A robot regrasping optimization method based on tactile primitive slip feature feedback

Abstract

The application belongs to the technical field of robot intelligent grasping, and relates to a robot regrasping optimization method based on tactile map sliding feature feedback, comprising the following steps: step 1, acquiring tactile modal information by using an optical tactile sensor arranged on a mechanical arm, that is, perceiving and acquiring tactile maps of a contact area; step 2, performing elliptical parameter fitting based on the acquired tactile modal information; step 3, constructing a trial grasping coordinate system of the mechanical arm; step 4, performing trial grasping on the object to make the object slide, calculating a tactile perception sliding vector of the object, and inferring a pose change of the object during the sliding; and step 5, performing corresponding optimization adjustment on different degrees of sliding of the object during the trial grasping, and reconstructing a regrasping pose of the mechanical arm. The application solves the problem that the existing grasping adjustment method depends on trial and error and cannot fully utilize tactile modal information.

Description

Technical Field

[0001] This invention belongs to the field of intelligent robot grasping technology, and relates to an optimized method for robot re-grasping based on tactile primitive sliding feature feedback. Background Technology

[0002] Vision-based grasping detection methods typically rely on information such as the object's size, texture, and color to obtain the grasping pose. However, they cannot perceive the contact surface characteristics of the grasped object, changes in its center of mass pose, changes in contact force, or the relative motion between the object and the gripper during the grasping process. Therefore, they cannot evaluate or provide feedback on existing grasping configurations. Consequently, when unstable grasping occurs, the vision-inferred grasping pose needs further optimization to meet requirements for stability, safety, and efficiency. Summary of the Invention

[0003] To address the aforementioned technical problems in existing technologies, this invention proposes a robot re-grasping optimization method based on tactile primitive slip feature feedback. Based on Canny edge detection and least-squares ellipse fitting, it extracts tactile primitives perceived by optical tactile sensors and further identifies tactile primitive slip features. These features are categorized into three types based on the severity of slippage during the initial grasping attempt: initial slippage, local slippage, and full slippage. By combining a visual grasping detection map, tactile feedback features, and a visual-tactile fusion quality analysis model, a corresponding adaptive optimization adjustment strategy is formulated to stably reconstruct the grasping pose. The specific technical solution is as follows:

[0004] A robot re-grasping optimization method based on tactile primitive sliding feature feedback includes the following steps:

[0005] Step 1: Use optical tactile sensors mounted on the robotic arm to acquire tactile modal information, that is, to sense and acquire tactile primitives of the contact area;

[0006] Step 2: Based on the acquired tactile modal information, perform ellipse parameter fitting;

[0007] Step 3: Construct the coordinate system for the robotic arm's trial grasping;

[0008] Step 4: Perform a trial grasp of the object to induce it to slide, calculate the tactile perception sliding vector of the object, and infer the pose change of the object during the sliding process.

[0009] Step 5: Optimize and adjust the different degrees of slippage that occur during the trial grasping process to reconstruct the robotic arm's grasping posture.

[0010] Further, step 1 specifically involves: using a vision-based tactile sensor to determine whether the robotic arm gripper is in contact with an object by observing changes in the contact image; if so, acquiring a sensor contact image; otherwise, acquiring an initial non-contact tactile image; acquiring the area, position, and direction of the object contact; and acquiring the movement of the contact part on the sensor surface through continuous video images.

[0011] The tactile sensor estimates the contact area of ​​an object through single-point and multi-point features, and realizes the perception of the contact area through the extraction and matching of geometric features, thereby obtaining a difference map with the initial image.

[0012] Utilizing the properties of elastomer materials and their slightly convex surface configuration, which manifest as elliptical or circular contact areas in imaging images, the contact area is described by extracting elliptical primitives from the tactile sensing modality and uniformly expressed as A(p,a,b,θ). Here, p(x,y) is the center coordinate of the contact ellipse in the tactile sensor image coordinate system, a is the semi-major axis of the contact ellipse, b is the semi-minor axis of the contact ellipse, and θ is the deflection angle of the contact ellipse, with the major axis of the ellipse as a reference. Its value is the angle of deflection of the major axis in the horizontal direction, and its range is [-π / 2,π / 2]. When the shape of the contact area is circular, the major axis is equal to the minor axis, i.e., a = b, and θ is any value within [-π / 2,π / 2].

[0013] Furthermore, step 2 specifically includes:

[0014] Step 2.1: After performing grayscale and morphological processing on the difference image, the Canny edge detection operator is used to extract the edge of the contact area;

[0015] Step 2.2: Fit the detected and extracted edge image using a geometric ellipse fitting method based on the least squares method.

