Model robot limb control method and system

Abstract

The application discloses a model robot limb control method and system, the method comprises the following steps: collecting the standardized fabric parameters of a target garment; inputting the standardized fabric parameters into a pre-constructed structure-fabric mapping model to obtain basic configuration data; superimposing category correction on the basic configuration data to obtain final configuration data; performing feedforward calculation on the final configuration data to generate feedforward configuration instructions for each actuator and drive each actuator to perform an action; comparing the actual execution result with the corresponding feedforward configuration instructions to obtain an execution deviation; iteratively correcting the feedforward configuration instructions for each actuator according to the execution deviation until the execution configuration deviation of all actuators meets a threshold value, and then the iteration is terminated. The method can generate an adaptive limb configuration according to the feedforward of the fabric parameters, cooperate with the encoder closed-loop iterative correction of the execution deviation, avoid garment wrinkles from the source, and balance the one-time forming speed and the positioning accuracy.

Description

Technical Field

[0001] This application relates to the field of robot control, specifically to a method and system for controlling the limbs of a mannequin robot. Background Technology

[0002] In the field of apparel e-commerce and brand photography, the quality of product display images directly affects purchasing decisions. Traditional methods using live models suffer from high costs, scheduling difficulties, and poor consistency. In recent years, humanoid mannequin robots have been widely used in apparel display photography; however, existing solutions have the following problems: Most existing mannequin robots use a fixed degree of freedom configuration. Their limb span, height, and shoulder line are fixed or linked according to a fixed kinematic ratio. They do not sense or input fabric parameters, and they do not distinguish between the draping requirements of light and thin draping fabrics such as silk and chiffon and heavy fabrics such as thick cotton and denim. This causes clothing to pile up, be pulled or hang in the air on the limb surface, which seriously affects the display effect of clothing.

[0003] The more advanced methods in the existing solutions detect wrinkles and adjust the robot's posture through sensing after the clothes are put on. These solutions follow a feedback loop of putting on clothes, then detecting, then fine-tuning, and then verifying. On the one hand, they rely on an iterative solution of multiple rounds of detection, adjustment, and verification, which takes a long time and is difficult to meet the high-efficiency shooting requirements of e-commerce. On the other hand, they can only passively correct wrinkles that have already formed, and cannot avoid them from the source. Especially for thin fabrics with poor drape, once wrinkles are formed, they are difficult to completely eliminate in a single fine-tuning.

[0004] Furthermore, the fabric drape coefficient, weight, and elastic modulus directly determine the garment's drape under natural gravity, which is the physical basis for whether the garment fits well. However, none of the above solutions incorporate fabric parameters as decision inputs for the mannequin robot's limb configuration before dressing, resulting in a disconnect between the robot's configuration and the actual drape requirements of the fabric. Although existing known technologies include fabric drape coefficient prediction models using backpropagation neural networks, their applications are limited to virtual simulation or fabric rating and are not coupled with the actuator configuration control of the mannequin robot.

[0005] Therefore, there is a need for an efficient and integrated solution that generates a body shape that allows clothing to drape naturally based on the characteristics of different fabrics before the model robot is dressed, thus avoiding wrinkles and stretching from the source. Summary of the Invention

[0006] In view of the shortcomings of the prior art described above, the purpose of this application is to provide a method and system for controlling the limbs of a mannequin robot. By constructing a structure-fabric drape mapping model, the limb configuration vector of the mannequin robot is generated based on the fabric parameters before dressing, so as to achieve the effect of making the clothing hang naturally and fit without pulling under the action of gravity.

[0007] Firstly, this application provides a method for controlling the limbs of a model robot, including: Collect standardized fabric parameters of the target garment, input the standardized fabric parameters into a pre-built structure-fabric mapping model, and obtain the basic configuration data of each actuator of the robot. Obtain the category information of the target garment, and correct the basic configuration data based on the category information to obtain the final configuration data; The final configuration data is processed by feedforward calculation to generate feedforward configuration instructions for each actuator, and the actuators are driven to perform actions according to the feedforward configuration instructions. The actual execution results are collected in real time by the encoder, and the actual execution results are compared with the corresponding feedforward configuration instructions to obtain the execution deviation. The feedforward configuration instructions of each actuator are iteratively corrected based on the execution deviation until the execution deviations of all actuators meet the threshold, thus completing the iteration.

[0008] Secondly, this application also provides a model robot limb control system, including: The fabric parameter acquisition module is used to acquire the basic fabric data of the target garment and to normalize the basic fabric data to obtain standardized fabric parameters. The mapping inference module is used to map the standardized fabric parameters into limb configuration data; The feedforward calculation module is used to perform feedforward calculation on the limb configuration data and generate feedforward configuration instructions for each actuator. The drive execution module is used to drive each actuator to perform actions according to the feedforward configuration instructions; The closed-loop verification module is used to collect the actual execution results in real time through the encoder, compare the actual execution results with the expected execution results corresponding to the feedforward configuration instructions, obtain the execution deviation, and iteratively correct the feedforward configuration instructions of each actuator based on the execution deviation.

