Grabbing control method and device based on pre-recognition and storage medium

Abstract

The invention belongs to the technical field of machine learning and robot grabbing, and provides a grabbing control method based on pre-recognition. Before grabbing operation is executed, sensing data obtained when a clamping jaw does not make contact with an object is collected based on a capacitive sensor and input into a pre-trained machine learning model, the category of the object to be grabbed is recognized, and grabbing target force is obtained through the recognized object category according to the mapping relation between the object category and grabbing force; and in the grabbing stage, based on the contact force collected by the capacitive sensor, optimization control is conducted on the opening amount of the clamping jaw, and the clamping jaw clamps the object with the grabbing target force. According to the invention, before a clamping behavior occurs, additional tactile information and supplementary information modes, different from visual information of a camera, of a clamped object can be provided, and the force application range of the clamping jaw can be expanded through the sensor, so that a clamping task can be better executed.

Description

Technical Field

[0001] The present invention belongs to the field of machine learning and robot grasping technology, and specifically relates to a grasping control method, device and storage medium based on pre-identification. Background Art

[0002] With the rapid advancement of modern technology, robotics is also developing rapidly. The need for human-machine interaction and environmental interaction in robotics has stimulated the development of sensor technology and sensor data processing. Existing sensors generally include lidar, cameras, ultrasonic radar, and contact force sensors. Camera-derived visual image processing technology, along with its analogous point cloud information processing, has gradually matured and improved with the development of deep network technology, resulting in a large number of application cases. When a single sensor performs environmental perception, the data collected inevitably contains errors. Furthermore, in complex environments, a single sensor may be subject to interference or even fail. Complex task requirements also render the information provided by a single sensor inadequate. Therefore, the integration of multiple sensors and the introduction of additional sensors are gaining increasing attention.

[0003] Inspired by the five senses of the human body, namely vision, touch, hearing, smell, and taste, people have begun to explore the application of other senses in the field of robotics, now that visual information processing is relatively complete. Among them, touch and hearing have received more attention due to their wider range of application scenarios. In the field of tactile perception, traditional tactile perception mainly focuses on contact tactile perception. According to the classification of common tactile sensors, it can be divided into capacitive, piezoelectric, piezoresistive, and visual tactile. Among them, due to the principle of pressure excitation, this type of tactile sensor is limited to contact tactile. However, based on the principle of pressure changing the electrode plate, capacitive sensors can also use the medium between the plates for detection, which makes non-contact short-distance tactile detection possible. At present, this type of short-distance detection has performed well in medical human body detection, pipeline liquid level detection and other non-destructive and close-fitting detection tasks of internal structures, and has produced many application examples.

[0004] Applying capacitive tactile sensors to contactless gripping pre-identification can provide effective, predictive tactile information for gripping tasks. Currently, since the gripper cannot fully integrate with the camera's vision, cameras deployed farther away often affect gripping due to occlusion. Cameras deployed on the gripper, due to their angle and position, are unable to function when the gripper is close to the object to be gripped. Currently, research has been conducted on the use of dual robotic arms to provide each other with scene visual information. However, being able to deploy additional sensors directly on a single robotic arm to supplement the missing visual information with tactile information for decision-making is also of great significance to robot scene perception.

[0005] Current research on tactile grasping is mainly based on contact tactile sense. In a similar study (the invention patent application is entitled "Multi-finger visual tactile track gripper and its control method", and the patent application publication number is CN117621133A), a multi-finger visual tactile track gripper is used to effectively combine the sensor structure and functional structure while providing tactile information. This not only solves the problem of the gripper's stable gripping of the object to be gripped, but also automatically adjusts the posture of the gripper according to the shape of the object. However, this type of gripper design lacks effective protection for the object. When the object is relatively fragile and the force range of the gripper is not within the force range that the gripper can apply, it is easy to cause damage to the object. The present invention uses an identification-based method to determine the target force of the object and supports the expansion of the original force range of the gripper through sensors. It can grip the object with a safe force value without being limited by the force range of the gripper itself. In addition, the present invention also expands the recognition function, which can expand the cognition of the gripper object while gripping, and can provide assistance in various intelligent applications.

[0006] Currently, designs exist that protect the function of gripping objects. For example, in a related research study (the invention patent application is titled "Multi-Mode Hybrid Grasping Manipulator Based on Tactile Perception," with patent publication number CN119017423A), a multi-mode hybrid grasping manipulator with tactile perception was developed to supplement tactile information during grasping and avoid damage to the gripped object. This manipulator comprises a gripping drive mechanism, an enveloping drive mechanism, a negative pressure mechanism, and a tactile sensing mechanism, supporting gripping using three modes: gripping, enveloping, and adsorption. A representative example of this type of device currently has the following features: modified grippers; the use of contact tactile sensors; and both targets are used for grasping tasks. A potential disadvantage of this device is that the expanded grasping method increases the size of the gripping device, making it more susceptible to obstacles. The tactile sensor used in this device is a traditional contact tactile sensor. To address these drawbacks, the present invention integrates the sensor with the gripper itself, significantly reducing the overall size of the gripper and enabling efficient operation even in environments with numerous obstacles. In addition, the non-contact pre-identification function and force-controlled gripping algorithm both rely on gripper fingertip replacement components with capacitive sensor functions, which can be easily transplanted to various existing grippers without the need to redesign the gripper.

[0007] Grippers with enhanced sensing capabilities and the ability to achieve more versatile gripping methods are playing a crucial role in the advancement of intelligent robots. Effectively grasping objects and acquiring information about them will be fundamental yet crucial tasks for robots to perform complex tasks.

[0008] The above information disclosed in this Background section is only for enhancement of understanding of the background of the invention and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0009] The present invention aims to solve one of the technical problems in the related art at least to a certain extent.

[0010] To this end, the present invention proposes a grasping control method, device and storage medium based on pre-identification, which combines non-visual auxiliary information before grasping an object, so that the robot has richer perceptual information about the grasped object and ensures the smooth progress of the grasping task.

[0011] In order to achieve the above object, the present invention adopts the following technical solutions:

[0012] A first aspect of the present invention provides a grasping control method based on pre-identification, wherein a capacitive sensor is provided on the gripping arm of the gripper facing the side of the object to be grasped, and the grasping control method comprises:

[0013] Step S1: Before performing a grasping operation, the capacitive sensor collects sensor data when the gripper is not in contact with the object, inputs the data into a pre-trained machine learning model, identifies the category of the object to be grasped, and derives the grasping target force from the identified object category based on the mapping relationship between the object category and the grasping force;

[0014] Step S2, the grasping stage, optimizes and controls the opening amount of the gripper based on the contact force collected by the capacitive sensor, so that the gripper grasps the object with the grasping target force.