[0016] Furthermore, step 3 specifically includes:

[0017] Step 3.1: Using the crawling configuration map G(x,y,z,r) inferred by the crawling detection algorithm at the current time... x ,r y ,r z Based on W), establish a grabbing coordinate system o-xyz with the grabbing center coordinates as the origin;

[0018] Step 3.2: Two face-to-face tactile sensors are installed on the working surface of the robotic arm gripper. The coordinate systems of the two contact images acquired by the tactile sensors are redefined. The right image coordinate system o-xz is established with the right sensor image as the reference and the center of the 320×240 imaging area as the coordinate system center. The left image is established with the left sensor image as a mirror image of the o-xyz coordinate system with the center of the image as the center. The contact areas generated by the left and right images are marked as A1 and A2, respectively, and the corresponding centers are p1(x1,z1) and p2(x2,z2), respectively. The coordinate values ​​of p(x,z) are the mapping values ​​from the tactile image space to the actual contact surface size.

[0019] Furthermore, step 4 specifically includes:

[0020] Step 4.1: Describe the initial state in the form of deviations; the initial deviations of the object include the initial position deviations (Δx, Δy, Δz) and the initial angle deviations (Δr). x ,Δr y ,Δr z ), where Δy is obtained from the camera's field of view, and its magnitude represents the distance of the object from the zox plane of the grasping coordinate system o-xyz; the pose deviation at the initial moment is shown in the following formula:

[0021]

[0022] Step 4.2: Obtain the motion characteristics of the contact ellipse by reading the video stream of the tactile image;

[0023] Step 4.3: Filter the ellipse parameters, describe the ellipse state vector, and finally determine the pose change of the object during the slippage:

[0024]

[0025] Furthermore, step 5 specifically includes:

[0026] Step 5.1: Optimize and adjust the initial slippage that occurs during the trial grasping process;

[0027] Step 5.2: Optimize and adjust the local slippage that occurs during the trial grasping process;

[0028] Step 5.3: Optimize and adjust the overall slippage that occurs during the trial grasping process.

[0029] Furthermore, step 5.1 specifically involves the following optimization adjustments based on the initial characteristics of convergent and bounded slip:

[0030] (1) The center of the desired contact area is within the sensor’s sensing range. If the center of the fitted ellipse of the contact detection is outside the sensor’s sensing range when the contact is made, then a full slip occurs. When adjusting, the center of the contact area should be moved along the direction closer to the center of the sensor’s sensing range.

[0031] (2) For small initial slip motions, i.e., the center of the fitted ellipse is within the sensor's sensing range, it is only necessary to adjust the pose of the end effector in the opposite direction of the object's pose change. In particular, due to the influence of gravity, the object naturally has a tendency to move along the negative z-axis. For changes in the z-axis direction, scaling is used, with a scaling factor of 0.6, i.e., capturing the configuration map G(x,y,z,r) x ,r y ,r z ,W) is adjusted to:

[0032] G′(x-dx,y-dy,z-dz×0.6,r x -dr x ,r y -dr y ,r z -dr z ,W).

[0033] Further, step 5.2 specifically involves: referring to the current object's grasping position map, searching along the x-axis of the current grasping coordinate system towards the object's center of gravity at non-zero positions on the grasping position map; reconstructing the grasping configuration based on the points with the highest confidence near this area; obtaining a new grasping configuration through the spatial mapping function f; the search direction being determined by the two tactile sensors; if the object's dr y If the value is positive, then search along the negative x-axis; if dr y If the value is negative, then the search proceeds along the positive x-axis.

[0034] Furthermore, step 5.3 specifically involves: if the current crawling configuration undergoes a complete shift, discarding the current crawling area, replacing the crawling point on the crawling configuration map, discarding the peak confidence level on the current crawling location map, and searching for another DR (Depth Ranking). y For local peak values ​​exceeding the threshold, a new coordinate is generated from the crawl map, and a new crawl configuration is inferred from this coordinate, resulting in an updated crawl configuration. Finally, through multiple crawl quality analyses and local adjustments, the overall slip problem is transformed into a local slip or initial slip problem, thereby adjusting the crawl posture near the new crawl configuration until a stable crawl state is achieved.

[0035] Beneficial effects:

[0036] This invention addresses the sensing mechanism and imaging characteristics of optical sensors by describing the features of tactile contact primitives using morphological methods and completing contact ellipse detection based on edge detection and geometric features. It also assesses the stability of the grasped object based on the visual input image and tactile sensing image of continuous motion, while using the continuous motion information to evaluate the unstable motion trend of the object in the robot gripper arm.