[0009] Thirdly, this application also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method described above.

[0010] Fourthly, this application also provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the method described above.

[0011] Fifthly, this application also provides a computer product, including a computer program that, when executed by a processor, implements the method described above.

[0012] The aforementioned mannequin robot limb control method and system, by normalizing fabric parameter input and superimposing clothing category correction amounts, generates differentiated configurations for fabrics with different drape characteristics, improving the adaptability and generalization ability to different types of clothing; by generating limb configurations based on fabric parameters before dressing, the clothing drapes naturally and fits snugly without pulling under gravity, avoiding wrinkles and pulling in the "dress first, detect later" approach from the source; by performing feedforward calculations through robot kinematic mapping, feedforward configuration instructions are generated and drive each actuator to perform actions, making the configuration generation path direct and without intermediate redundant links, thus improving control efficiency; by using the actuator's built-in encoder to iteratively correct execution deviations in a closed loop, convergence to the preset threshold range is achieved in only one or two rounds, balancing the speed of configuration execution and the accuracy of placement. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is an application environment diagram of a mannequin robot limb control method in one embodiment; Figure 2 A flowchart of a mannequin robot limb control method in one embodiment; Figure 3 This is a structural block diagram of the mannequin robot limb control system in one embodiment; Figure 4 This is a pipeline architecture diagram of the mannequin robot limb control system in one embodiment; Figure 5 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0016] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0017] This application provides a method for feedforward control of limb configuration in a mannequin robot, which can be applied to, for example... Figure 1 In the application environment shown, the real-time control host 102 communicates with the actuator system 104 via the fieldbus 101 and interacts with the edge server 103 via the network. The actuator system 104 includes 12 electric cylinder actuators, shoulder torsion springs, and ankle cross shaft mechanisms. Each electric cylinder actuator has a built-in encoder for real-time feedback of the actual stroke. The fabric database 105 stores fabric parameters for various clothing fabrics and is integrated on the edge server 103. The real-time control host 102 executes the model robot limb configuration feedforward control method provided in this embodiment. After execution, it outputs a shooting ready signal and pushes it to the shooting system 106.

[0018] In one exemplary embodiment, such as Figure 2 As shown, a feedforward control method for the limb configuration of a mannequin robot is provided, which is then applied to... Figure 1 Taking the real-time control host 102 as an example, the explanation includes the following steps 201 to 205. Wherein: Step 201: Collect standardized fabric parameters of the target garment, input the standardized fabric parameters into the pre-built structure-fabric mapping model, and obtain the basic configuration data of each actuator of the robot.

[0019] Standardized fabric parameters refer to a set of parameters characterizing the mechanical properties of clothing fabrics, including weight per unit area, drape coefficient, elastic modulus, thickness, and fabric-limb surface friction coefficient. Weight per unit area refers to the mass of fabric per unit area, measured in grams per square meter; drape coefficient refers to the degree to which the fabric bends under natural gravity, ranging from 0 to 1, with smaller values ​​indicating softer and more draping fabric; elastic modulus is the ratio of stress to strain within the elastic deformation range of the fabric, reflecting the fabric's resistance to deformation; thickness refers to the physical thickness of the fabric, measured in millimeters; and the fabric-limb surface friction coefficient is the coefficient of friction between the fabric surface and the surface of the mannequin robot's limbs.

[0020] Specifically, basic fabric data for the target garment is collected, including weight, drape coefficient, elastic modulus, thickness, and fabric-limb surface friction coefficient. This data is then normalized to map each parameter to a uniform numerical range, resulting in standardized fabric parameters. The structure-fabric mapping model is a regression function from fabric mechanical properties to actuator configuration quantities. A sample set is constructed by collecting various fabric parameters and obtaining the corresponding optimal limb configurations. The model is trained using fabric parameters as input and the optimal limb configuration as output until the error between the model output and the samples meets the preset accuracy requirements.

[0021] For example, the real-time control host 102 receives the basic fabric data of the target garment input by the operator through the human-machine interface, or reads the fabric parameters corresponding to the target garment through the fabric database 105 to obtain the fabric parameter vector. ,in For weight, For the sag coefficient, For elastic modulus, For thickness, This refers to the coefficient of friction between the fabric and the limb surface. The real-time control host 102 normalizes each parameter according to its rated range, mapping each parameter to a uniform numerical range. To obtain standardized fabric parameters .

