Intelligent medicine identifying and grabbing system based on robot arm

Abstract

The invention relates to the technical field of intelligent grabbing, in particular to an intelligent medicine identifying and grabbing system based on a robot arm, which comprises an identifying and checking module, a grabbing and positioning module, a grabbing and executing module, a buffering control module and a distributing and transferring module. According to the method, three groups of independent parameters including the medicine packaging size, the bar code character structure and the near-infrared reflectivity are extracted, a multi-dimensional recognition reference is established, the misrecognition risk caused by single-dimensional judgment is avoided, and multiple contact area indexes such as the curvature radius change rate, the texture mean value and the reflectivity gradient are calculated; the optimal adsorption point position is selected to improve the adsorption stability, the adsorption success rate is improved by matching the quality estimated value and the adsorption requirement grade of the medicine packaging unit with the applicable adsorption pressure grade, the grabbing integrity and the distribution stability are guaranteed, and the release time and the target goods shelf coordinate linkage setting are combined. And the certainty of a distribution path and the consistency of a control closed loop are enhanced.

Description

Technical Field

[0001] The present invention relates to the field of intelligent grasping technology, and in particular to a medicine intelligent recognition and grasping system based on a robot arm. Background Art

[0002] The field of intelligent grasping technology mainly involves core technologies such as machine vision, robotic arm control and path planning, target recognition and positioning, object posture estimation, end-effector control, and multi-sensor fusion, to achieve precise positioning, recognition, and manipulation of designated targets in automated processing scenarios. This technology is widely used in manufacturing, logistics, medicine, warehousing and other scenarios, and supports efficient grasping of target objects with complex structures, diverse types, and randomly placed positions. The system usually includes an image acquisition module, a visual recognition algorithm module, a grasping decision module, and an execution control module, aiming to achieve high-accuracy and high-robust dynamic target operations. Its key technical indicators include target recognition accuracy, grasping success rate, response time, system repeatability, error control range, etc.

[0003] The Intelligent Drug Identification and Grabbing System is an intelligent recognition and mechanical grasping system for automated drug distribution. It uses machine vision technology to identify various types of drugs, then uses control algorithms to drive a robotic arm to grasp and transfer them to designated distribution areas. It then collaborates with drug loading equipment to ensure orderly drug shelving. This system improves drug sorting and replenishment efficiency, reduces manual intervention, and is suitable for use in hospital pharmacies, smart medicine cabinets, automated drug refill lines, and other scenarios, enabling precise, efficient, and automated drug management.

[0004] Traditional grasping systems generally rely on image acquisition and visual algorithms for target recognition, and lack a comprehensive comparison mechanism for dimensional errors, character distribution structure and packaging materials, which can easily lead to misrecognition problems when the barcode information is obscured or the visual features are similar. Grasping and positioning mainly rely on visual image plane information, and fail to fully evaluate the degree of deformation of the target surface and the mechanical properties of the adsorption area, which will cause grasping offset or detachment. The execution actions are mostly based on fixed path control without considering the differences in landing materials and contact area constraints. In the landing scenario of soft-packaged medicines, there is a risk of excessive impact causing damage to the packaging, and the release action is not accurately linked to the transfer target, which will cause distribution positioning errors and affect the stability of the overall system. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a drug intelligent recognition and grasping system based on a robotic arm.

[0006] In order to achieve the above objectives, the present invention adopts the following technical solution: a drug intelligent identification and grasping system based on a robot arm, the system comprising:

[0007] The recognition and verification module receives the instruction issued by the refill task, calls three pieces of information about the candidate drug in the robot's field of view: the length, width, and height values, the total number of barcode characters and their spatial distribution, and the near-infrared reflectivity curve value. It determines whether the difference values of the three items are all below the set matching threshold and generates a matching recognition result.

[0008] Based on the matching and recognition results, the grasping and positioning module obtains the curvature radius change rate, grayscale texture mean and variance, and reflectivity gradient distribution value of each area on the packaging surface, calculates the adsorption effect score, selects the coordinates of the area with the highest score as the adsorption grasping contact point, and generates optimized adsorption coordinate information;

[0009] The grabbing execution module obtains the standard unit mass value of the drug packaging unit corresponding to the position based on the optimized adsorption coordinate information, selects the corresponding pressure value level, performs an adsorption grabbing action, moves the drug to the specified hovering position above, and generates an adsorption movement completion signal;

[0010] The buffer control module determines the drug packaging type based on the suction movement completion signal. Based on the Z-axis displacement value from the current robot end to the target drop point, it sets the Z-axis servo speed sequence according to the gradient, loads the deceleration command, and executes the slow drop, generating a buffer control sequence.

[0011] The present invention has improvements in that the matching recognition result includes the drug size offset value, the character structure consistency coefficient and the material spectrum difference ratio; the optimized adsorption coordinate information includes the target area spatial coordinates, the area surface roughness level and the surface reflection distribution range; the adsorption movement completion signal is specifically the adsorption action state identifier, the adsorption point pressure value record and the end effector feedback status code; the buffer control sequence includes the deceleration interval displacement length, the end speed change gradient and the release instantaneous tension level.

[0012] The present invention is improved in that the identification and verification module includes:

[0013] The benchmark extraction submodule receives the instruction issued by the replenishment task and extracts three sets of parameters from the drug demand record: the standard value of the outer packaging size, the barcode encoding character sequence, and the standard value of the packaging material reflectance spectrum. These parameters are stored as a size benchmark group, a character benchmark group, and a spectrum benchmark group, respectively. The benchmark value set is generated based on the contents of the three sets of parameter records.

[0014] The difference determination submodule, based on the reference value set, calls the three contents of the length, width and height dimensions of the candidate drug, the total number of barcode characters and the character spatial distribution, and the near-infrared reflectivity curve value, calculates the individual difference values between the length, width and height dimensions and the dimension reference group, calculates the structural correspondence between the character spatial distribution and the character reference group according to the position mapping rule, calculates the bandwidth difference range between the peak position of the spectral curve and the spectral reference group according to the main peak band comparison formula, and generates a multi-dimensional difference index set;

[0015] The identification and confirmation submodule calls the multidimensional difference indicator set, sets the difference value threshold standard and performs numerical threshold judgment on the three indicators respectively, performs Boolean cross operation on the judgment results, and if all three judgments are in the pass state, the current candidate drug is recorded as the identification pass object, and the matching recognition result is obtained.

[0016] The present invention is improved in that the grabbing and positioning module includes:

[0017] The point cloud extraction submodule obtains the matching recognition results and, based on the drug image marked as a candidate target, collects the surface three-dimensional point cloud information of the corresponding area of the image, divides the area into a grid according to the image coordinate system, integrates the coordinate channel and depth channel data in the three-dimensional point set, extracts the point density and normal distribution offset amplitude corresponding to the unit area of the grid, and generates spatial point cloud distribution information;

[0018] The surface parameter identification submodule calls the spatial point cloud distribution information, detects the curvature radius change rate, grayscale texture mean and variance, and reflectivity gradient distribution value corresponding to the surface of each area, determines whether the curvature radius change rate is in a flat state range based on the set reference value range, compares the grayscale texture mean and variance with the texture grading range, identifies whether the reflectivity gradient distribution value is less than the material adsorption critical value, calculates and obtains the adsorption adaptation score corresponding to the area, and generates an adsorption area scoring result;

[0019] The adsorption coordinate decision submodule is based on the adsorption area scoring results and the area index with the highest score, and locates the center coordinate point of the area in the image coordinate system. It combines the area center of gravity data to determine the actual coordinate position of the adsorption contact point in three-dimensional space, outputs the adsorption center coordinates and records the corresponding surface type identification, and establishes optimized adsorption coordinate information.

