple processors. Therefore, the steps performed by one processor described in the present invention may also be jointly performed or separately performed by multiple processors. For example, if the processor of the automated data fitting system 100 for combining device position calibration performs step A and step B, it should be understood that step A and step B may also be jointly performed by two different processors or separately performed in one processor. For example, the first processor performs step A, the second processor performs step B, or the first processor and the second processor jointly perform steps A and B.

[0175] In addition, an embodiment of the present invention further provides a readable storage medium, in which computer-executable instructions are preset. When the processor executes the computer-executable instructions, the automated data fitting method for combining device position calibration as described above is implemented.

[0176] It should be noted that, in order to simplify the description of the present invention disclosure and thus help the understanding of one or more embodiments of the invention, in the foregoing description of the embodiments of the present invention, sometimes multiple features are merged into one embodiment, drawing, or description thereof.

Claims

1. An automated data fitting method combined with device position calibration, characterized in that, The method includes: Collecting a multi-source sensor data set of the target device in a three-dimensional space, where the multi-source sensor data set includes device motion trajectory data, position coordinate offsets, and environmental interference parameters; Performing feature extraction on the multi-source sensor data set to generate a target feature set, where the target feature set includes spatial position association features, dynamic trajectory fluctuation features, and environmental coupling features; Invoking a pre-trained adaptive fitting model, inputting the target feature set into the adaptive fitting model for dynamic parameter matching processing to generate an initial calibration parameter set of the target device; Generating a dynamic path sequence for device position calibration based on the initial calibration parameter set, where the dynamic path sequence includes spatial coordinate adjustment parameters of multiple consecutive calibration points; Triggering the target device to perform an automated position calibration operation according to the dynamic path sequence and feeding back the collected and calibrated verification data set to the adaptive fitting model to update the model parameters.

2. The automated data fitting method combined with device position calibration according to claim 1, wherein The collecting the multi-source sensor data set of the target device in a three-dimensional space includes: Real-time collecting the displacement increment data of the target device on the X-axis, Y-axis, and Z-axis by a laser ranging sensor array deployed on the motion plane of the target device; Invoking an inertial measurement unit to obtain the angular velocity offset and acceleration offset of the target device, and recording the temperature interference parameter and electromagnetic interference intensity at each sampling moment; Performing timestamp alignment processing on the displacement increment data, the angular velocity offset, and the acceleration offset to obtain synchronized motion trajectory data; Performing normalization processing on the temperature interference parameter and the electromagnetic interference intensity to obtain normalized environmental interference parameters; Performing spatial coordinate mapping on the synchronized motion trajectory data and the normalized environmental interference parameters to generate a multi-source sensor data set including device motion trajectory data, position coordinate offsets, and environmental interference parameters.

3. The automated data fitting method combined with equipment position calibration according to claim 1, wherein The performing feature extraction on the multi-source sensor data set to generate a target feature set includes: Extracting the trajectory change rate within a continuous time window from the device motion trajectory data, and constructing spatial position association features based on the trajectory change rate, where the spatial position association features include a coordinate transformation matrix between adjacent calibration points; Performing frequency domain transformation processing on the position coordinate offsets, extracting low-frequency components as reference position features, and extracting high-frequency components as dynamic trajectory fluctuation features; Performing coupling degree analysis on the environmental interference parameters and the dynamic trajectory fluctuation features, calculating the contribution weight of environmental parameters to trajectory fluctuations, and generating environmental coupling features based on the contribution weight; Performing dimension alignment processing on the spatial position association features, the dynamic trajectory fluctuation features, and the environmental coupling features to generate a target feature set with a unified timestamp and spatial coordinate reference system.