[0015] In some embodiments, step S1 specifically includes:

[0016] Step S11: Before performing the grasping operation, the gripper is in an idle and stationary state, and a round of capacitance data is collected as reference capacitance data to be identified. Then, the gripper is moved near the object to be grasped, and a round of capacitance data is collected in a stationary state as static capacitance data to be identified.

[0017] Step S12: performing data enhancement on the reference capacitance data to be identified and the static capacitance data to be identified to obtain enhanced static capacitance data to be identified;

[0018] Step S13: Input the enhanced static capacitance data to be identified into the pre-trained first machine learning model to perform non-contact object classification pre-identification to obtain the object category L p ;

[0019] Step S14: According to the mapping relationship between object category and grasping force, the object category L p Get the grasping target force Fp .

[0020] In some embodiments, the data enhancement includes: filtering the reference capacitance data to be identified and the static capacitance data to be identified, then obtaining the mean of the data, and subtracting the mean of the reference capacitance data to be identified from the mean of the static capacitance data to be identified to obtain the enhanced static capacitance data to be identified.

[0021] In some embodiments, the first machine learning model is trained using a static classification dataset, where the static classification dataset is obtained according to the following steps:

[0022] Step S1311, data acquisition basic parameter settings: set the number e of the current round of data, initialize e=1; set the number of repeated acquisitions of the current round of data to K c ; Set the collection interval between two adjacent data to t c ; Set the cutoff frequency f of the frequency domain filter c ; Set the jaw opening to the maximum value and keep the jaws in an empty state;

[0023] Step S1312: for the object S collected in the current round e , static data collection is carried out in two stages:

[0024] The first stage, reference signal acquisition: make the object S e Located outside the detection area of ​​all capacitive sensors on the gripper, based on the acquisition interval t c and the number of repeated acquisitions K c Perform a single round of data acquisition to obtain the reference signal C of the capacitive sensor i bi ;

[0025] The second stage, static signal acquisition: move the gripper to the object S e Nearby, make the object S e Located within the detection area of ​​all capacitive sensors on the gripper, based on the acquisition interval t c and the number of repeated acquisitions K c Perform a single round of data acquisition to obtain the static signal C of the capacitive sensor i si ;

[0026] Step S1313: Based on the cutoff frequency f c For reference signal C bi and static signal C si Filter them separately to get the reference mutual capacitance mean μ of capacitance sensor i bi and the static mutual capacitance μ si , and make the difference between the two to get the enhanced static capacitance data μ of capacitance sensor i sdi :

[0027] μ sdi = μ bi -μ si

[0028] Based on the object S to be recognized corresponding to the data number e e , the static classification data D is obtained s : [μ sdi , L e , where L e is the class label of the object to be recognized corresponding to the data number e of this data;

[0029] Step S1314: If e < E, then set e = e + 1, and repeat Step S1312 to Step S1313 until e = E, and end the acquisition of the static classification data D s All the obtained static classification data are used as the static classification data set, and E is the number of object categories contained in the static classification data set.

[0030] In some embodiments, Step S1 specifically includes:

[0031] Step S11': Before performing the grasping operation, when the gripper is in the vacant and stationary state, collect a round of capacitance data as the capacitance data of the benchmark to be recognized; then move the gripper to near the object to be recognized, and control the gripper to gradually close until it just fits the object to be recognized, and collect the capacitance data of the dynamic object to be recognized during this process;

[0032] Step S12': Perform data enhancement on the capacitance data of the benchmark to be recognized and the capacitance data of the dynamic object to be recognized, and obtain the enhanced capacitance data of the dynamic object to be recognized;

[0033] Step S13': Input the enhanced capacitance data of the dynamic object to be recognized into the pre-trained second machine learning model for non-contact object classification pre-identification, and obtain the object category L p ;

[0034] Step S14': According to the mapping relationship between the object category and the grasping force, obtain the grasping target force F from the object category L p p .

[0035] In some embodiments, the data enhancement includes: filtering the capacitance data of the benchmark to be recognized, and then calculating the mean value of the data; calculating the mean value and variance of the capacitance data of the dynamic object to be recognized, performing normalization, and flipping the overall sequence forward and backward to perform polynomial fitting to obtain the fitting parameters;

[0036] ​The enhanced dynamic capacitance data to be identified is constructed using the fitting parameters, a mean difference between the reference capacitance data to be identified and the dynamic capacitance data to be identified, and a variance of the dynamic capacitance data to be identified.

[0037] In some embodiments, the second machine learning model is trained using a dynamic classification dataset, where the dynamic classification dataset is obtained according to the following steps:

[0038] Step S1311', data acquisition basic parameter settings: set the number e of the current round of data, initialize e=1; set the number of repeated acquisitions of the current round of data to K c ; Set the collection interval between two adjacent data to t c ; Set the end force value of the current round data to F e ; Set the step length l of the gripper closing s ; Set the cutoff frequency f of the frequency domain filter c ;Data fixed interval d;

[0039] Step S1312': for the object S collected in the current round e , dynamic data collection is carried out in two stages:

[0040] The first stage, the reference signal acquisition: keep the gripper in the empty state with the maximum opening amount, and make the object S e Located outside the detection area of ​​all capacitive sensors on the gripper, based on the acquisition interval t c and the number of repeated acquisitions K c Perform a single round of data acquisition to obtain the reference signal C of the capacitive sensor i bi ;

[0041] The second stage, sequence signal acquisition: move the gripper to the object S e Nearby, make the object S e Located in the detection area of ​​all capacitive sensors on the gripper, the gripper starts from the maximum opening and follows the set step length l s Close step by step, collect data of the closing process after each step, and wait for a collection interval t c Then, the data of the next closing process is collected; after each step of the jaw closing, if the maximum value of the normal force data collected by each capacitive sensor is higher than the end force value F e , or the gripper is completely closed, the collected closing process data of the capacitance sensor i is combined into a sequence signal C of the capacitance sensor i di ;

[0042] Step S1313', based on the cutoff frequency f c Reference signal C for capacitance sensor i bi Filter and obtain the reference mutual capacitance mean μ of capacitance sensor ibi ; Represent the sequence signal C of the capacitive sensor i based on the data fixed interval d di in the time domain, and perform data fitting after flipping the head and tail to obtain the fitting parameter p of the capacitive sensor i i , and calculate the mutual capacitance mean value μ of the sequence signal of the capacitive sensor i bi and the standard deviation σ di , and then obtain the difference μ ddi :

[0043] μ ddi =μ bi -μ di

[0044] Based on the object S corresponding to the data number e e , obtain the dynamic classification data D d : [p i , μ ddi , σ di , L e , where L e is the category label of the object to be recognized corresponding to the data number e of this data;

[0045] Step S1314': If e < E, then set e = e + 1, and repeat steps S1312' to S1313' until e = E, and end the dynamic classification data D s acquisition, and use all the obtained dynamic classification data as the dynamic classification data set, and E is the number of object categories contained in the dynamic classification data set.