[0020] The present invention is improved in that the capture execution module includes:

[0021] The adsorption evaluation submodule aligns the terminal adsorption axis with the coordinate position based on the optimized adsorption coordinate information, obtains the standard unit mass value of the drug packaging unit corresponding to the coordinate, extracts the material coding type of the part and the average deviation value in the surface normal direction, combines the friction characteristic level of the packaging material and the stability level of the packaging structure, calculates the adsorption action level value required for adsorption grasping, matches the adsorption vacuum pressure value corresponding to the level, and generates adsorption requirement level information;

[0022] The path planning submodule calls the adsorption requirement level information and loads it into the end effector control unit according to the corresponding vacuum pressure parameter level. It reads the three-dimensional coordinate values of the hovering point set in the target distribution area, performs end trajectory planning based on the linear interpolation principle, constructs a three-axis coordinate path sequence, and generates a three-dimensional path coordinate group.

[0023] The motion control submodule controls the end effector to move sequentially according to the three-dimensional path coordinate group, gradually performs three-axis interpolation instructions through servo control, tracks the change in the offset distance between the current robot arm end position and the path node position, determines whether the offset value is lower than the threshold setting, enters the position holding state, and establishes an adsorption movement completion signal.

[0024] The present invention is improved in that the buffer control module includes:

[0025] The attribute recognition submodule confirms the completion of the end-grabbing action based on the adsorption movement completion signal, calls the three contents of the current drug packaging bottom surface contact area, the material Young's modulus reference value, and the material reflectivity main peak band value, extracts the matching material combination index from the material attribute database, determines the drug packaging type based on the material structure group, and obtains the packaging material category information;

[0026] The path setting submodule determines whether the packaging material is classified as a brittle material based on the packaging material category information, identifies whether the contact area of the packaging bottom surface is less than the lower area limit, and if so, sets the starting and ending point coordinates of the deceleration section in the terminal descent path. It calculates the length of the deceleration section by combining the current height of the terminal and the height of the landing point reference surface to generate deceleration path length information;

[0027] The speed adjustment submodule calls the deceleration path length information, combines the Z-axis displacement value from the current manipulator end to the target landing point, collects the material's Young's modulus, contact area and peak symmetry factor of the packaging reflection band, sets the exponential buffer adjustment coefficient and constructs a dynamic control factor, calculates and obtains the terminal Z-axis servo speed grading reference value, sets the speed sequence of each interval of the descent section based on the grading standard, loads the sequence to the controller output end, performs the displacement action, and establishes a buffer control sequence.

[0028] The present invention is improved in that the system further comprises:

[0029] The distribution and transfer module extracts the recorded release position coordinates according to the buffer control sequence, obtains the landing point status feedback signal at the coordinates after the medicine is dropped, and determines whether it is consistent with the coordinate grid number of the distribution unit. If it is consistent, it calls the assigned target shelf number and grid positioning code, controls the transfer rail to execute the corresponding path switching instruction, and generates a distribution completion identification code;

[0030] The distribution completion identification code specifically refers to the distribution shelf number, the grid docking position number and the track switching path index.

[0031] The present invention is improved in that the distribution and transfer module includes:

[0032] The release control submodule extracts the recorded release position coordinate information according to the buffer control sequence, binds the tension release moment of the end suction device to the release coordinate point, detects the spatial coincidence between the current end position of the manipulator and the coordinate point, and determines whether the coordinate match reaches the release judgment threshold. If the set conditions are met, the end effector is triggered to execute the tension release command and establish the release action trigger information;

[0033] The landing point verification submodule collects the landing point coordinate feedback signal after the drug is dropped according to the release action trigger information, obtains the grid number of the current coordinate, calls the defined distribution unit grid number sequence, and compares the two sets of numbers. If the landing point number is consistent with the target grid number, the current delivery behavior is marked as a positioning compliance state and positioning consistency identification information is generated;

[0034] The path switching submodule extracts the target shelf number and slot positioning code bound in the task allocation record based on the positioning consistency identification information judgment result, combines them into a path scheduling address code, controls the transfer rail to execute the path switching instruction according to the address code index instruction sequence, and establishes a distribution completion identification code.

[0035] Compared with the prior art, the advantages and positive effects of the present invention are:

[0036] In the present invention, a multi-dimensional recognition benchmark is established by extracting three sets of independent parameters, namely, drug packaging size, barcode character structure, and near-infrared reflectivity, to avoid the risk of misidentification caused by single-dimensional judgment. Accurate matching of drug targets is achieved through comparison of numerical deviation, structural similarity ratio, and main peak position bandwidth. Combined with the surface three-dimensional point cloud information, multiple contact area indicators such as the curvature radius change rate, texture mean, and reflectivity gradient are calculated, and the optimal adsorption point is selected to improve the adsorption stability. The quality valuation of the drug packaging unit and the adsorption demand level are matched to the applicable adsorption pressure level to improve the adsorption success rate. The Z-axis speed sequence is dynamically adjusted in combination with the landing point contact area and the packaging material properties to achieve landing point buffering control, ensure the grasping integrity and distribution smoothness, and combine the release timing with the target shelf coordinate linkage setting to enhance the certainty of the distribution path and the consistency of the control closed loop. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is a system flow chart of the present invention;

[0038] Figure 2 This is a flow chart of the identification and verification module of the present invention;

[0039] Figure 3 This is a flowchart of the grabbing and positioning module of the present invention;

[0040] Figure 4 A flowchart of the capture execution module of the present invention;

[0041] Figure 5 This is a flow chart of the buffer control module of the present invention;

[0042] Figure 6 Flowchart of the distribution transshipment module for the present invention. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0044] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings and are only for the convenience of describing the present invention and simplifying the description. They do not indicate or imply that the devices or elements referred to must have a specific direction, be constructed and operate in a specific direction, and therefore should not be understood as limiting the present invention. In addition, in the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0045] See also Figure 1 The present invention provides a technical solution: a drug intelligent recognition and grasping system based on a robot arm, the system comprising:

[0046] The recognition and verification module obtains the instruction issued by the replenishment task, extracts three sets of parameters from the drug demand record: the outer packaging size standard value, the barcode encoding character sequence, and the packaging material reflectance spectrum standard value, and stores them as three sets of benchmark data. It calls the three pieces of information of the length, width, and height dimensions of the candidate drugs in the robot's operating field of view, the total number of barcode characters and the character spatial distribution, and the near-infrared reflectivity curve value. According to the size deviation calculation rule, the numerical differences between the length, width, and height dimensions and the outer packaging size standard values are compared. According to the character structure comparison rule, the structural similarity ratio between the total number of characters and the character spatial distribution and the barcode encoding character sequence is compared. The difference values are compared based on the main peak position and bandwidth parameters of the near-infrared reflectivity curve value and the packaging material reflectance spectrum standard value. It determines whether the three difference values are all lower than the set matching threshold and generates a matching recognition result.