4. The automated data fitting method combined with device position calibration according to claim 1, characterized in that The invoking a pre-trained adaptive fitting model, inputting the target feature set into the adaptive fitting model for dynamic parameter matching processing to generate an initial calibration parameter set of the target device includes: Inputting the spatial position association features into the first feature encoding layer of the adaptive fitting model to generate spatial position encoding vectors; Input the dynamic trajectory fluctuation feature into the second feature encoding layer of the adaptive fitting model to generate a trajectory fluctuation encoding vector; Input the environmental coupling feature into the third feature encoding layer of the adaptive fitting model to generate an environmental coupling encoding vector; Perform attention weight allocation processing on the spatial position encoding vector, the trajectory fluctuation encoding vector, and the environmental coupling encoding vector to obtain a weighted fusion feature vector; Call the parameter decoder of the adaptive fitting model to perform inverse mapping processing on the weighted fusion feature vector to generate an initial calibration parameter set including position compensation parameters, motion acceleration correction coefficients, and environmental interference suppression coefficients.

5. The automated data fitting method combined with device position calibration according to claim 4, characterized in that The performing attention weight allocation processing on the spatial position encoding vector, the trajectory fluctuation encoding vector, and the environmental coupling encoding vector to obtain a weighted fusion feature vector includes: Calculate a first similarity score between the spatial position encoding vector and the trajectory fluctuation encoding vector; Calculate a second similarity score between the trajectory fluctuation encoding vector and the environmental coupling encoding vector; Calculate a third similarity score between the environmental coupling encoding vector and the spatial position encoding vector; Perform softmax normalization processing on the first similarity score, the second similarity score, and the third similarity score to obtain a spatial position attention weight, a trajectory fluctuation attention weight, and an environmental coupling attention weight; Scale the spatial position encoding vector according to the spatial position attention weight, scale the trajectory fluctuation encoding vector according to the trajectory fluctuation attention weight, scale the environmental coupling encoding vector according to the environmental coupling attention weight, and perform element-wise fusion on the scaled spatial position encoding vector, the trajectory fluctuation encoding vector, and the environmental coupling encoding vector to generate a weighted fusion feature vector.

6. The automated data fitting method combined with equipment position calibration according to claim 4, wherein The generating a dynamic path sequence for device position calibration based on the initial calibration parameter set includes: Calculate the coordinate difference between the calibration path start point and the end point of the target device in three-dimensional space according to the position compensation parameter; Perform piecewise linear interpolation processing on the coordinate difference based on the motion acceleration correction coefficient to generate theoretical coordinate values of multiple intermediate calibration points; Dynamically adjust the theoretical coordinate values of each intermediate calibration point in combination with the environmental interference suppression coefficient to generate a dynamic path sequence including actual coordinate values, adjustment time intervals, and motion direction parameters; Perform collision detection and path smoothing processing on multiple consecutive calibration points in the dynamic path sequence to generate an optimized dynamic path sequence.

7. The automated data fitting method combined with device position calibration according to claim 6, wherein The dynamically adjusting the theoretical coordinate values of each intermediate calibration point in combination with the environmental interference suppression coefficient to generate a dynamic path sequence including actual coordinate values, adjustment time intervals, and motion direction parameters includes: Obtain the theoretical coordinate value of the current intermediate calibration point and the corresponding environmental interference suppression coefficient, and perform scalar multiplication operations on each spatial dimension component of the environmental interference suppression coefficient and the theoretical coordinate value to generate a three-dimensional environmental compensation vector for each intermediate calibration point; Determine the dynamic adjustment priorities in the X-axis, Y-axis, and Z-axis directions according to the component magnitudes of the three-dimensional environmental compensation vector, and superimpose the three-dimensional environmental compensation vector on the corresponding axial components of the theoretical coordinate values in the order from the highest priority to the lowest priority to generate the actual coordinate values; Calculate the adjustment time interval for each intermediate calibration point based on the ratio of the total magnitude of the three-dimensional environmental compensation vector to the preset reference adjustment speed, where the total magnitude of the three-dimensional environmental compensation vector is the result of the square root operation of the sum of the squares of the axial components; Determine the motion direction parameters according to the positive and negative signs of the axial components of the three-dimensional environmental compensation vector, and the motion direction parameters include the X-axis positive or negative movement flag, the Y-axis positive or negative movement flag, and the Z-axis positive or negative movement flag; Arrange the actual coordinate values, adjustment time intervals, and motion direction parameters in chronological order to generate the initial dynamic path sequence; Perform kinematic feasibility verification on each actual coordinate value in the initial dynamic path sequence. When it is detected that the displacement amount between adjacent actual coordinate values exceeds the product of the maximum motion speed of the device and the adjustment time interval, perform geometric scaling on the adjustment time interval until the displacement constraint condition is satisfied; Update the time parameters in the initial dynamic path sequence according to the scaled adjustment time interval to generate the intermediate dynamic path sequence; Insert device attitude stability detection points in the intermediate dynamic path sequence. The insertion position of the attitude stability detection points is the median moment of the adjustment time interval between every two adjacent intermediate calibration points, and add an attitude maintenance instruction at the attitude stability detection points to suppress inertial drift; Reallocate the adjustment time intervals of each intermediate calibration point according to the number of inserted attitude stability detection points to ensure that the execution time of the attitude maintenance instruction is included in the total adjustment time interval; Merge the updated intermediate dynamic path sequence with the attitude stability detection points to generate a dynamic path sequence including actual coordinate values, adjustment time intervals, and motion direction parameters.