[0022] Furthermore, the real-time control host 102 inputs the standardized fabric parameters P into the pre-constructed structure-fabric drape mapping model M, whose linearized reference form is as follows: ,in For the baseline configuration, The mapping coefficient matrix is ​​6×5, with elements determined by calibration experiments. Basic configuration data, represented as a six-dimensional limb configuration vector, is obtained through model inference. .in, To extend the electric cylinder's stroke for leg extension, The electric cylinder is used to move the waist. This is the amount of gravity compensation for the shoulder torsion spring. The downward angle of the elbow. Ankle pitch angle, The ankle tilt angle is defined by the mapping model M, which is obtained through offline calibration. Its output dimension corresponds one-to-one with the actuator's degrees of freedom.

[0023] Among them, the fabric parameter vector Each parameter can have a coupled effect on multiple configuration components, but a dominant influencing component exists in the calibration data. Sensitivity assumptions used for model initialization and explanation: gram weight. Main impact and And usually when g increases and Increased demand; overhang coefficient d mainly affects and ,and An increase in elastic modulus indicates a decrease in the fabric's drape ability, leading to a corresponding increase in the demand for drape configurations; Main impact and and to Secondary effects; thickness It is mainly affected by the fabric's bending stiffness, weight, and contact condition. , and coefficient of friction Main impact and and to , It has a secondary effect.

[0024] In a preferred embodiment, the structure-fabric drape mapping model M is implemented using an RBF radial basis function network, with a 5-node input layer. The hidden layer has 24 RBF basis function nodes; the output layer has 6 linear nodes, corresponding to... This means that a 5-24-6 network structure is used. The hidden layer centers are determined by the clustering results of the labeled samples, the basis function widths are determined based on the sample spacing or cross-validation, and the output layer weights are obtained by least squares with regularization.

[0025] The offline calibration process of the mapping model M includes: S1, establishing a fabric sample database. It is recommended to select at least 30 to 40 fabric samples with different weights, drape coefficients, elastic moduli, thicknesses, and surface friction characteristics, and record the unique number and five-dimensional parameters of each sample. S2, conduct multiple sets of candidate configuration experiments for each sample, for and Within a preset safety range, conduct orthogonal experiments, D-optimal experiments, or phased searches; S3, calculate the garment morphology evaluation index J for each configuration, where J can be weighted by wrinkle score, shoulder tension score, hem height difference, left-right symmetry error, and human posture deviation; S4, record the configuration that minimizes J and satisfies the robot's mechanical limits as the optimal configuration for that fabric / garment category sample. Thus, training data is obtained. , , S5, divide the samples into training, validation, and test sets. Use the training set to determine the RBF center, width, and output layer weights, and use the validation set to determine the number of network nodes and hyperparameters. S6, calculate the six-dimensional configuration prediction error using the independent test set. If the error exceeds the preset threshold, add boundary samples or recalibrate. S7, perform a feedforward configuration on the new fabric on the robot and verify it through encoder and garment shape evaluation. When several consecutive batches of samples exceed the threshold, enter the recalibration process.

[0026] Step 202: Obtain the category information of the target garment, and correct the basic configuration data based on the category information to obtain the final configuration data.

[0027] The category information refers to the type of garment to which the target clothing belongs, including bottoms, dresses, tops, or trousers / skirts. Correction refers to adding correction amounts to the corresponding configuration components in the basic configuration data based on the garment category, so that the limb configuration adapts to the draping requirements of different garment categories. The final configuration data includes six-dimensional limb configuration vectors for each actuator of the robot. These vectors include the leg extension cylinder stroke, waist shift cylinder displacement, shoulder torsion spring gravity compensation, elbow drop angle, ankle pitch angle, and ankle tilt angle. The leg extension cylinder stroke determines the lower limb extension distance; the waist shift cylinder displacement determines the torso's anterior-posterior alignment; the shoulder torsion spring gravity compensation counteracts the downward pull of the garment's weight on the shoulder line; the elbow drop angle controls the elbow joint's drop angle; and the ankle pitch and tilt angles control the ankle cross-axis rotation angles in the pitch and tilt directions, respectively.

[0028] Specifically, by acquiring the category information of the target garment, determining the corresponding category correction amount, and then adding the category correction amount to the basic configuration data, the final configuration data is obtained. For example, the operator selects the category information of the currently displayed garment through a human-machine interface. The real-time control host determines the corresponding category correction amount based on the product category information. When the target garment is a bottom, apply corrections corresponding to the bottom category to the leg extension cylinder stroke, pitch angle, and ankle tilt angle; when the target garment is a dress, apply corrections corresponding to the dress category to the leg extension cylinder stroke and waist shift cylinder displacement; when the target garment is a top, apply corrections corresponding to the top category to the shoulder torsion spring gravity compensation and elbow drop angle; when the target garment is a trouser leg or skirt hem, apply corrections corresponding to the trouser leg or skirt hem category to the ankle pitch angle and ankle tilt angle; the real-time control host superimposes the category corrections onto the basic configuration data to obtain the final configuration data. ,in Based on basic configuration data, According to category A definite six-dimensional positive vector.