[0046] In some embodiments, step S2 specifically includes:

[0047] Step S21: Set the proximity force value threshold F l1 , read the data F of the capacitive sensor c , and count the maximum value F of the force value data of all capacitive sensors on the current gripper arms max ;

[0048] Step S22: Based on the maximum value F of the force value data stored in the current capacitive sensors max , judge the current gripping state: If F max ≤F l1 , then the gripper is in the state of not being close to the object to be grasped, and execute step S23; If F max >F l1 , then the gripper is in the state of contacting the object to be grasped, and execute step S24;

[0049] Step S23: The gripper gradually closes based on the step length l s , and the interval time between two adjacent steps of the gripper closing is set to tc Then, according to step S21, the maximum value F of the force data of each capacitance sensor is updated. max , and judging the current clamping state based on step S22;

[0050] Step S24: Use the optimization method to close the gripper to the target gripping force F. p The closest position, the steps are as follows:

[0051] Step S241: Set the optimization step length l j ; Set the optimized attenuation coefficient β and initialize it to β0=1;

[0052] Step S242: Construct a step size optimization function J(ΔF) to represent the change Δl of the clamping jaw closing position each time to control the clamping jaw closing. The expression is as follows:

[0053]

[0054] J(ΔF)=β n ×m(ΔF)×l j

[0055] Among them, l min Indicates the minimum closing amount that the gripper can perform; the force difference ΔF = F p -F max ;

[0056] m(ΔF) is the sign of the optimization direction, which is converted from the force difference ΔF:

[0057]

[0058] β n To optimize the iterative value of the attenuation coefficient β:

[0059]

[0060] Where n is the number of times the force difference ΔF changes its sign, initially n=0, N m is the set number of times to start decay, β min is the lower limit of the attenuation coefficient;

[0061] Step S243: Get the jaw opening amount feedback or obtain the current jaw opening amount x based on the accumulated step length. c , and the current opening amount of the gripper x c Save to data pool X l In the data pool X l The latest N l The jaw opening data;

[0062] Step S244: The gripper moves under the control of the position change Δl. When the saved gripper opening amount data has satisfied N l When requested, a stop detection flag is performed stop :

[0063]

[0064] Where x is the data pool X l Data in

[0065] Step S245: If flag stop = False, then repeat steps S242 to S244 to continue the force-controlled gripping process; if flag stop = True, it indicates that the grasping target force F has been reached p The corresponding nearest position is clamped.

[0066] A second aspect of the present invention provides a pre-identification based grasping control device, comprising:

[0067] a capacitive sensor mounted on the gripping arm of the gripper on the side facing the object to be gripped;

[0068] a pre-identification module configured to, before performing a grasping operation, collect sensor data from the gripper when it is not in contact with an object based on the capacitive sensor, input the data into a pre-trained machine learning model, identify the category of the object to be grasped, and derive a grasping target force from the identified object category based on a mapping relationship between the object category and the grasping force;

[0069] The optimization control module is configured to optimize the opening amount of the clamping jaws based on the contact force collected by the capacitive sensor, so that the clamping jaws can clamp the object with the grasping target force.

[0070] A third aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the grasping control method according to any embodiment of the first aspect of the present invention.

[0071] The characteristics and beneficial effects of the present invention are:

[0072] The feature of the present invention is that it can use capacitive sensors to provide the required additional information when the robot performs gripping. In addition, the present invention supports providing pre-identification information in a non-contact manner by maintaining a certain distance before the gripper formally applies force to the gripped object, or providing pre-identification information based on the sequence sensing information of the approaching process during the gripping process, so that the robot has richer perception information about the gripped object, performs preliminary classification judgment of the gripped object, or pre-estimates the gripping force value to avoid excessive gripping force damaging the gripped object. The present invention can also use sensors to expand the force application range of the gripper itself, expand the force detection range of the sensor to the gripper's gripping force range, and accurately complete the force control gripping task by relying on the force measurement accuracy of the sensor.

[0073] The present invention has the following beneficial effects:

[0074] 1. The identification information provided by the present invention belongs to the tactile information of the robot, which is an effective supplement to the robot's visual information. It can also provide independent tactile information judgment when the visual information cannot fully play a role.

[0075] 2. The tactile information provided by the present invention can be obtained without completely clamping the object. It has two sets of recognition frameworks: static non-contact recognition and recognition during approach. It can obtain additional information about the object being recognized without damaging or interfering with the object being recognized.

[0076] 3. The present invention can classify and identify the object to be identified, or estimate the gripping force value based on the recognition information, and can provide effective assistance to the robot grasping task.

[0077] 4. The present invention can use a force sensor to expand the force application range of the clamp itself, which can break through the range limitation of the clamp's force application. Through algorithm control and sensor force information, the sensor's force detection range is expanded to the clamping force range of the clamp, realizing force-controlled clamping tasks based on the sensor's force measurement accuracy, thereby increasing the versatility of the equipment.

[0078] 5. The present invention uses a gripper fingertip component combined with a capacitive sensor, which can be easily transplanted onto various grippers without the need to redesign the gripper. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] Figure 1 An overall flow chart of a grasping control method based on pre-identification provided in an embodiment of the first aspect of the present invention;

[0080] Figure 2 A schematic structural diagram of an electronic device provided in accordance with an embodiment of the third aspect of the present invention. DETAILED DESCRIPTION

[0081] In order to make the purpose, technical solutions and advantages of this application more clearly understood, this application is 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 this application and are not intended to limit this application.