[0047] Based on the matching recognition results, the grasping and positioning module collects the surface 3D point cloud information of the drug image marked as the candidate target, obtains the curvature radius change rate, grayscale texture mean and variance, and reflectivity gradient distribution value of each area on the packaging surface, determines whether the curvature radius change rate is within the set flatness range, compares the grayscale texture mean and variance with the grading standard range, determines whether the reflectivity gradient distribution value is less than the critical threshold for material adsorption, calculates the adsorption effect score, selects the coordinates of the area with the highest score as the adsorption grasping contact point, and generates optimized adsorption coordinate information;

[0048] The grasping execution module aligns the end-effector's grasping axis with the coordinate position based on the optimized grasping coordinate information, obtains the standard unit mass value of the drug packaging unit corresponding to the position, and evaluates the grasping demand level based on the grasping location characteristics. It then calls the preset grasping vacuum pressure level table in the grasping actuator, selects the corresponding pressure value level, and loads it into the end-effector control signal to perform the grasping action. Based on the reference center point coordinate group of the target distribution area, the robot is controlled to execute a three-dimensional linear interpolation path translation action, moving the drug to the designated hovering position above, and generating a grasping movement completion signal.

[0049] The standard unit mass value is derived from the net weight of a single drug package provided in the drug packaging specification table; the adsorption vacuum pressure level table is a reference table of vacuum suction levels provided by the adsorption device manufacturer;

[0050] Based on the adsorption movement completion signal, the buffer control module confirms that the grasping action is completed, calls the three contents of the packaging bottom contact area, the material Young's modulus reference value, and the material reflectivity main peak band value of the drug packaging, retrieves the material combination from the material property database, determines the drug packaging category, and identifies whether the packaging bottom contact area is less than the set area lower limit. If the judgment is true, the deceleration interval range in the terminal descent path is set. Combined with the Z-axis displacement value from the current robot end to the target landing point, the deceleration interval length is calculated, and the Z-axis servo speed sequence is set according to the gradient. After loading the deceleration instruction, the slow descent is executed to generate the buffer control sequence;

[0051] The contact area of the package bottom surface is derived from the bottom surface area recognition of the vision system; the Young's modulus of the material is the coefficient of the material's resistance to elastic deformation; the Z-axis servo speed sequence is the speed step array of the servo controller controlling the end effector in the Z-axis direction;

[0052] The distribution and transfer module extracts the recorded release position coordinates according to the buffer control sequence, sets the release timing of the terminal adsorber in conjunction with the release coordinates, obtains the landing point status feedback signal at the coordinates after the medicine is dropped, and determines whether it is consistent with the coordinate grid number of the distribution unit. If it is consistent, it calls the assigned target shelf number and grid positioning code, controls the transfer rail to execute the corresponding path switching instruction, and generates a distribution completion identification code;

[0053] The matching recognition results include the drug size offset value, character structure consistency coefficient and material spectrum difference ratio. The optimized adsorption coordinate information includes the spatial coordinates of the target area, the surface roughness level of the area and the surface reflection distribution range. The adsorption movement completion signal is specifically the adsorption action status identification, the adsorption point pressure value record and the end effector feedback status code. The buffer control sequence includes the deceleration interval displacement length, the terminal speed change gradient and the release instantaneous tension level. The distribution completion identification code specifically refers to the distribution shelf number, the grid docking position number and the track switching path index.

[0054] See also Figure 2 , the identification and verification module includes:

[0055] The benchmark extraction submodule receives the instruction issued by the replenishment task and extracts three sets of parameters from the drug demand record: the standard value of the outer packaging size, the barcode encoding character sequence, and the standard value of the packaging material reflectance spectrum. These parameters are stored as a size benchmark group, a character benchmark group, and a spectrum benchmark group, respectively. The benchmark value set is generated based on the contents of the three sets of parameter records.

[0056] When obtaining the three sets of parameters of the outer packaging size standard value, barcode encoding character sequence, and packaging material reflectance spectrum standard value recorded in the drug replenishment task issuance instruction, the task number should be extracted through the task scheduling system interface and the corresponding drug demand item should be located. The packaging length, width, and height values, barcode information field, and reflectance spectrum identification number specified in the item should be read in turn. The number corresponds to a set of predefined reflectance spectrum database indexes. The storage format of each parameter is set to a standardized array type. Units must be unified before storage. The size values are uniformly encoded in millimeters and normalized to the interval [0,1]. The barcode sequence is converted into a character vector form. The main peak band of the curve corresponding to the reflectance spectrum information is classified within the range of [400,1100] nm and the main peak center wavelength and left and right half width values are extracted as triplets for encoding. The three sets of standard data are stored in the size benchmark group, character benchmark group, and spectrum benchmark group respectively. Before forming the benchmark value set, the three sets of parameters need to be checked for consistency. The verification method is to determine whether the size data is a three-dimensional vector and whether the character vector meets the length requirement. Between 8 and 16 bits, check whether the spectral triplet has a central peak. If there is a missing item, a parameter error prompt is returned and the recognition process is suspended. For example, in a task, the drug standard is set to 100mm in length, 50mm in width, and 30mm in height. The corresponding barcode sequence is ABX9802713, and the spectral main peak triplet is 740nm, ±35nm. This group of data is normalized and converted into a size group [1.00, 0.50, 0.30], a character group [65, 66, 88, ...], and a spectral group [0.671, 0.047, 0. 047], wherein the wavelength is normalized to the range of 0–1, and the character group conversion adopts ASCII encoding. Finally, the three groups of structures form a three-dimensional structure as a reference parameter set, wherein the setting of the size reference value is based on the nominal size in the drug packaging specification data manual, and the size standard calibrated in the product packaging design in the batch is preferably used as the main value, and its consistency within the allowable range of production deviation is confirmed by on-site measurement and sampling. After statistics, the standard size value set is obtained, and the allowable error setting range is ±5%, and this upper limit is used as the boundary during the normalization process;

[0057] The difference determination submodule, based on the benchmark value set, calls the three contents of the candidate drug's length, width, and height dimensions, the total number of barcode characters and the character spatial distribution, and the near-infrared reflectivity curve value. It calculates the individual difference values between the length, width, and height dimensions and the dimension benchmark group, calculates the structural correspondence between the character spatial distribution and the character benchmark group according to the position mapping rule, and calculates the bandwidth difference range between the peak position of the spectral curve and the spectral benchmark group according to the main peak band comparison formula, thereby generating a multi-dimensional difference index set.

[0058] When calling the information of candidate drugs based on the benchmark parameter set, the industrial vision system first needs to detect all visible drug areas in the current operation area image, obtain the spatial bounding box of the object to be compared through edge extraction and segmentation operations, and measure the boundary length of the packaging object in the X, Y, and Z directions, which is recorded as 90mm in length, 49mm in width, and 30mm in height. Its normalized size is [0.90, 0.49, 0.30]. When compared with the size benchmark group [1.00, 0.50, 0.30], the difference operations Δx = 0.10, Δy = 0.01, and Δz = 0.00 are performed in sequence. The three difference values are all less than the preset size threshold range [0.15, 0.05, 0.05]. The next step is to calculate the character structure correspondence. The barcode character is ABX9802712, which is equal to the benchmark character ABX9802713 in terms of the total number of characters and the character spatial distribution is completely consistent. Only the last character does not match. The structural matching definition formula is used:

[0059]