[0029] Step 203: Perform feedforward calculation on the final configuration data to generate feedforward configuration instructions for each actuator, and drive each actuator to perform actions according to the feedforward configuration instructions.

[0030] Feedforward calculation refers to the process of converting limb configuration vectors into target strokes, target displacements, or target joint angles for each actuator using a preset function. Feedforward configuration commands are control commands containing target parameters for each actuator, used to drive each actuator to perform actions according to the target configuration.

[0031] Specifically, the six-dimensional limb configuration vector is converted into the corresponding action type and action parameters for each actuator through a preset function, and feedforward configuration instructions are generated based on the action type and action parameters for each actuator. For example, the real-time control host 102 uses a robot kinematics mapping function... For limb configuration vectors Perform feedforward calculations to generate feedforward configuration instructions for each actuator. The feedforward configuration instructions include the target stroke of each electric cylinder, the target compensation amount of the shoulder torsion spring, and the target joint angle of the ankle cross axis.

[0032] Furthermore, the real-time control host 102 sends feedforward configuration commands to the corresponding actuators via the fieldbus 101, driving 12 electric cylinders, shoulder torsion springs, and ankle cross shafts to perform actions. The thigh and calf use carbon fiber tubes in conjunction with extension electric cylinders, totaling four cylinders on both sides, rated at 300N·50mm, equipped with ball linear bearings, and their stroke... The lower limb extension is determined to give the skirt or tube a natural outward flare and landing shape; a displacement electric cylinder is installed at the waist, its displacement... The abduction distance of the lower limbs is combined with the calculation of the "abduction distance-height" drop configuration to achieve the anterior-posterior alignment of the torso; an electric cylinder is installed on each side of the shoulder to complete the shoulder extension freedom movement, while the shoulder torsion spring is set with the compensation amount according to the fabric feed. This counteracts the downward pull of the garment's weight on the shoulder line, achieving zero tension on the shoulder line; the elbow joint achieves natural hanging through passive freedom, allowing the sleeve to form... The ankle joint's pitch and roll angles are driven by two vertical electric cylinders, totaling four electric cylinders for both feet. and It generates rotation angles in the pitch and lateral directions, so that the trouser hem or skirt hem lands naturally and without tilting.

[0033] Step 204: The actual execution results are collected in real time by the encoder, and the actual execution results are compared with the corresponding feedforward configuration instructions to obtain the execution deviation.

[0034] The encoder refers to the position / displacement sensor built into each electric cylinder actuator, used to provide real-time feedback on the actuator's actual stroke, displacement, or angle. Execution deviation refers to the difference between the actual execution result of each actuator and its target configuration, including stroke deviation, displacement deviation, and angle deviation.

[0035] Specifically, the actual execution results are collected by the built-in encoders of each actuator to obtain the expected execution results corresponding to the feedforward configuration instructions. The actual execution results are then compared with the expected execution results to obtain the execution deviation. For example, during execution, each actuator's built-in encoder collects the actual execution results in real time. The actual execution results are fed back to the real-time control host 102 via the fieldbus 101. The real-time control host 102 then transmits the actual execution results... Expected execution result corresponding to feedforward configuration instructions Compare and calculate the execution deviation of each actuator separately. Execution deviations include stroke deviation, displacement deviation, and angular deviation.

[0036] Step 205: Iteratively correct the feedforward configuration instructions of each actuator according to the execution deviation until the execution deviation of all actuators meets the threshold to complete the iteration.

[0037] Specifically, the correction amount is calculated by combining the execution deviation and the preset correction gain. Based on the correction amount, the feedforward configuration instructions of each actuator are iteratively corrected round by round according to the iterative learning law. When the execution deviation of all actuators is less than the deviation threshold, the iterative process terminates and the system determines that the configuration has converged.

[0038] For example, the real-time control host 102 adjusts the execution deviation. Adjust gain according to preset settings The feedforward configuration instructions of each actuator are modified round by round, and the modification method is as follows: ,in To correct the gain, the value ranges from 0 to 1, and the correction amplitude is determined by the execution deviation and the preset correction gain. The real-time control host 102 reissues the corrected feedforward configuration command to the actuators for execution, and compares the actual execution results again by collecting data from the encoder. When the absolute value of the execution deviation of all actuators is less than the preset deviation threshold, the correction is completed. Or the number of iterations reaches the preset maximum number of iterations. The real-time control host 102 determines that the limb configuration has reached the target state, terminates the iteration process, and pushes a shooting ready signal to the shooting system 106 to execute the shooting.