8. The automated data fitting method combined with device position calibration according to claim 4, characterized in that Trigger the target device to perform an automated position calibration operation according to the dynamic path sequence, and collect the calibrated verification data set and feedback it to the adaptive fitting model to update the model parameters, including: Convert the dynamic path sequence into a device control instruction set, and the control instruction set includes the motor drive parameters and servo control parameters of each calibration point; Real-time monitor the actual motion trajectory data when the target device performs the calibration operation according to the control instruction set; Collect the actual position coordinate data after calibration, and calculate the deviation from the theoretical coordinate values in the dynamic path sequence to generate position calibration error data; Combine the actual motion trajectory data, position calibration error data, and real-time environmental interference parameters into a verification data set; Input the verification data set into the error backpropagation link of the adaptive fitting model to update the weight parameters of the first feature encoding layer, the second feature encoding layer, and the parameter decoder.

9. The automated data fitting method combined with equipment position calibration according to claim 8, characterized in that The error backpropagation link that inputs the verification data set into the adaptive fitting model and updates the weight parameters of the first feature encoding layer, the second feature encoding layer, and the parameter decoder includes: Extracting the root mean square error between the actual motion trajectory data and the theoretical trajectory data from the verification data set as the first loss function; Calculating the maximum offset of the position calibration error data in different spatial dimensions as the second loss function; Generating a third loss function based on the prediction error of the environmental coupling coding vector according to the real-time environmental interference parameters; Performing weighted summation on the first loss function, the second loss function, and the third loss function to obtain a comprehensive loss value; Using the adaptive momentum optimization algorithm to update the weights of the fully connected layer and the attention allocation parameters of the adaptive fitting model according to the comprehensive loss value; Among them, the parameter update process of the adaptive momentum optimization algorithm includes: Calculating the first gradient component of the comprehensive loss value with respect to the spatial position coding vector; Calculating the second gradient component of the comprehensive loss value with respect to the trajectory fluctuation coding vector; Calculating the third gradient component of the comprehensive loss value with respect to the environmental coupling coding vector; Determining the momentum update direction of each parameter node according to the exponential weighted summation result of the historical gradient moving average and the current gradient; Iteratively updating the weights of the fully connected layer of the adaptive fitting model according to the momentum update direction, and at the same time adjusting the temperature coefficient in the attention allocation parameters to control the smoothness of the softmax normalization.

10. An automated data fitting system combined with device position calibration, characterized in that, The automated data fitting system combined with device position calibration includes a processor and a memory. The memory is connected to the processor. The memory is used to store programs, instructions, or codes, and the processor is used to execute the programs, instructions, or codes in the memory to implement the automated data fitting method combined with device position calibration according to any one of claims 1-9 above.

Images