[0060] The structural matching result is compared with the set threshold After comparison, if the conditions are met, the character structure is determined to be acceptable. Finally, the main peak data of the spectrum is detected. The near-infrared spectrum reflectance image of the packaging surface is received by the visual system, and the central peak 720nm and ±35nm are extracted. It is normalized to [0.655, 0.047, 0.047] and compared with the reference spectrum group for inter-element difference. The maximum difference is 0.016, which is less than the preset spectral difference threshold Δλ max =0.03, so the three types of parameter differences all meet the standards. The three indicators are recorded as the size difference vector [0.10, 0.01, 0.00], the character matching degree 0.909, and the spectral difference vector [0.016, 0.000, 0.000], which are respectively used as the size difference set, the character matching factor and the spectral shift set, and integrated into the multidimensional difference indicator set, where the size difference threshold is set to Δx th =0.15, Δy th =0.05, Δz th =0.05, based on the fact that the maximum molding error ratio of the packaging machine is approximately within 15%, and this is used as the maximum size deviation boundary; character matching threshold The reference barcode misreading rate is controlled within a safety margin of 10% and the upper limit of the spectrum difference value is Δλ. max =0.03 is derived from the maximum allowable deviation of the standard blister material due to changes in angle and incident waveband in the reflection test;

[0061] The identification and confirmation submodule calls the multi-dimensional difference indicator set, sets the difference value threshold standard, and performs numerical threshold judgment on the three indicators respectively. The judgment results are subjected to Boolean cross operation. If all three judgments are in the pass state, the current candidate drug is recorded as the identification pass object and the matching recognition result is obtained;

[0062] After calling the three indicator values in the multi-dimensional difference indicator set, it is necessary to set a unified judgment threshold standard based on the three-dimensional indicator, and the size difference is expressed as Δ max ≤0.15 is the judgment basis, and the character matching degree is C s ≥0.85 is the passing standard, and the maximum element of the spectral difference vector shall not exceed Δλ max =0.03, that is, all indicators should meet the threshold rules of the dimensions to which they belong. The maximum value of the size difference vector 0.10 < 0.15 is qualified, the character matching degree 0.909 > 0.85 is qualified, and the maximum value of the spectrum difference 0.016 < 0.03 is also qualified. The three Boolean judgment values are all 1. Logical multiplication is used in the Boolean cross operation process, that is:

[0063] B=b1·b2·b3=1·1·1=1;

[0064] It is concluded that the candidate drug meets the identification criteria, its ID information is recorded and the identification status field in the operation instruction table is updated to confirm that it is an identified object, and the matching identification result is obtained. The above judgment thresholds are all obtained by setting the tolerance band of the normalized indicator and do not rely on external data inference. The judgment value used in each dimension can be dynamically configured in the initial configuration of the system and is bound to the error sensitivity level. For example, when the accuracy level is set to medium, the size threshold Δx is taken. th =0.15, character threshold Reflection error upper limit Δλ max =0.03, if the system recognition accuracy level is improved, the threshold can be adjusted to Δx th =0.10, Δλ max =0.02, to adapt to scenarios with stricter resolution requirements.

[0065] Table 1 Identification sample data table

[0066]

[0067] As shown in Table 1, all parameter values extracted from sample A001 are within the preset threshold range, and the difference between its values and the baseline set is within the allowable range, making it a valid identification sample. This result demonstrates that under the current system configuration, the executed data judgment logic has controllable parameters and that each threshold setting has a clear basis and reasonable explanation path.

[0068] See also Figure 3 , the grasping and positioning module includes:

[0069] The point cloud extraction submodule obtains matching recognition results and, based on the drug image marked as a candidate target, collects the surface 3D point cloud information of the corresponding area of the image. It divides the area into grids according to the image coordinate system, integrates the coordinate channel and depth channel data of the 3D point set, extracts the point density and normal distribution offset amplitude corresponding to the unit area of the grid, and generates spatial point cloud distribution information.

[0070] After obtaining the image of the drug marked as a candidate target, the system calls the industrial camera to perform structured light illumination on the image to acquire the three-dimensional point cloud information corresponding to its surface. The point cloud file uses (x, y, z) coordinates as the basic unit, and embeds the reflection intensity of each point and the color channel information corresponding to the image pixel in the point structure. Then, the entire image is divided into fixed-dimensional grid areas according to the image coordinate system. For example, each image is divided into 20×20 areas in the horizontal and vertical directions respectively, and each grid unit is 16px×16px. According to this division scheme, the number of all points and the corresponding three-dimensional coordinates are collected grid by grid, and the total number of points is divided by the projected area to obtain the point density per unit area A i , its unit is pt / mm 2 For example, if the number of points in a grid is 160, the corresponding physical area is 2.5mm 2 , then the point density is 64pt / mm 2 , then calculate the angle between the normal vector of all points in the grid and the neighborhood mean normal, and calculate the standard deviation of all angles, which is recorded as the normal distribution offset amplitude D i , if the angle between the normal lines of each point is 3°, 5°, 6°, ..., then the mean square deviation is:

[0071]

[0072] When executing this process, the system generates all the areas (A i ,D i ) numerical combination to form a grid matrix, each position in the matrix represents the structural density and curvature change reference value of different positions on the surface of the candidate drug. Combined with these point cloud distribution information, the system uses it as the basic data set of the adsorption area adaptation factor in subsequent modules to generate spatial point cloud distribution information;

[0073] The surface parameter identification submodule calls the spatial point cloud distribution information, detects the curvature radius change rate, grayscale texture mean and variance, and reflectivity gradient distribution value corresponding to each area of the surface, and judges whether the curvature radius change rate is in the flat state range according to the set reference value interval. It compares the grayscale texture mean and variance with the texture grading interval, and identifies whether the reflectivity gradient distribution value is less than the material adsorption critical value using the formula:

[0074]

[0075] The adsorption adaptation score corresponding to the obtained area is calculated to generate the adsorption area scoring result;

[0076] Among them, K i It represents the normalized value of the curvature radius change rate of the i-th region, which is obtained by dividing the original curvature change by the set maximum change reference value. K represents the normalized value of the reference curvature flatness value, which is the normalized curvature change rate of the target region under the corresponding reference state. T i It represents the normalized value of the grayscale texture comprehensive value of the i-th region, which is normalized to the interval [0,1] after calculation by combining the grayscale mean and standard deviation. T represents the normalized value of the texture level standard value, which is the data after normalization of the set texture reference level. Represents the normalized value of the reflectivity gradient amplitude, which is the absolute value of the reflectivity change per unit distance divided by the set maximum gradient value, D i represents the mean square value of the normal deviation angle in the i-th region, using the dimensionless angle difference calculation, A i Represents the normalized value of the point cloud density per unit area of the i-th region, which is obtained by dividing the point density of the region by the maximum density;

[0077] After calling the spatial point cloud distribution information, the system detects three types of structural attribute parameters region by region: curvature radius change rate, grayscale texture characteristics, reflectivity gradient, curvature change rate K i The numerical estimation is performed by fitting the second-order derivative curve of the surface in each region and normalized to the set maximum reference value K max =0.25, if the local variation is 0.05, then K i =0.20, set the base curvature K = 0.15, grayscale texture eigenvalue T i It is formed by linear weighting of local grayscale mean and standard deviation, such as grayscale mean G μ =120, standard deviation G σ =15, using weighted formula:

[0078]

[0079] Among them G max =255, the texture reference value T is set to 0.35, and the reflectivity gradient By comparing the grayscale change amplitudes of two adjacent rows of pixels and normalizing them to the unit distance gradient value, the maximum reflection gradient is set to If the local change is 10%, the normalized value is 0.333. The above three items are combined with the D obtained from the point cloud information. i =1.91, A i =64 / 100=0.64Substitute into the formula:

[0080]

[0081] The system calculates S for all areas i The records are sorted, and the maximum score represents the optimal adsorption area. The score of this area is the adsorption area scoring result. The curvature reference value K = 0.15 is derived from the curvature of the middle section of the common surface fitting contour of the circular blister package structure. The texture reference value T = 0.35 is taken from the control area corresponding to the texture grayscale distribution of the common smooth surface area of PVC transparent packaging. The upper limit of the reflectivity gradient set by the material adsorption critical value is It comes from the statistical measurement of the region where sudden changes in reflection intensity in the experiment are prone to adsorption slip;

[0082] In the formula,

[0083] S i : represents the adsorption adaptation score of the i-th grid area. The larger the value, the less favorable the surface state of the area is for adsorption, and the smaller the value, the more suitable it is.