[0039] In the aforementioned method for controlling the limbs of a mannequin robot, fabric parameters are collected and mapped to basic configuration data. Then, corrections are added based on clothing category information to obtain the final configuration data. This allows different types of clothing with the same fabric parameters to achieve differentiated limb configuration adaptations, improving the fit and natural drape of the clothing display. By generating the limb configuration based on fabric parameters before dressing, wrinkles and stretching are avoided from the source, overcoming the difficulty in eliminating wrinkles caused by traditional methods that involve dressing first and then detecting and correcting. Feedforward calculations are performed using robot kinematic mapping to generate feedforward configuration commands and drive each actuator to perform actions. This makes the configuration generation path direct, without intermediate redundant links, improving control efficiency. The encoder's closed-loop iterative correction of execution deviations only requires one or two rounds to converge to a preset threshold range, ensuring actuator positioning accuracy and balancing the speed and positioning accuracy of one-time configuration formation.

[0040] Based on the same inventive concept, this application also provides a model robot limb control system for implementing the above-mentioned model robot limb control method. The solution provided by this system is similar to the solution described in the above method. Therefore, the specific limitations in one or more system embodiments provided below can be found in the limitations of the model robot limb control method above, and will not be repeated here.

[0041] In one exemplary embodiment, such as Figure 3 As shown, a mannequin robot limb control system is provided, including: a fabric parameter acquisition module 301, a mapping inference module 302, a feedforward calculation module 303, a drive execution module 304, and a closed-loop verification module 305, wherein: The fabric parameter acquisition module 301 is used to acquire the fabric parameters of the target garment and normalize the fabric parameters to obtain standardized fabric parameters. The mapping inference module 302 is used to map the standardized fabric parameters into limb configuration data; The feedforward calculation module 303 is used to perform feedforward calculation on the limb configuration data and generate feedforward configuration instructions for each actuator. The drive execution module 304 is used to drive each actuator to perform actions according to the feedforward configuration instructions; The closed-loop verification module 305 is used to collect the actual execution results in real time through the encoder, compare the actual execution results with the corresponding feedforward configuration instructions to obtain the execution deviation, and iteratively correct the feedforward configuration instructions of each actuator according to the execution deviation.

[0042] The mannequin robot limb control system proposed in this application can acquire and normalize the fabric parameters of the target garment through a fabric parameter acquisition module, map them into limb configuration data through a mapping inference module, generate feedforward configuration instructions for each actuator through a feedforward calculation module, and drive the actuators to perform actions. Then, through a closed-loop verification module, the execution deviation is iteratively corrected in real time through encoder feedback until the positioning accuracy of all actuators meets the threshold. The solution provided in this application can generate an adaptive limb configuration based on the fabric parameters, so that the garment drapes naturally and fits without stretching under gravity, avoiding the generation of garment wrinkles from the source. The encoder closed-loop iterative correction balances the one-time forming speed and positioning accuracy.

[0043] The various modules in the aforementioned mannequin robot limb control system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the computer device's memory as software, so that the processor can call and execute the corresponding operations of each module.

[0044] In an exemplary test embodiment, the system based on the mannequin robot limb control system was tested in a standard indoor environment, specifically: ambient temperature 23±2℃, relative humidity 45%±10%RH, uniform diffuse reflection illumination of 200~800Lux in a standard shooting studio, power supply 220V±10% / 50Hz with UPS protection; the test used prototype number ZM-ROBOT-2026-005, system version V1.0.0-Alpha, and mapping model version DrapeMap-v1.0; the prototype under test contained 12 time-dependent electric cylinders, including a lower limb extension electric cylinder, a waist shifting electric cylinder, a shoulder torsion spring mechanism, and an ankle cross shaft mechanism.

[0045] In the calibration and verification test of the mapping model, samples of five types of standard fabrics—cotton, chemical fiber, chiffon, knitwear, and denim—were collected, and their fabric parameter vectors were measured. And normalized to the rated range of each dimension. Three experienced pattern makers labeled the baseline limb configuration for each sample. The mean value is used as the baseline. A structure-fabric drape mapping model M is fitted using offline data, and predictions are made on the reserved validation set. Calculate the prediction bias for each dimension.

[0046] In five types of fabrics and a total of 250 validation samples, the configuration prediction bias of the structure-fabric drape mapping model M is: leg elongation stroke. Typical ±0.6mm, maximum ±0.9mm (target ≤ ±1.0mm); waist displacement Typical ±0.5mm (target ≤±0.8mm); joint angle Typical ±0.5°, maximum ±0.7° (target ≤ ±0.8°).

[0047] In the actuator positioning accuracy test, the 12 time-dependent electric cylinders, under 20 target poses and 10 repeated acquisitions each, achieved a maximum positioning deviation of ±0.05mm and a repeatability of ±0.04mm, meeting the target of ≤±0.05mm. The waist-shifting electric cylinder achieved a displacement accuracy of ±0.04mm. The shoulder torsion spring mechanism and ankle cross-axis angle accuracy both met the target of ≤±0.5°. The average closed-loop sampling period for all states was 0.8ms, lower than the upper limit of 1ms. In the feedforward + encoder closed-loop convergence test, the initial target is solved by the feedforward solution of the mapping model M. It drives execution and records the actual value fed back from the electric cylinder's built-in encoder. According to the iterative learning law （ ,0< ≤1) Iterative correction, statistics reach termination condition. or The required number of iterations k and the final positioning error. Feedforward solution for the initial target. After closed-loop iterative correction by the encoder, the typical convergence iteration is 8 times, with a maximum of 11 times (≤12 times for the target), and the final positioning error is <0.04°.