[0082] On the contrary, this application covers any alternatives, modifications, equivalents, and solutions made within the spirit and scope of this application as defined by the claims. Furthermore, to facilitate a better understanding of this application, certain specific details are described in detail below in the detailed description of this application. Those skilled in the art will be able to fully understand this application without these details.

[0083] According to a first aspect of the present invention, an embodiment of a grasping control method based on pre-identification is proposed. A capacitive sensor is provided on each gripping arm of the gripper facing the side of the object to be grasped, and the primary sensing area of ​​the capacitive sensor faces the inner side of the gripper, that is, the direction of the object when grasping. The capacitive sensor is specifically a capacitive force sensor, which has two structural layers arranged opposite each other. Each structural layer is provided with a mutual capacitor located in the middle and four self-capacitors evenly distributed around the mutual capacitor. The grasping control method of this embodiment of the present invention includes the following steps:

[0084] Step S1: Pre-identification phase. Before performing a grasping operation, the capacitive sensor collects sensor data when the gripper is not in contact with the object to be grasped. This data is then fed into a pre-trained machine learning model to identify the category of the object to be grasped. Based on the mapping relationship between object category and grasping force, the grasping target force is derived from the identified object category.

[0085] Step S2, the grasping stage, optimizes and controls the opening amount of the gripper based on the contact force collected by the capacitive sensor, so that the gripper grasps the object with the grasping target force.

[0086] In some embodiments, during the pre-recognition phase, the present invention proposes different pre-recognition methods based on the specific usage scenarios of the gripper. Specifically, for scenarios where recognition accuracy is not high but recognition speed is required, and some objects to be recognized must be inaccessible, a static pre-recognition method is used for pre-recognition; for scenarios where recognition accuracy is high and all objects to be recognized are accessible, a dynamic pre-recognition method is used for pre-recognition.

[0087] In some embodiments, see Figure 1 In the left part of the middle dotted line, the static pre-identification method in step S1 includes the following steps:

[0088] Step S11: Before performing the grasping operation, the gripper is in an idle state and collects a round of capacitance data as the reference capacitance data to be identified. Then, the gripper is moved to the vicinity of the object to be identified and collects a round of capacitance data as the static capacitance data to be identified.

[0089] Step S12: performing data enhancement on the reference capacitance data to be identified and the static capacitance data to be identified collected in step S11 to obtain enhanced static capacitance data to be identified;

[0090] Step S13: Input the enhanced static capacitance data to be identified into the pre-trained first machine learning model to perform non-contact object classification pre-identification to obtain the object category L p ;

[0091] Step S14: Based on the mapping relationship between object category and grasping force, the object category L identified in step S13 is p Get the grasping target force F p .

[0092] Furthermore, in step S11, before performing the grasping operation, the gripper is kept at its maximum opening and in a stationary state in an idle state, and capacitance sensor data on each gripping arm of the gripper is collected as reference capacitance data to be identified.

[0093] Furthermore, in step S11, before performing the grasping operation, the gripper is moved to the vicinity of the object to be identified, the gripper is kept at the maximum opening and in a stationary state, and capacitance sensor data on each gripping arm of the gripper is collected as static capacitance data.

[0094] Furthermore, step S12 performs data enhancement on the reference capacitance data to be identified and the static capacitance data to be identified collected in step S11, specifically: filtering the reference capacitance data to be identified and the static capacitance data to be identified to reduce the influence of high-frequency noise on the data, then obtaining the mean of the data, and subtracting the mean of the reference capacitance data to be identified from the mean of the static capacitance data to be identified to obtain the enhanced static capacitance data to be identified.

[0095] Furthermore, in step S13, the pre-trained first machine learning model is obtained according to the following steps:

[0096] Step S131: static classification data set construction, specifically including:

[0097] Step S1311, data acquisition basic parameter settings: set the number e of the current round of data, initialize e=1; set the number of repeated acquisitions of the current round of data to K c ; Set the collection interval between two adjacent data to t c , flexibly select according to the sampling frequency and response speed of the capacitance sensor; set the cutoff frequency f of the frequency domain filterc ; Set the jaw opening to the maximum value and keep the jaws in an empty state; the collected data includes: sensor mutual capacitance data C i , i is the sensor number;

[0098] In one embodiment of the present invention, the number of repeated data collections per round is set to K. c =100; Set the interval t between two adjacent data collections c 0.03s≤t c ≤0.1s; After experimental testing, the cutoff frequency f of the frequency domain filter is set c =15; There are two capacitive force sensors on the clamping jaw, each installed on the inner side of a corresponding clamping arm;

[0099] Step S1312: for the object S collected in the current round e , static data collection is carried out in two stages:

[0100] The first stage, reference signal acquisition: make the object S e Located outside the detection area of ​​all capacitive sensors, the gripper maintains the maximum opening amount, based on the acquisition interval t c Perform a single round of data acquisition to obtain the reference signal C of the capacitive sensor i bi , if the number of single-round collections has reached K c times, then the reference signal C bi Collection completed;

[0101] The second stage, static signal acquisition: move the gripper to the object S e Nearby, make the object S e Located within the detection area of ​​all capacitive sensors, the gripper maintains the maximum opening, based on the acquisition interval t c Perform a single round of data acquisition to obtain the static signal C of the capacitive sensor i si , if the number of single-round collections reaches K c times, then the static signal C si Collection completed;

[0102] Step S1313: Based on the cutoff frequency f c For reference signal C bi and static signal C si Filter them separately to get the reference mutual capacitance mean μ of capacitance sensor i bi and the static mutual capacitance μ si , and make the difference between the two to get the enhanced static capacitance data μ sdi :

[0103] μ sdi =μ bi -μ si

[0104] Based on the object S to be recognized corresponding to the data number e e , obtain the static classification data D s :[μ sdi ,L e , where L e is the class label of the object to be recognized corresponding to the data number e of this data;

[0105] Step S1314: If e < E, then set e = e + 1, and repeat steps S1312 to S1313 until e = E, and end the acquisition of the static classification data D s Collect all the obtained static classification data as a static classification data set. Among them, the static classification data set needs to be divided into a static classification training set and a static classification test set. Generally, the static classification training set and the static classification test set can be divided according to a ratio of 8:2 or 9:1;

[0106] In an embodiment of the present invention, for the use scenario of the gripper, the static classification data set constructed contains a total of E = 8 types of static classification data of different objects.