[0084] K i : The normalized value of the rate of change of the curvature radius of the ith region, which is obtained by normalizing the rate of change of the second-order derivative of the surface profile in the region, and reflects the degree of geometric fluctuation of the surface of the region;

[0085] K: The normalized value of the base curvature change rate set by the system, usually extracted from a common flat area (such as the middle of a PVC blister), representing the standard curvature under the ideal adsorption surface state;

[0086] T i : The normalized value of the grayscale texture comprehensive value of the i-th region, which is the grayscale mean G μ and standard deviation G σ Linear weighted calculation to evaluate texture complexity;

[0087] T: The base value of the grayscale texture level, which is set as the reference value under the grayscale of the ideal smooth surface texture and is derived from the standard material illumination grayscale distribution;

[0088] The normalized value of the reflectivity gradient at a unit distance represents the degree of grayscale mutation of adjacent pixels and captures the local light variation characteristics caused by surface reflection or material mutation;

[0089] D i : The mean square value of the normal vector deviation angle in the area, which is used to indicate the directional consistency of the surface structure;

[0090] A i : The normalized value of the number of point clouds per unit area in the i-th region, used as an evaluation indicator of structural density.

[0091] The operation structures of the formula are described as follows:

[0092] The molecular part takes the absolute value difference of the three error indicators, which respectively reflect the geometric curvature deviation, texture deviation, and the intensity of optical fluctuation. These three are the main factors affecting the adsorption success rate;

[0093] The three differences were directly summed to form an overall measure of structural error;

[0094] Use square root for the denominator is the stability correction factor of the adsorption region, where Emphasize the importance of normal direction consistency, A i Reflects the density of the local point cloud. The denser the point cloud, the more reliable the structure. The larger the denominator, the higher the overall S i tends to decrease;

[0095] Taking the form of square sum plus one and then taking the root can avoid decimal instability, making the formula discriminative in areas with small structural deviations and avoiding distortion in areas with high disturbances;

[0096] The benefit of the formula is that by uniformly mapping the differences in structure, texture and optical features into weighted absolute deviation accumulation and using structural stability factors for normalization and compression, the surface continuity, image grayscale stability and surface material adsorption risk are taken into account during the adsorption point screening process, thereby supporting the effective evaluation and ranking of special-shaped packaging surfaces.

[0097] The results show that the comprehensive adsorption fitness score of candidate region i is S i =0.192, which is lower than the upper critical value of adsorption risk S set by the system th = 0.25, indicating that the region is in good condition in terms of structural flatness, image stability, and optical adsorption judgment, making it a preferred adsorption point in the current image. This score serves as the core input for the subsequent "Adsorption Coordinate Decision Submodule," which sorts regions from low to high by score. The coordinate point corresponding to the minimum score is the adsorption center. The mathematical logic between the various factor combinations in the formula and the scoring results effectively reveals the combined impact of different parameters on the quality of adsorption point selection and quantifies the direct control variables that can be used for scheduling.

[0098] The adsorption coordinate decision submodule locates the center coordinate point of the adsorption area in the image coordinate system based on the adsorption area scoring results and the area index with the highest score. It then determines the actual coordinate position of the adsorption contact point in three-dimensional space by combining the area center of gravity data, outputs the adsorption center coordinates, and records the corresponding surface type identifier to establish the optimized adsorption coordinate information.

[0099] According to the adsorption area scoring result, the area with the highest score is indexed as i * , the system reads the coordinates of its center point in the image coordinate system Combined with the image-to-world coordinate system mapping calibration parameters, it is transformed into coordinate points (X, Y, Z) in physical 3D space. The center of gravity data of the center point and surrounding points in the region are then extracted from the point cloud data to verify their structural symmetry and normal vector consistency, ultimately determining the location of the adsorption contact point. The system also records the surface material type corresponding to that point, such as blister, paper box, or glass surface structure, indexes the material type, generates a structural label, and outputs the adsorption coordinate point value and corresponding label to jointly establish the optimized adsorption coordinate information. This result is used in the subsequent adsorption execution module as a reference for path and angle matching.

[0100] See also Figure 4 , the crawl execution module includes:

[0101] The adsorption evaluation submodule aligns the terminal adsorption axis with the coordinate position based on the optimized adsorption coordinate information, obtains the standard unit mass value of the drug packaging unit corresponding to the coordinate, extracts the material coding type of the part and the average deviation value in the surface normal direction, and combines the friction characteristic level of the packaging material and the stability level of the packaging structure to use the formula:

[0102]

[0103] Calculate and obtain the adsorption action level value required for adsorption grabbing, match the adsorption vacuum pressure value corresponding to the level, and generate adsorption requirement level information;

[0104] Among them, L s Represents the adsorption action level value, M is the normalized value of the standard unit mass value, θ is the normalized value of the normal direction deviation of the coordinate point, δ is the normalized value of the surface stability level of the area, ζ is the normalized value of the surface texture density of the material, η is the normalized value of the adsorption friction grading coefficient corresponding to the material code, and λ is the normalized value of the effective adsorption area ratio of the material at the adsorption site;

[0105] According to the optimized adsorption coordinate information, the terminal adsorption axis is aligned with the coordinate position, and the standard unit mass value M of the pharmaceutical packaging unit at the corresponding coordinate point is obtained. The material coding type of the coordinate area is recorded, and the normal direction information in the surface point cloud model is collected. The average deviation value θ is calculated in combination with the normal angle of the neighboring points. The friction grading standard library of packaging materials is called to obtain its friction characteristic grade parameter η. The structural stability grade is calculated based on the surface geometric structure boundary morphology and thickness distribution data and normalized to obtain δ. The texture density normalization value ζ is derived from the image grayscale texture frequency data. The effective adsorption area ratio λ is calculated based on the ratio of the number of adsorption contact points per unit area to the regional projection area. After normalizing the above six data, it is inserted into the following formula:

[0106]

[0107] The adsorption action level value is obtained by calculation, and this value is used to match the corresponding vacuum pressure level, which is used as the input signal for controlling the adsorption device and generating adsorption demand level information.

[0108] Here, ln(1+ζ) is used to introduce a nonlinear gain influenced by texture density. Its influence on the overall adsorption level decreases as ζ approaches zero, while its contribution to stable adsorption performance increases as ζ increases. The denominator, the product of the adsorption friction grading coefficient η and the area ratio λ, controls the slip risk level in the adsorption environment, ensuring a stable coordination between the package's bulk characteristics and surface performance in the numerator. The novelty of this formula lies in embedding center of mass stability, surface texture, adsorption area, and friction response into a nonlinear nested structure. This allows for a dynamic characterization of the adsorption stability state without increasing structural complexity, while also taking into account the impact of material differences on the ability to adapt to the adsorption environment.