[0048] In the component configuration achievement rate test, the prototype was dressed in four types of standard clothing: bottoms, dresses, tops, and trouser hems / skirts. Each type was displayed 50 times. Five evaluators independently judged whether the configuration was achieved based on three criteria: "natural drape of clothing, no obvious wrinkles, and a close fit to the shoulder / waistline." The configuration achievement rates for each of the four types of clothing (bottoms, dresses, tops, and trouser hems / skirts) were as follows: bottoms 96.0%, dresses 94.0%, tops 96.0%, and trouser hems / skirts 92.0%, with an overall achievement rate of 94.5%, all meeting the target of ≥92%.

[0049] In the above tests, the average time for the entire configuration to be ready was as follows: power-on self-calibration + clothing parameter building average 2.0s, fabric parameter normalization + mapping inference average 0.4s, feedforward calculation average 0.3s, encoder closed-loop verification iteration average 4.5s, shooting trigger average 0.3s, AI weakening processing average 1.6s, total average 9.1s, maximum 9.6s, which meets the internal target requirement of ≤10s.

[0050] In the aforementioned feedforward control system for the robot's limb configuration, through mapping model calibration and verification testing, the configuration prediction deviations in all dimensions are better than the preset target values, indicating that the mapping model can accurately map fabric parameters to the robot's limb configuration, providing an accurate configuration benchmark for subsequent feedforward control. Through actuator positioning accuracy testing, the positioning accuracy and repeatability of each electric cylinder actuator, as well as the accuracy of each joint angle, all meet design requirements, ensuring precise execution of feedforward configuration commands. Through encoder closed-loop convergence testing, the number of convergence iterations and the termination positioning error both meet design targets, balancing one-time forming speed and placement accuracy, and shortening configuration adjustment time. Through component configuration achievement rate testing, the achievement rates of all four configuration types meet design target requirements, effectively adapting to the draping needs of different types of clothing. Through full-process configuration time testing, the total process time meets internal target requirements, demonstrating efficient engineering response speed and meeting the high-efficiency shooting needs of e-commerce clothing.

[0051] In one exemplary embodiment, such as Figure 4 As shown, an architecture diagram of the core control pipeline of a model robot limb control system is provided. The control pipeline adopts a closed-loop control structure of "parameter acquisition → mapping inference → feedforward calculation → drive execution → closed-loop verification → morphological evaluation → readiness determination". Each module coordinates serially with a 1ms period through a shared memory ring buffer.

[0052] Specifically, the parameter acquisition module M1 acquires the standardized fabric parameters of the target garment. Product category information and will and Transferred to mapping inference module M2; The mapping inference module M2, based on the received standardized fabric parameters Product category information The basic configuration data is obtained through the structure-fabric drape mapping model, and then based on the product category information. Determine the category correction amount Superimposed on the basic configuration data, a six-dimensional limb configuration vector is obtained. Then Transmitted to feedforward solution module M3; The feedforward solution module M3 uses the robot's kinematic mapping function to convert the limb configuration vectors Solving into feedforward configuration instructions for each actuator And transmit it to the driver execution module M4; The drive execution module M4 executes the feedforward configuration instructions. It drives 12 electric cylinders, shoulder torsion springs, and ankle cross shafts to perform actions. During the actuator's operation, the encoder built into the actuator collects the actual execution results in real time. And feed back to the closed-loop verification module M5; The closed-loop verification module M5 will display the actual execution results. With current feedforward configuration instructions Compare and calculate execution deviation and according to the iterative learning law After the feedforward configuration instructions are corrected round by round and the closed-loop verification is completed, the closed-loop verification module M5 transmits the final converged actual position information to the morphological evaluation module M6. The morphological evaluation module M6 evaluates the final convergence location. The system determines whether the current garment's form on the mannequin robot meets the requirements. When the garment's form evaluation index does not meet the preset form threshold, the form evaluation module M6 calculates and generates a configuration correction amount driven by form error. and will Feedback is sent to the mapping inference module M2, which then performs the following steps: Readjust configuration vectors This forms a complete closed loop; when the clothing form evaluation index meets the preset form threshold, the form evaluation module M6 transmits the evaluation result to the ready judgment module M7. Readiness determination module M7 receives the execution deviation from closed-loop verification module M5 and the evaluation results from morphological evaluation module M6. When the actuator positioning error... When the preset accuracy threshold and the morphology evaluation both meet the preset morphology threshold, the ready determination module M7 outputs a shooting ready signal.