[0107] [[ID=​​​​​​​​​​​​​​​​Step S11': Before performing the grasping operation, the gripper is in an idle state and collects a round of capacitance data as the reference capacitance data to be identified. The gripper is then moved near the object to be identified and gradually closed until it fits the object to be identified, collecting the dynamic capacitance data to be identified during this process.

[0113] Step S12', performing data enhancement on the reference capacitance data to be identified and the dynamic capacitance data to be identified collected in step S11' to obtain enhanced dynamic capacitance data to be identified;

[0114] Step S13', input the enhanced dynamic capacitance data to be identified into the pre-trained second machine learning model, perform non-contact object classification pre-identification, and obtain the object category L p ;

[0115] Step S14': Based on the mapping relationship between object category and grasping force, the object category L identified in step S13' is p Get the grasping target force F p .

[0116] Furthermore, in step S11 ′, before performing the grasping operation, the gripper is kept in a maximum opening amount and in a static state in an idle state, and capacitance sensor data on each gripping arm of the gripper is collected as reference capacitance data to be identified.

[0117] Furthermore, in step S11', before performing the grasping operation, the gripper is moved to the vicinity of the object to be identified, and the gripper is controlled to open gradually from the maximum amount until it fits the object to be identified. During each closing step, the capacitance sensor data on each clamping arm of the gripper is collected as the dynamic capacitance data to be identified and stored in a sequence.

[0118] Furthermore, in step S12', data enhancement is performed on the reference capacitance data to be identified and the dynamic capacitance data to be identified collected in step S11', specifically: filtering the reference capacitance data to be identified to reduce the influence of high-frequency noise on the data, and then obtaining the mean of the data; calculating the mean and variance of the dynamic capacitance data to be identified, normalizing it, and flipping the entire sequence back and forth to perform polynomial fitting to obtain fitting parameters; using the fitting parameters, the mean difference between the reference capacitance data to be identified and the dynamic capacitance data to be identified, and the variance of the dynamic capacitance data to be identified to construct the enhanced dynamic capacitance data to be identified.

[0119] Furthermore, in step S13', the pre-trained second machine learning model is obtained according to the following steps:

[0120] Step S131′, dynamic classification data set construction, specifically includes:

[0121] Step S1311', data acquisition basic parameter settings: set the number e of the current round of data, initialize e=1; set the number of repeated acquisitions of the current round of data to K c ; Set the collection interval between two adjacent data to t c , flexibly select according to the sampling frequency and response speed of the capacitance sensor; set the end force value of the current round data to F e ; Set the step length l of the gripper closing s ; Set the cutoff frequency f of the frequency domain filter c ; Data fixed interval d; The collected data includes: sensor mutual capacitance data C i , the force data F output by the sensor based on the self-capacitance data i and the current opening amount of the gripper l c , i is the sensor number;

[0122] In one embodiment of the present invention, the basic parameters are set as follows:

[0123] Set the number of repeated data collections K for each round c =100;

[0124] Set the interval between two consecutive data collections to t c It is necessary to flexibly select the sampling frequency of the mutual capacitance sensor and the response speed of the force sensor. In this case, when collecting the benchmark data, 0.03s≤t c ≤0.1s; When collecting dynamic data, adjust the collection interval t between two adjacent data c , t c =0.1s;

[0125] It is necessary to consider the noise fluctuation of the capacitive sensor in the idle state. In order to avoid damaging the object to be identified when the sensor is close to the object, the design is 0.2N≤F e ≤0.5N;

[0126] Set the step length l of the gripper closing s , l s =0.475mm, the step length is 5 times the minimum opening of the jaws;

[0127] After experimental testing, the cutoff frequency f of the frequency domain filter is set c =15; data fixed interval d=1s;

[0128] Step S1312': for the object S collected in the current round e , dynamic data collection is carried out in two stages:

[0129] The first stage, the reference signal acquisition: keep the gripper in the empty state with the maximum opening amount, and make the object S eLocated outside the detection area of ​​all capacitive sensors, based on the acquisition interval t c Perform a single round of data acquisition to obtain the reference signal C of the capacitive sensor i bi , if the number of acquisitions has reached K c times, then the reference signal C bi Collection completed;

[0130] The second stage, sequence signal acquisition: move the gripper to the object S e Nearby, make the object S e Located within the detection area of ​​all capacitive sensors, the gripper starts from the maximum opening and follows the set step length l s Close step by step, collect data of the closing process after each step, and wait for a collection interval t c Then, the data of the next closing process is collected; after each step of the jaw closing, if the maximum value of the normal force data collected by each capacitive sensor is higher than the end force value F e , or the gripper is completely closed, the data of each closing process of the capacitance sensor i is collected to form the sequence signal C of the capacitance sensor i di , the second stage ends, otherwise the sequence signal acquisition continues;

[0131] Step S1313', based on the cutoff frequency f c Reference signal C for capacitance sensor i bi Filter and obtain the reference mutual capacitance mean μ of capacitance sensor i bi Based on the fixed interval d of the data, the sequence signal C of the capacitance sensor i is di Represent it in the time domain, and perform data fitting after flipping the head and tail to obtain the fitting parameter p of the capacitance sensor i i , and calculate the mutual capacitance mean μ of the sequence signal of capacitance sensor i di and standard deviation σ di , and then get the difference μ between the two ddi :

[0132] μ ddi =μ bi -μ di

[0133] Based on the object S corresponding to the data number e e , get dynamic classification data D d :[p i ,μ ddi ,σ di ,L e ], where L e The category label of the object to be identified corresponding to the data number e of the data; the fixed interval d of the data is taken from the time interval t used when collecting the data c ;

[0134] In a specific embodiment of the present invention, the fitting parameter p i is obtained by polynomial fitting. The order of the polynomial is set to 2, and three fitting parameters will be generated for each sequence signal;

[0135] Step S1314': If e < E, let e = e + 1, and repeat steps S1312' to S1313' until e = E, and end the acquisition of dynamic classification data D s Collect all the obtained dynamic classification data as a dynamic classification data set. Among them, the dynamic classification data set needs to be divided into a dynamic classification training set and a dynamic classification test set. Generally, the static classification training set and the static classification test set can be divided according to a ratio of 8:2 or 9:1;

[0136] In an embodiment of the present invention, for the usage scenario of the gripper, the dynamic classification data set constructed contains dynamic classification data of E = 8 different types of objects.