[0109] If the example data is: M = 0.6, θ = 0.1, δ = 0.15, ζ = 0.3, η = 0.4, λ = 0.5, then:

[0110]

[0111] The result shows that the current adsorption area needs to match the adsorption level of 0.76, which corresponds to the fifth-level vacuum adsorption pressure value in the system. This parameter is required in the subsequent path loading process to complete the adsorption device control.

[0112] The path planning submodule calls the adsorption requirement level information and loads it into the end effector control unit according to the corresponding vacuum pressure parameter level. It reads the three-dimensional coordinate values of the hovering point set in the target distribution area, performs end trajectory planning based on the linear interpolation principle, constructs a three-axis coordinate path sequence, and generates a three-dimensional path coordinate group.

[0113] In the path planning submodule, the adsorption requirement level information obtained by the previous module is called, and the corresponding vacuum adsorption pressure level parameter is written into the control interface of the end effector. The three-dimensional coordinate group of the specified hovering point in the system distribution target area database is read and set as the coordinate of the path end point. On the basis of the path starting point being the current end posture, the linear interpolation path planning principle is used to use the straight line segments between the coordinates as the interpolation basis to divide the path node group, and record the X, Y, and Z coordinate component values and moving step lengths of each path segment respectively to complete the segmented normalization processing of each segment in the path. A sequence label is set for each path node sequence and a posture sequence queue is established. The path generation process does not involve environmental avoidance or multi-target interference, but only considers the translation linear interpolation behavior to generate a three-dimensional path coordinate group, which is subsequently used for loading and execution in the end action control stage to generate action timing signals to coordinate the execution rhythm of the drive module.

[0114] The motion control submodule controls the end effector to move sequentially according to the three-dimensional path coordinate group, gradually executes three-axis interpolation instructions through servo control, tracks the offset distance change between the current robot arm end position and the path node position, determines whether the offset value is lower than the threshold setting, enters the position holding state, and establishes the adsorption movement completion signal;

[0115] In the motion control submodule, the end effector is controlled to execute path movement according to the generated three-dimensional path coordinate group. The servo system reads the path node data in sequence and sends it to the target displacement register of the drive motor. After each node action is completed, the encoder feedback value of the actual position of the current end of the robotic arm is recorded, and the offset distance between the current end position and the set path node is calculated by the three-axis coordinate difference. If the offset distance value is less than the action error threshold ∈ set by the system, for example, ∈ = 0.5mm is set, it is determined that the current node has completed the movement and entered the stable position holding stage. The adsorption action status under this node is recorded and the state completion signal is marked. If all node states meet the error threshold requirements, the state holding result is recorded in the end state machine, and the adsorption movement completion signal is generated. The system enters the waiting process for the next path switching or distribution action scheduling.

[0116] See also Figure 5 , the buffer control module includes:

[0117] The attribute recognition submodule confirms the end-of-line grasping action has ended based on the adsorption movement completion signal. It then calls the three items of the current drug packaging: the bottom contact area, the reference value of the material Young's modulus, and the main peak band value of the material reflectivity. It then extracts the matching material combination index from the material attribute database, determines the drug packaging type based on the material structure group, and obtains the packaging material category information.

[0118] After confirming that the end-grabbing action has ended based on the adsorption movement completion signal, the three data items of the current drug packaging bottom contact area, material Young's modulus reference value and material reflectivity main peak band value are obtained, and the three items are matched item by item with the preset material attribute index table in the material attribute database. Each matching operation is compared in the form of attribute value normalization, such as standardizing the contact area value in square millimeters and normalizing it to the [0,1] interval, and making a difference judgment based on the area reference value set in the database. If the error between the normalized value and the area value of the index item does not exceed 0.05, it is judged as an area matching item. For Young's modulus, the standard value of the material elastic modulus in Pa is collected and normalized. The normalization target interval is set to [0.1,1.0], and the error is less than 5% of the matching items; the matching of the main peak band value of the material reflectivity is based on the wavelength unit nm. After identifying the center position and the left and right half widths of the main peak of the reflection curve, the main peak center band value is calculated. The matching difference is less than 10nm for a valid match. The above three matching index items must meet the conditions at the same time and then be combined and classified to form a material combination index group. According to the combination index, the corresponding material structure group number is extracted from the structure group dictionary. The structure group number is set from 1 to 9 according to the material constitutive model type and the hierarchical structure complexity number. Combined with the classification rules set in the database, it is further determined whether its material group belongs to the corresponding range of the pharmaceutical packaging structure. If it meets the requirements, the pharmaceutical packaging type label code is established according to the mapping rule. The code is used for the subsequent setting of the path constraint parameters to finally obtain the packaging material category information.

[0119] The path setting submodule determines whether the packaging material is classified as a brittle material based on the packaging material category information and identifies whether the contact area of the packaging bottom surface is less than the lower area limit. If the judgment result is true, the starting and ending point coordinates of the deceleration section in the terminal descent path are set. The length of the deceleration section is calculated by combining the current height of the terminal and the height of the landing point reference surface to generate the deceleration path length information;

[0120] Based on the packaging material category information obtained in the previous section, identify whether the current packaging belongs to the brittle material category. The classification rule is based on the residual deformation of the packaging material under the impact of the drop point. The brittle material judgment condition is that the residual deformation under unit impact force is less than 0.2mm. When the material category is glass, ceramic or thin plastic composite material, it is defaulted to brittle type. If the judgment is established, enter the buffer path construction process; continue to check whether the contact area of the bottom surface of the package is less than the set area lower limit, which is X min =150mm 2If the actual contact area is smaller than this value, it is necessary to set a deceleration section in the descent path. The deceleration section is defined as the interval in the descent path where the Z-axis servo speed of the end of the robot arm is reduced. The starting point coordinate is set to the Z-axis coordinate of the current position, and the end point is set to the coordinate point after the reference coordinate of the landing point is offset 20mm upward. The actual total height of the fall is obtained by the difference between the Z value of the current position and the Z value of the landing reference surface, which is recorded as H. z The length of the deceleration section is the difference between the difference and the height of the end point, recorded as U, that is: U = H z -20mm. After completing the value calculation, the deceleration path length information is generated.

[0121] The speed adjustment submodule calls the deceleration path length information, combines the Z-axis displacement value from the current manipulator end to the target landing point, collects the normalized value of the material's Young's modulus, the normalized value of the contact area, and the peak symmetry factor of the package reflection band, sets the exponential buffer adjustment coefficient, and constructs a dynamic control factor using the formula:

[0122]

[0123] Obtain the terminal Z-axis servo speed classification reference value through calculation, set the speed sequence for each interval of the descending section based on the classification standard, load the sequence to the controller output end, perform displacement action, and establish a buffer control sequence;

[0124] Among them, V z It represents the servo speed classification benchmark value in the Z-axis direction, E is the normalized value of Young's modulus, X is the normalized value of contact area, U is the deceleration path length value, ψ is the ratio of the left and right half-widths of the main reflection peak of the material, ρ is the normalized value of the package surface roughness, κ is the buffer response attenuation exponent parameter, and e is the natural logarithm base constant.