[0053] In the aforementioned feedforward control system for the limb configuration of the model robot, standardized fabric parameters and category information are acquired through a parameter acquisition module. A mapping inference module then generates a limb configuration vector adapted to the current garment in one go. The feedforward calculation module directly obtains the feedforward configuration instructions for each actuator, significantly improving the efficiency of configuration generation. A closed-loop verification module compares the actual execution results fed back from the encoder in real time with the current feedforward instructions and iteratively corrects the instructions until the execution deviation meets a preset accuracy threshold, effectively ensuring the actuator's placement accuracy and balancing one-time forming speed and placement accuracy. A morphology evaluation module evaluates the actual drape of the garment and automatically generates a configuration correction amount when the preset morphology threshold is not met, feeding it back to the mapping inference module to readjust the configuration, further ensuring the visual effect of the garment display and achieving full-link closed-loop control from parameter input to morphology effect. A ready-to-shoot module only outputs a shooting ready signal when both the actuator placement error and the garment morphology evaluation meet preset thresholds, ensuring that the system guarantees placement accuracy while ensuring the garment display effect meets shooting requirements, providing reliable conditions for subsequent shooting.

[0054] In one exemplary embodiment, a computer device is provided, which may be a real-time control host, and its internal structure diagram may be as follows: Figure 5 As shown, the computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface communicates with 12 electric cylinder actuators, shoulder torsion springs, and ankle cross-axis mechanisms via a fieldbus (such as FDCAN), and can interact with an edge server via Ethernet or WebSocket. When executed by the processor, the computer program implements a method for controlling the limbs of a mannequin robot. The display unit of the computer device is used to display a human-computer interaction interface. It can be a touch screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0055] Those skilled in the art will understand that Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0056] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps: Collect target clothing fabric data, and after data processing, obtain standardized fabric parameters. Input the standardized fabric parameters into a pre-built structure-fabric mapping model to obtain the basic configuration data of each actuator of the robot. Obtain the category information of the target garment, and modify the basic configuration data according to the category information to obtain the final configuration data; The final configuration data is processed by feedforward calculation to generate feedforward configuration instructions for each actuator, and each actuator is driven to perform actions according to the feedforward configuration instructions. The actual execution results are collected in real time by the encoder, and the actual execution results are compared with the corresponding feedforward configuration instructions to obtain the execution deviation. The feedforward configuration instructions of each actuator are iteratively corrected based on the execution deviation until the execution deviations of all actuators meet the threshold, at which point the iteration terminates.

[0057] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0058] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor: Collect target clothing fabric data, and after data processing, obtain standardized fabric parameters. Input the standardized fabric parameters into a pre-built structure-fabric mapping model to obtain the basic configuration data of each actuator of the robot. Obtain the category information of the target garment, and modify the basic configuration data according to the category information to obtain the final configuration data; The final configuration data is processed by feedforward calculation to generate feedforward configuration instructions for each actuator, and each actuator is driven to perform actions according to the feedforward configuration instructions. The actual execution results are collected in real time by the encoder, and the actual execution results are compared with the corresponding feedforward configuration instructions to obtain the execution deviation. The feedforward configuration instructions of each actuator are iteratively corrected based on the execution deviation until the execution deviations of all actuators meet the threshold, at which point the iteration terminates.

[0059] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0060] For example, the aforementioned computer-readable storage media may include, but are not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0061] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps: Collect target clothing fabric data, and after data processing, obtain standardized fabric parameters. Input the standardized fabric parameters into a pre-built structure-fabric mapping model to obtain the basic configuration data of each actuator of the robot. Obtain the category information of the target garment, and modify the basic configuration data according to the category information to obtain the final configuration data; The final configuration data is processed by feedforward calculation to generate feedforward configuration instructions for each actuator, and each actuator is driven to perform actions according to the feedforward configuration instructions. The actual execution results are collected in real time by the encoder, and the actual execution results are compared with the corresponding feedforward configuration instructions to obtain the execution deviation. The feedforward configuration instructions of each actuator are iteratively corrected based on the execution deviation until the execution deviations of all actuators meet the threshold, at which point the iteration terminates.

[0062] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0063] For example, the computer program product may exist in the form of a software installation package, application (APP), plug-in, patch, etc., and may be stored in a computer-readable storage medium or transmitted or distributed to multiple electronic devices via a network.

[0064] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0065] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for controlling the limbs of a mannequin robot, characterized in that, include: Collect standardized fabric parameters of the target garment, input the standardized fabric parameters into a pre-built structure-fabric mapping model, and obtain the basic configuration data of each actuator of the robot. Obtain the category information of the target garment, and correct the basic configuration data based on the category information to obtain the final configuration data; The final configuration data is processed by feedforward calculation to generate feedforward configuration instructions for each actuator, and the actuators are driven to perform actions according to the feedforward configuration instructions. The actual execution results are collected in real time by the encoder, and the actual execution results are compared with the corresponding feedforward configuration instructions to obtain the execution deviation. The feedforward configuration instructions of each actuator are iteratively corrected based on the execution deviation until the execution deviations of all actuators meet the threshold, thus completing the iteration.