[0137] Step S132': Use the dynamic classification training set to train the second machine learning model, and save the second machine learning model in the form of parameters;

[0138] Preferably, the second machine learning model can adopt the support vector machine SVC model for multi-classification tasks. By finding the separation boundary between different classes, the classifier is trained, where the kernel function is set to a linear kernel function;

[0139] Step S133': Use the dynamic classification test set to test the second machine learning model obtained in step S132'. Based on the test results, draw a confusion matrix. If the data is completely separated without any confusion, or the confusion is serious, then additionally and discontinuously, supplement the data volume for each object category to be recognized according to steps S1312' to S1313', and retrain according to step S132'.

[0140] Furthermore, in step S14', the mapping relationship between the object category and the grasping force is the same as that in step S14, which will not be elaborated here.

[0141] It should be noted that the pre-identification method provided in the embodiment of the present invention expands the modalities that can be recognized by the capacitive sensor installed on the gripper, specifically the object category. In the embodiment of the present invention, the self-capacitance of the capacitive sensor is used to measure the contact force, and the mutual capacitance is used to predict the object category.

[0142] In some embodiments, in the grasping stage of step S2, based on the grasping target force F obtained in step S1 p , perform a force-controlled grasping operation, including the following steps:

[0143] Step S21: Setting the approach force threshold F l1 , read the data of the capacitance sensor F c , and calculate the maximum value F of all sensor force data on each gripping arm max ;

[0144] Step S22: Based on the maximum value F of the current force data of each sensor max , judge the current gripping state: if F max ≤F l1 , the gripper is not close to the object to be grasped, and step S23 is executed; if F max >F l1 , the gripper is in contact with the object to be grasped, and step S24 is executed;

[0145] Step S23: The gripper is based on the step length l s , closing gradually, the interval time between two adjacent closing steps of the gripper is set to t c Then, the maximum value F of each sensor force value data is updated again according to step S21. max , and judging the current clamping state based on step S22;

[0146] Step S24: Use the optimization method to close the gripper to the target gripping force F. p The reason why the position corresponding to the target force cannot be reached directly is that the jaw opening amount is a discrete amount in the contact state. Each index movement corresponds to a range of force values, and the contact force in the closing direction and the opening direction are different. The specific steps are as follows:

[0147] Step S241: Set the optimization step length l j , l j =0.19mm; set the optimized attenuation coefficient β, and its initial value is set to β0=1;

[0148] Step S242: Construct a step size optimization function J(ΔF) to represent the change Δl of the clamping jaw closing position each time to control the clamping jaw closing. The expression is as follows:

[0149]

[0150] J(ΔF)=β n ×m(ΔF)×l j

[0151] Among them, l min Indicates the minimum closing amount that the gripper can perform. In this embodiment, l min =0.095mm; the difference ΔF is based on the maximum value F of the current force data max With the target force F pget:

[0152] ΔF=F p -F max

[0153] m(ΔF) is the sign of the optimization direction, which is converted from the force difference ΔF:

[0154]

[0155] β n To optimize the iterative value of the attenuation coefficient β:

[0156]

[0157] Where n is the number of times the force difference ΔF changes its sign, initially n=0, N m is the set number of times to start decay, β min is the lower limit of the attenuation coefficient; in this embodiment, N is set m =5,β min =0.4;

[0158] Step S243: Get the jaw opening amount feedback, or get the current jaw opening amount x based on the accumulated step length. c , and the current opening amount of the gripper x c Save to data pool X l In the data pool X l The latest N l The jaw opening amount data, in this embodiment, N l =10;

[0159] Step S244: The gripper moves under the control of the position change Δl. When the saved gripper opening amount data has satisfied N l When requested, a stop detection flag is performed stop :

[0160]

[0161] Where x is the data pool X l Data in

[0162] Step S245: If flag stop = False, then repeat steps S242 to S244 to continue the force-controlled gripping process; if flag stop = True, it means that the target force F has been reached p The corresponding nearest position is clamped.

[0163] It is understandable that when the robot is performing gripping, in order to deal with the information loss caused by the camera image being blocked, or to deal with the environment where the camera cannot be observed due to the deployment position, it is necessary to use non-visual additional information to assist in the gripping judgment. The first embodiment of the present invention proposes a gripping control method based on pre-identification. The present invention can maintain a certain distance before the gripper formally applies force to the gripped object, and provide pre-identification information in a non-contact form, or provide pre-identification information based on the sequence sensing information of the approaching process during the gripping approach process, so that the robot has richer perception information about the gripped object, performs preliminary classification judgment of the object to be gripped, or pre-estimates the gripping force value to avoid excessive gripping force damaging the object to be gripped. In addition, the present invention can use a force sensor to expand the force application range of the gripper itself, break through the range limit of the gripper's force application, and expand the sensor's force detection range to the gripper's gripping force range through algorithm control and sensor force information, thereby realizing a force-controlled gripping task based on the sensor's force measurement accuracy, thereby achieving the purpose of increasing the versatility of the equipment. For example, a gripper with a force range of 32N-160N can, after being equipped with a sensor with a range of 0N-20N, grip fragile objects such as tofu with a force of 2N. However, using the original gripper, the minimum force of 32N will still crush and destroy the tofu.

[0164] A second embodiment of the present invention provides a pre-identification based grasping control device, comprising:

[0165] a capacitive sensor mounted on the gripping arm of the gripper on the side facing the object to be gripped;

[0166] a pre-identification module configured to, before performing a grasping operation, collect sensor data from the gripper when it is not in contact with an object based on the capacitive sensor, input the data into a pre-trained machine learning model, identify the category of the object to be grasped, and derive a grasping target force from the identified object category based on a mapping relationship between the object category and the grasping force;

[0167] The optimization control module is configured to optimize the opening amount of the clamping jaws based on the contact force collected by the capacitive sensor, so that the clamping jaws can clamp the object with the grasping target force.

[0168] It should be noted that the above explanations of the embodiment of the grasping control method are also applicable to the grasping control device of this embodiment and will not be repeated here.

[0169] In order to implement the above embodiment, the embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored. The program is executed by a processor to perform the grasping control method of the above embodiment.