[0125] Call the deceleration path length information and collect the Z-axis displacement value from the current robot end to the landing point reference surface, recorded as H z

[0126] =58mm, and at the same time obtain the normalized Young's modulus value E = 0.45, the normalized contact area value X = 0.36, the left and right half-width ratio of the package reflection band ψ = 1.3, and the normalized roughness value ρ = 0.4 of the package surface. Set the buffer response index parameter κ = 2.5 and substitute the above parameters into the following formula:

[0127]

[0128] Substituting the data into the equation:

[0129]

[0130] The final Z-axis servo speed classification benchmark value is 0.0647. Based on this value, four speed levels are divided within the deceleration path segment, which are set to 0.9, 0.7, 0.5, and 0.3 times the benchmark value, namely 0.0582, 0.0453, 0.0324, and 0.0194 respectively. According to the distance interval of the Z-axis falling path segment, the speed value of each segment is allocated and recorded in the path speed instruction table, and finally the buffer control sequence is established.

[0131] In this formula, the product of E and X serves as the upper factor, representing the adsorption buffer foundation formed by the material's elastic response and contact area. The denominator is composed of three components: path length U, material reflectivity asymmetry |ψ-1|, and material roughness influence factor exp(-κ·ρ), reflecting the need for adjustment of the falling path, the influence of surface structure on adsorption stability, and the degree of dynamic response suppression, respectively. The entire formula, through square root extraction, improves convergence to velocity amplitude, enhancing control smoothness and feedback elasticity.

[0132] The benefit of the formula is that by introducing the material's reflection band symmetry and surface roughness response index, the influence of surface microscopic properties on the buffering process is quantified and embedded into the speed calculation logic, effectively avoiding the simplified method of traditional path buffering that only relies on contact area and modulus.

[0133] This result shows that the calculated V z =0.0647 as the graded speed value reflects the descent speed control benchmark that the end effector should execute under the current path structure and material properties. This result will directly affect the speed sequence setting, thereby affecting the execution effect of the buffering action and the maintenance of packaging stability.

[0134] See also Figure 6 , the distribution and transfer module includes:

[0135] The release control submodule extracts the recorded release position coordinate information based on the buffer control sequence, binds the tension release moment of the end suction device to the release coordinate point, detects the spatial overlap between the current robot end position and the coordinate point, and determines whether the coordinate match reaches the release judgment threshold. If the set conditions are met, the end effector is triggered to execute the tension release command and establish the release action trigger information;

[0136] According to the multiple servo speed intervals and target landing point coordinates recorded in the buffer control sequence, the release position coordinate value at the end of the release path is extracted and set as the midpoint coordinate (x r ,y r ,z r ), and establish a binding relationship between the end suction device tension release action and the release point coordinates. The system sets the binding behavior as the reference benchmark for triggering the release action; combined with the current position coordinates (x c,y c ,z c ), collect the Euclidean distance between the end coordinates of the robot arm and the reference release point coordinates in real time, use the spatial coincidence threshold judgment rule, define the release judgment threshold ∈=3mm, when the formula is satisfied:

[0137]

[0138] That is, the current position is considered to have reached the trigger release condition. In actual tasks, for example, when the release point coordinates are (150, 210, 320) mm and the current acquisition value is (152.1, 211.4, 318.7) mm, the calculated distance is:

[0139]

[0140] If the pressure is less than the set threshold of 3mm, the release condition is met and the system immediately triggers the tension release command. The controller changes the state of the end adsorber from negative pressure adsorption to tension release state. At the moment of release, the solenoid valve is reversed to release the pressure difference inside the negative pressure cavity. After the operation is completed, the release action trigger information is established.

[0141] The drop point verification submodule collects the drop point coordinate feedback signal based on the release action trigger information, obtains the grid number of the current coordinate, calls the defined distribution unit grid number sequence, and compares the two sets of numbers. If the drop point number is consistent with the target grid number, the current delivery behavior is marked as a positioning compliance state and positioning consistency identification information is generated;

[0142] According to the release action trigger information, the coordinate feedback value of the landing point of the drug touching the ground after release is collected in real time. d ,y d ,z d ) The position signal is fed back to the control center by the lower computer. The system maps the landing point coordinates to the distribution grid number space according to the spatial grid mapping rule, and records it as the current grid number G d , the target grid number sequence G is preset in the distribution task scheduling system t The numbering sequence is constructed based on the distribution unit positioning matrix in the task distribution table; for example, the landing point coordinates are (405.3, 279.2, 12.6) mm, and the corresponding numbering rule is that each unit side is 40 mm long, then G d =(10,7), and the preset target grid number of the current task is G t=(10,7), compare the two sets of numbers to determine whether the numbers of each dimension are completely consistent. If they are consistent, mark this delivery behavior as positioning compliance; if the numbers are inconsistent, no positioning identification information is generated and the current result is invalidated. After the number comparison result is output, the system generates positioning consistency identification information when the number consistency condition is met.

[0143] The path switching submodule extracts the target shelf number and slot location code bound to the task assignment record based on the positioning consistency identification information judgment result, combines them into a path scheduling address code, and controls the transfer rail to execute the path switching instruction according to the address code index instruction sequence to establish the distribution completion identification code;

[0144] Based on the positioning consistency identification information obtained above, the system determines that the positioning result is compliant, and then extracts the target shelf number and slot positioning code assigned to the current drug in the task scheduling order to construct a path scheduling address code. The system sets the shelf number to a four-digit integer code, and the slot positioning adopts a three-digit letter + number mixed code. For example, if the shelf number is 0312 and the slot code is G7B, the combined address code is 0312G7B. The controller calls the address code to the path instruction library to find the corresponding transfer rail track switching instruction sequence. The track switching instruction is mapped into a string of continuous codes according to the rail control logic, for example: 030→070→135→end point. The track point is dynamically set as a path node according to the address code; the control system controls the track motor to perform the corresponding drive action, and pushes the trolley to the target position in the order of the path nodes. After the process is completed, the system generates and records the distribution completion identification code of the current task. The identification code is composed of the task number + shelf number + timestamp, such as:

[0145] T2025061902_0312_20250620104531.

[0146] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A drug intelligent recognition and grasping system based on a robot arm, characterized in that: The system comprises: The recognition and verification module receives the instruction issued by the refill task, calls three pieces of information about the candidate drug in the robot's field of view: the length, width, and height values, the total number of barcode characters and their spatial distribution, and the near-infrared reflectivity curve value. It determines whether the difference values of the three items are all below the set matching threshold and generates a matching recognition result. Based on the matching and recognition results, the grasping and positioning module obtains the curvature radius change rate, grayscale texture mean and variance, and reflectivity gradient distribution value of each area on the packaging surface, calculates the adsorption effect score, selects the coordinates of the area with the highest score as the adsorption grasping contact point, and generates optimized adsorption coordinate information; The grabbing execution module obtains the standard unit mass value of the drug packaging unit corresponding to the position based on the optimized adsorption coordinate information, selects the corresponding pressure value level, performs an adsorption grabbing action, moves the drug to the specified hovering position above, and generates an adsorption movement completion signal; The buffer control module determines the drug packaging category based on the adsorption movement completion signal, combines the Z-axis displacement value from the current robot end to the target landing point, sets the Z-axis servo speed sequence according to the gradient, loads the deceleration instruction, and executes the slow drop to generate a buffer control sequence.