2. The method according to claim 1, characterized in that, The standardized fabric parameters of the target garment collected include: Collect basic fabric data for the target garment, including weight, drape coefficient, elastic modulus, thickness, and fabric-limb surface friction coefficient. The basic fabric data is normalized to map each parameter to a uniform value range, thus obtaining the standardized fabric parameters.

3. The method according to claim 1, characterized in that, The structure-fabric mapping model is a regression function from fabric mechanical properties to actuator configuration quantities. The training process of the structure-fabric mapping model includes: A sample set was constructed by collecting various fabric parameters and obtaining the corresponding optimal limb configurations. The training is performed using the fabric parameters as input and the optimal limb configuration as output. The process continues until the error between the model output and the sample meets the preset accuracy requirements.

4. The method according to claim 1, characterized in that, The step of obtaining the category information of the target garment and correcting the basic configuration data based on the category information to obtain the final configuration data includes: Obtain the category information of the target garment and determine the category correction amount corresponding to the category information; The category correction is superimposed on the basic configuration data to obtain the final configuration data.

5. The method according to claim 4, characterized in that, The final configuration data includes six-dimensional limb configuration vectors for each actuator of the robot. These six-dimensional limb configuration vectors include the leg extension cylinder stroke, the waist shift cylinder displacement, the shoulder torsion spring gravity compensation, the elbow drop angle, the ankle pitch angle, and the ankle tilt angle. The process of superimposing the category corrections onto the basic configuration data includes: When the target clothing category is bottoms, apply the correction amount corresponding to the bottom category to the leg extension cylinder stroke, ankle pitch angle and ankle lateral tilt angle in the limb configuration vector; When the target clothing category is dress, apply the correction amount corresponding to the dress category to the leg extension electric cylinder stroke and waist shift electric cylinder displacement in the limb configuration vector; When the target clothing category is top, apply the correction amount corresponding to the top category to the shoulder torsion spring gravity compensation amount and elbow droop angle in the limb configuration vector; When the target clothing category is trouser leg or skirt hem, apply the correction amount corresponding to the trouser leg or skirt hem category to the ankle pitch angle in the limb configuration vector.

6. The method according to claim 5, characterized in that, The leg extension cylinder stroke determines the lower limb extension distance; the waist shifting cylinder displacement is used to determine the front-to-back alignment of the torso; the shoulder torsion spring gravity compensation is used to counteract the downward pull of the garment's weight on the shoulder line; the elbow drop angle is used to control the drop angle of the elbow joint; and the ankle pitch angle and ankle tilt angle are used to control the rotation angle of the ankle cross axis in the pitch and tilt directions, respectively.

7. The method according to claim 1, characterized in that, The step of performing feedforward calculation on the final configuration data to generate feedforward configuration instructions for each actuator includes: The six-dimensional limb configuration vector is converted into the action type and action parameters corresponding to each actuator through a preset function; The feedforward configuration instruction is generated based on the action type and action parameters corresponding to each actuator.

8. The method according to claim 1, characterized in that, The step of acquiring actual execution results in real time through the encoder, comparing the actual execution results with the corresponding feedforward configuration instructions to obtain the execution deviation includes: The actual execution results are collected through the built-in encoders of each actuator; Obtain the expected execution result corresponding to the feedforward configuration instruction; The execution deviation is obtained by comparing the actual execution result with the expected execution result. The execution deviation includes stroke deviation, displacement deviation and angle deviation.

9. The method according to claim 1, characterized in that, The step of iteratively correcting the feedforward configuration instructions of each actuator based on the execution deviation until the execution deviations of all actuators meet the threshold, and then terminating the iteration, includes: The correction amount is calculated by combining the execution deviation with the preset correction gain; According to the correction amount, the feedforward configuration instructions of each actuator are iteratively corrected round by round using the iterative learning law; When the execution deviation of all actuators is less than the deviation threshold, the iteration process terminates and the system determines that the configuration has converged.

10. A control system for the limbs of a mannequin robot, characterized in that, include: The fabric parameter acquisition module is used to acquire the basic fabric data of the target garment and to normalize the basic fabric data to obtain standardized fabric parameters. The mapping inference module is used to map the standardized fabric parameters into limb configuration data; The feedforward calculation module is used to perform feedforward calculation on the limb configuration data and generate feedforward configuration instructions for each actuator. The drive execution module is used to drive each actuator to perform actions according to the feedforward configuration instructions; The closed-loop verification module is used to collect the actual execution results in real time through the encoder, compare the actual execution results with the expected execution results corresponding to the feedforward configuration instructions, obtain the execution deviation, and iteratively correct the feedforward configuration instructions of each actuator based on the execution deviation.

11. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 9.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 9.

13. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 9.

Images