[0170] Reference below Figure 2, which shows a schematic diagram of the structure of an electronic device suitable for implementing an embodiment of the present invention. It should be noted that the electronic devices in the embodiments of the present invention may include, but are not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), and in-vehicle terminals (e.g., in-vehicle navigation terminals), as well as fixed terminals such as digital TVs, desktop computers, and servers. Figure 2 The electronic device shown is only an example and should not limit the functions and scope of use of the embodiments of the present invention.

[0171] like Figure 2 As shown, the electronic device may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 101, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 102 or a program loaded from a storage device 108 into a random access memory (RAM) 103. Various programs and data required for the operation of the electronic device are also stored in the RAM 103. The processing device 101, the ROM 102, and the RAM 103 are connected to each other via a bus 104. An input / output (I / O) interface 105 is also connected to the bus 104.

[0172] Typically, the following devices may be connected to the I / O interface 105: an input device 106 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, etc.; an output device 107 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 108 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 109. The communication device 109 may allow the electronic device to communicate with other devices wirelessly or by wire to exchange data. Figure 2 The electronic device is shown with various devices, but it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed instead.

[0173] In particular, according to an embodiment of the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, this embodiment includes a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via the communication device 109, or installed from the storage device 108, or installed from the ROM 102. When the computer program is executed by the processing device 101, the above-mentioned functions defined in the method of the embodiment of the present invention are performed.

[0174] It should be noted that the computer-readable medium of the present invention can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or component, or any combination thereof. More specific examples of computer-readable storage media can include, but are not limited to, an electrical connection having one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device, or component. In the present application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. This propagated data signal can take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium may be transmitted using any suitable medium, including but not limited to wires, optical cables, RF (radio frequency), etc., or any suitable combination thereof.

[0175] The computer-readable medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device.

[0176] The computer-readable medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device executes the grasping control method.

[0177] Computer program code for performing the operations of the present invention may be written in one or more programming languages, or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, Python, and conventional procedural programming languages ​​such as "C-" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on the remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0178] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0179] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of such features. Throughout the description of this application, "plurality" means at least two, for example, two, three, etc., unless otherwise specifically defined.

[0180] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code comprising one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may be performed out of the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present application belong.

[0181] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection having one or more wires (electronic devices), a portable computer disk cartridge (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disc read-only memory (CDROM). Furthermore, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or otherwise processing it in a suitable manner if necessary, and then storing it in a computer memory.

[0182] It should be understood that various parts of the present application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used to implement: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0183] Those skilled in the art will understand that all or part of the steps carried out in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a program, and the developed program can be stored in a computer-readable storage medium. When the program is executed, it includes one of the steps of the method embodiment or a combination thereof.

[0184] In addition, the functional units in the various embodiments of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into a module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0185] The storage medium mentioned above may be a read-only memory, a magnetic disk, or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present application. Persons skilled in the art may make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present application.

Claims

1. A grasping control method based on pre-identification, characterized in that: A capacitive sensor is provided on the gripping arm of the gripper facing the object to be gripped. The gripping control method includes: Step S1: Before performing a grasping operation, the capacitive sensor collects sensor data when the gripper is not in contact with the object, inputs the data into a pre-trained machine learning model, identifies the category of the object to be grasped, and derives the grasping target force from the identified object category based on the mapping relationship between the object category and the grasping force; Step S2, the grasping stage, optimizes and controls the opening amount of the gripper based on the contact force collected by the capacitive sensor, so that the gripper grasps the object with the grasping target force.

2. The grasping control method according to claim 1, characterized in that: The step S1 specifically includes: Step S11: Before performing the grasping operation, the gripper is in an idle and stationary state, and a round of capacitance data is collected as reference capacitance data to be identified. Then, the gripper is moved near the object to be grasped, and a round of capacitance data is collected in a stationary state as static capacitance data to be identified. Step S12: performing data enhancement on the reference capacitance data to be identified and the static capacitance data to be identified to obtain enhanced static capacitance data to be identified; Step S13: Input the enhanced static capacitance data to be identified into the pre-trained first machine learning model to perform non-contact object classification pre-identification to obtain the object category L p ; Step S14: According to the mapping relationship between object category and grasping force, the object category L p Get the grasping target force F p .

3. The grasping control method according to claim 2, characterized in that: The data enhancement includes: filtering the reference capacitance data to be identified and the static capacitance data to be identified, then calculating the mean of the data, and subtracting the mean of the reference capacitance data to be identified from the mean of the static capacitance data to be identified to obtain the enhanced static capacitance data to be identified.

4. The grasping control method according to claim 2, characterized in that: The first machine learning model is trained using a static classification dataset, where the static classification dataset is obtained according to the following steps: Step S1311, data acquisition basic parameter settings: set the number e of the current round of data, initialize e=1; set the number of repeated acquisitions of the current round of data to K c ; Set the collection interval between two adjacent data to t c ; Set the cutoff frequency f of the frequency domain filter c ; Set the jaw opening to the maximum value and keep the jaws in an empty state; Step S1312: for the object S collected in the current round e , static data collection is carried out in two stages: The first stage, reference signal acquisition: make the object S e Located outside the detection area of ​​all capacitive sensors on the gripper, based on the acquisition interval t c and the number of repeated acquisitions K c Perform a single round of data acquisition to obtain the reference signal C of the capacitive sensor i bi ; The second stage, static signal acquisition: move the gripper to the object S e Nearby, make the object S e Located within the detection area of ​​all capacitive sensors on the gripper, based on the acquisition interval t c and the number of repeated acquisitions K c Perform a single round of data acquisition to obtain the static signal C of the capacitive sensor i si ; Step S1313: Based on the cutoff frequency f c For reference signal C bi and static signal C si Filter them separately to get the reference mutual capacitance mean μ of capacitance sensor i bi and the static mutual capacitance μ si , and make the difference between the two to get the enhanced static capacitance data μ of capacitance sensor i sdi : m sdi =μ bi -m si Based on the object S to be identified corresponding to the data number e e , get static classification data D s :[μ sdi ,L e ], where L e The category label of the object to be identified corresponding to the data number e of the data; Step S1314: If e < E, then set e = e + 1, and repeat Step S1312 to Step S1313 until e = E, and end the static classification data D s Collect, and use all the obtained static classification data as a static classification data set. E is the number of object categories contained in the static classification data set 5. The grasping control method according to claim 1, characterized in that: The step S1 specifically includes: Step S11': Before performing the grasping operation, the gripper is in an idle and stationary state, and a round of capacitance data is collected as the reference capacitance data to be identified; then the gripper is moved near the object to be identified, and the gripper is controlled to gradually close until it just fits the object to be identified, and the dynamic capacitance data to be identified in this process is collected; Step S12′, performing data enhancement on the reference capacitance data to be identified and the dynamic capacitance data to be identified to obtain enhanced dynamic capacitance data to be identified; Step S13', input the enhanced dynamic capacitance data to be identified into the pre-trained second machine learning model, perform non-contact object classification pre-identification, and obtain the object category L p ; Step S14': According to the mapping relationship between object category and grasping force, the object category L p Get the grasping target force F p .