2. The drug intelligent recognition and grasping system based on a robot according to claim 1 is characterized in that: The matching recognition result includes the drug size offset value, the character structure consistency coefficient and the material spectrum difference ratio; the optimized adsorption coordinate information includes the target area spatial coordinates, the area surface roughness level and the surface reflection distribution range; the adsorption movement completion signal is specifically the adsorption action state identifier, the adsorption point pressure value record and the end effector feedback status code; the buffer control sequence includes the deceleration interval displacement length, the end speed change gradient and the release instantaneous tension level.

3. The drug intelligent recognition and grasping system based on a robot according to claim 2 is characterized in that: The identification and verification module includes: The benchmark extraction submodule receives the instruction issued by the replenishment task and extracts three sets of parameters from the drug demand record: the standard value of the outer packaging size, the barcode encoding character sequence, and the standard value of the packaging material reflectance spectrum. These parameters are stored as a size benchmark group, a character benchmark group, and a spectrum benchmark group, respectively. The benchmark value set is generated based on the contents of the three sets of parameter records. The difference determination submodule, based on the reference value set, calls the three contents of the length, width and height dimensions of the candidate drug, the total number of barcode characters and the character spatial distribution, and the near-infrared reflectivity curve value, calculates the individual difference values between the length, width and height dimensions and the dimension reference group, calculates the structural correspondence between the character spatial distribution and the character reference group according to the position mapping rule, calculates the bandwidth difference range between the peak position of the spectral curve and the spectral reference group according to the main peak band comparison formula, and generates a multi-dimensional difference index set; The identification and confirmation submodule calls the multidimensional difference indicator set, sets the difference value threshold standard and performs numerical threshold judgment on the three indicators respectively, performs Boolean cross operation on the judgment results, and if all three judgments are in the pass state, the current candidate drug is recorded as the identification pass object, and the matching recognition result is obtained.

4. The drug intelligent recognition and grasping system based on a robot according to claim 3 is characterized in that: The grabbing and positioning module includes: The point cloud extraction submodule obtains the matching recognition results and, based on the drug image marked as a candidate target, collects the surface three-dimensional point cloud information of the corresponding area of the image, divides the area into a grid according to the image coordinate system, integrates the coordinate channel and depth channel data in the three-dimensional point set, extracts the point density and normal distribution offset amplitude corresponding to the unit area of the grid, and generates spatial point cloud distribution information; The surface parameter identification submodule calls the spatial point cloud distribution information, detects the curvature radius change rate, grayscale texture mean and variance, and reflectivity gradient distribution value corresponding to the surface of each area, determines whether the curvature radius change rate is in a flat state range based on the set reference value range, compares the grayscale texture mean and variance with the texture grading range, identifies whether the reflectivity gradient distribution value is less than the material adsorption critical value, calculates and obtains the adsorption adaptation score corresponding to the area, and generates an adsorption area scoring result; The adsorption coordinate decision submodule is based on the adsorption area scoring results and the area index with the highest score, and locates the center coordinate point of the area in the image coordinate system. It combines the area center of gravity data to determine the actual coordinate position of the adsorption contact point in three-dimensional space, outputs the adsorption center coordinates and records the corresponding surface type identification, and establishes optimized adsorption coordinate information.

5. The drug intelligent recognition and grasping system based on a robot according to claim 4 is characterized in that: The capture execution module includes: The adsorption evaluation submodule aligns the terminal adsorption axis with the coordinate position based on the optimized adsorption coordinate information, obtains the standard unit mass value of the drug packaging unit corresponding to the coordinate, extracts the material coding type of the part and the average deviation value in the surface normal direction, combines the friction characteristic level of the packaging material and the stability level of the packaging structure, calculates the adsorption action level value required for adsorption grasping, matches the adsorption vacuum pressure value corresponding to the level, and generates adsorption requirement level information; The path planning submodule calls the adsorption requirement level information and loads it into the end effector control unit according to the corresponding vacuum pressure parameter level. It reads the three-dimensional coordinate values of the hovering point set in the target distribution area, performs end trajectory planning based on the linear interpolation principle, constructs a three-axis coordinate path sequence, and generates a three-dimensional path coordinate group. The motion control submodule controls the end effector to move sequentially according to the three-dimensional path coordinate group, gradually performs three-axis interpolation instructions through servo control, tracks the change in the offset distance between the current robot arm end position and the path node position, determines whether the offset value is lower than the threshold setting, enters the position holding state, and establishes an adsorption movement completion signal.

6. The drug intelligent recognition and grasping system based on a robot according to claim 5 is characterized in that: The buffer control module includes: The attribute recognition submodule confirms the completion of the end-grabbing action based on the adsorption movement completion signal, calls the three contents of the current drug packaging bottom surface contact area, the material Young's modulus reference value, and the material reflectivity main peak band value, extracts the matching material combination index from the material attribute database, determines the drug packaging type based on the material structure group, and obtains the packaging material category information; The path setting submodule determines whether the packaging material is classified as a brittle material based on the packaging material category information, identifies whether the contact area of the packaging bottom surface is less than the lower area limit, and if so, sets the starting and ending point coordinates of the deceleration section in the terminal descent path. It calculates the length of the deceleration section by combining the current height of the terminal and the height of the landing point reference surface to generate deceleration path length information; The speed adjustment submodule calls the deceleration path length information, combines the Z-axis displacement value from the current manipulator end to the target landing point, collects the material's Young's modulus, contact area and peak symmetry factor of the packaging reflection band, sets the exponential buffer adjustment coefficient and constructs a dynamic control factor, calculates and obtains the terminal Z-axis servo speed grading reference value, sets the speed sequence of each interval of the descent section based on the grading standard, loads the sequence to the controller output end, performs the displacement action, and establishes a buffer control sequence.

7. The drug intelligent recognition and grasping system based on a robot according to claim 6 is characterized in that: The system further comprises: The distribution and transfer module extracts the recorded release position coordinates according to the buffer control sequence, obtains the landing point status feedback signal at the coordinates after the medicine is dropped, and determines whether it is consistent with the coordinate grid number of the distribution unit. If it is consistent, it calls the assigned target shelf number and grid positioning code, controls the transfer rail to execute the corresponding path switching instruction, and generates a distribution completion identification code; The distribution completion identification code specifically refers to the distribution shelf number, the grid docking position number and the track switching path index.

8. The drug intelligent recognition and grasping system based on a robot according to claim 7 is characterized in that: The distribution and transfer module includes: The release control submodule extracts the recorded release position coordinate information according to the buffer control sequence, binds the tension release moment of the end suction device to the release coordinate point, detects the spatial coincidence between the current end position of the manipulator and the coordinate point, and determines whether the coordinate match reaches the release judgment threshold. If the set conditions are met, the end effector is triggered to execute the tension release command and establish the release action trigger information; The landing point verification submodule collects the landing point coordinate feedback signal after the drug is dropped according to the release action trigger information, obtains the grid number of the current coordinate, calls the defined distribution unit grid number sequence, and compares the two sets of numbers. If the landing point number is consistent with the target grid number, the current delivery behavior is marked as a positioning compliance state and positioning consistency identification information is generated; The path switching submodule extracts the target shelf number and slot positioning code bound in the task allocation record based on the positioning consistency identification information judgment result, combines them into a path scheduling address code, controls the transfer rail to execute the path switching instruction according to the address code index instruction sequence, and establishes a distribution completion identification code.

Images