6. The grasping control method according to claim 5, characterized in that: The data enhancement includes: filtering the reference capacitance data to be identified and then obtaining the mean of the data; calculating the mean and variance of the dynamic capacitance data to be identified, normalizing the data, and flipping the entire sequence forward and backward to perform polynomial fitting to obtain fitting parameters; The enhanced dynamic capacitance data to be identified is constructed using the fitting parameters, a mean difference between the reference capacitance data to be identified and the dynamic capacitance data to be identified, and a variance of the dynamic capacitance data to be identified.

7. The grasping control method according to claim 5, characterized in that: The second machine learning model is trained using a dynamic classification dataset, where the dynamic classification dataset is obtained according to the following steps: Step S1311', data acquisition basic parameter settings: set the number e of the current round of data, initialize e=1; set the number of repeated acquisitions of the current round of data to K c ; Set the collection interval between two adjacent data to t c ; Set the end force value of the current round data to F e ; Set the step length l of the gripper closing s ; Set the cutoff frequency f of the frequency domain filter c ;Data fixed interval d; Step S1312': for the object S collected in the current round e , dynamic data collection is carried out in two stages: The first stage, the reference signal acquisition: keep the gripper in the empty state with the maximum opening amount, and make the object S e Located outside the detection area of ​​all capacitive sensors on the gripper, based on the acquisition interval t c and the number of repeated acquisitions K c Perform a single round of data acquisition to obtain the reference signal C of the capacitive sensor i bi ; The second stage, sequence signal acquisition: move the gripper to the object S e Nearby, make the object S e Located in the detection area of ​​all capacitive sensors on the gripper, the gripper starts from the maximum opening and follows the set step length l s Close step by step, collect data of the closing process after each step, and wait for a collection interval t c Then, the data of the next closing process is collected; after each step of the jaw closing, if the maximum value of the normal force data collected by each capacitive sensor is higher than the end force value F e , or the gripper is completely closed, the collected closing process data of the capacitance sensor i is combined into a sequence signal C of the capacitance sensor i di ; Step S1313', based on the cutoff frequency f c Reference signal C for capacitance sensor i bi Filter and obtain the reference mutual capacitance mean μ of capacitance sensor i bi Based on the fixed interval d of the data, the sequence signal C of the capacitance sensor i is di Represent it in the time domain, and perform data fitting after flipping the head and tail to obtain the fitting parameter p of the capacitance sensor i i , and calculate the mutual capacitance mean μ of the sequence signal of capacitance sensor i di and standard deviation σ di , and then get the difference μ between the two ddi : m ddi =μ bi -m di Based on the object S corresponding to the data number e e , get dynamic classification data D d :[p i ,μ ddi ,σ di ,L e ], where L e The category label of the object to be identified corresponding to the data number e of the data; Step S1314': If e < E, then set e = e + 1, and repeat Step S1312' to Step S1313' until e = E, and end the dynamic classification data D s Collect, and use all the obtained dynamic classification data as the dynamic classification data set. E is the number of object categories contained in the dynamic classification data set 8. The grasping control method according to claim 1, characterized in that: Step S2 specifically includes: Step S21: Setting the approach force threshold F l1 , read the data of the capacitance sensor F c , and count the maximum value F of the force data of all capacitive sensors on each gripping arm max ; Step S22: Based on the maximum value F of the current force value data stored in each capacitance sensor max , judge the current gripping state: if F max ≤F l1 , the gripper is not close to the object to be grasped, and step S23 is executed; if F max >F l1 , the gripper is in contact with the object to be grasped, and step S24 is executed; Step S23: The gripper is based on the step length l s , closing gradually, the interval time between two adjacent closing steps of the gripper is set to t c Then, according to step S21, the maximum value F of the force data of each capacitance sensor is updated. max , and judging the current clamping state based on step S22; Step S24: Use the optimization method to close the gripper to the target gripping force F. p The closest position, the steps are as follows: Step S241: Set the optimization step length l j ; Set the optimized attenuation coefficient β and initialize it to β0=1; Step S242: Construct a step size optimization function J(ΔF) to represent the change Δl of the clamping jaw closing position each time to control the clamping jaw closing. The expression is as follows: J(ΔF)=β n ×m(ΔF)×l j Among them, l min Indicates the minimum closing amount that the gripper can perform; the force difference ΔF = F p -F max ; m(ΔF) is the sign of the optimization direction, which is converted from the force difference ΔF: β n To optimize the iterative value of the attenuation coefficient β: Where n is the number of times the force difference ΔF changes its sign, initially n=0, N m is the set number of times to start decay, β min is the lower limit of the attenuation coefficient; Step S243: Get the jaw opening amount feedback or obtain the current jaw opening amount x based on the accumulated step length. c , and the current opening amount of the gripper x c Save to data pool X l In the data pool X l The latest N l The jaw opening data; Step S244: The gripper moves under the control of the position change Δl. When the saved gripper opening amount data has satisfied N l When requested, a stop detection flag is performed stop : Where x is the data pool X l Data in; Step S245: If flag stop = False, then repeat steps S242 to S244 to continue the force-controlled gripping process; if flag stop = True, it indicates that the grasping target force F has been reached p The corresponding nearest position is clamped.

9. A grasping control device based on pre-identification, characterized in that: include: a capacitive sensor mounted on the gripping arm of the gripper facing the object to be gripped; a pre-identification module configured to, before performing a grasping operation, collect sensor data from the gripper when it is not in contact with an object based on the capacitive sensor, input the data into a pre-trained machine learning model, identify the category of the object to be grasped, and derive a grasping target force from the identified object category based on a mapping relationship between the object category and the grasping force; The optimization control module is configured to optimize the opening amount of the clamping jaws based on the contact force collected by the capacitive sensor, so that the clamping jaws can clamp the object with the grasping target force.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the gripping control method according to any one of claims 1 to 8.

Images