[0037] This invention addresses different degrees and types of slippage trends by combining global information from a vision-based grasping map with local pose perception information based on tactile feedback. This effectively suppresses the unstable trend, allowing for a more effective grasping configuration during subsequent grasping operations, thereby enabling the robot to perform more stable and safer grasping tasks.

[0038] This invention solves the problem that existing grasping and adjustment methods rely on trial and error and cannot fully utilize tactile modal information. Attached Figure Description

[0039] Figure 1 This is a flowchart of a robot re-grasping optimization method based on tactile primitive sliding feature feedback according to the present invention;

[0040] Figure 2 Fitting image for contact area primitives;

[0041] Figure 3 To capture the coordinate system and tactile image coordinate system diagram;

[0042] Figure 4 A schematic diagram of the sliding feature of the contact ellipse between two frames of a video stream;

[0043] Figure 5 This is a schematic diagram illustrating the specific process of optimizing and adjusting the slippage of an object during a trial grasp, based on the present invention.

[0044] Figure 6 This is a schematic diagram of the force analysis when a robotic arm grasps a rod-shaped object;

[0045] Figure 7 This is a schematic diagram of the object model used in the grasping experiment of this invention embodiment;

[0046] Figure 8 This is a flowchart of the re-grabbing simulation experiment according to an embodiment of the present invention;

[0047] Figure 9 This invention provides a grasping prediction and an optimized grasping pose graph. Detailed Implementation

[0048] To make the objectives, technical solutions, and technical effects of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0049] like Figure 1 As shown, this invention proposes a robot re-grasping optimization method based on tactile primitive sliding feature feedback. This method achieves grasping state perception through the extraction and analysis of tactile primitives, thereby optimizing the grasping strategy. It can be applied to conventional six-axis industrial robotic arms equipped with optical tactile sensors. The specific implementation steps of this method are as follows:

[0050] Step 1: When using a robotic arm to grasp an object, optical tactile sensors are used to acquire tactile modal information, that is, to sense and acquire tactile primitives of the contact area.

[0051] Specifically, by using a vision-based tactile sensor to obtain tactile modal information through changes in the contact image, the following inferences are made:

[0052] (1) Whether there is contact with an object; if so, obtain the sensor contact image; otherwise, obtain the initial non-contact tactile image.

[0053] (2) The area, position, and direction of contact between the objects;

[0054] (3) Obtain the motion status of the contact part on the sensor surface through continuous video images;

[0055] Among them, when using a tactile sensor, the contact area of ​​an object is estimated by single-point and multi-point features, and the contact area is perceived by extracting and matching geometric features, thereby obtaining a difference map with the initial image.

[0056] Leveraging the characteristic that elastomer materials and their micro-convex surface configurations often manifest as elliptical or circular contact areas in imaging images, the contact area is described by extracting elliptical primitives from the tactile sensing modality and uniformly representing the contact area in the format A(p,a,b,θ). Here, p(x,y) represents the center coordinates of the contact ellipse in the tactile sensor image coordinate system, a is the semi-major axis length of the contact ellipse, b is the semi-minor axis length of the contact ellipse, and θ is the deflection angle of the contact ellipse, with the major axis as a reference, and its value is the angle of horizontal deflection of the major axis, ranging from [-π / 2, π / 2]. Specifically, when the contact area is circular, the major axis length is equal to the minor axis length, i.e., a = b, and θ is any value within [-π / 2, π / 2].

[0057] Step 2: Based on the acquired tactile modal information, perform ellipse parameter fitting, specifically including:

[0058] Step 2.1: After performing grayscale and morphological processing on the difference image, the Canny edge detection operator is used to extract the edge of the contact area;

[0059] Step 2.2: Fit the detected and extracted edge image using a geometric ellipse fitting method based on the least squares method, such as... Figure 2 As shown, the accuracy of ellipse parameter fitting is maintained while improving the detection speed.

[0060] Step 3: Construct the robotic arm's trial grasping coordinate system, specifically including:

[0061] Step 3.1: Using the crawling configuration map G(x,y,z,r) inferred by the crawling detection algorithm at the current time... x ,r y ,r z Based on W), a grasping coordinate system o-xyz is established with the grasping center coordinates as the origin. This system is used to conveniently describe the motion characteristics of the object in the visual and tactile sensing link after the gripper of the robotic arm has been stably closed.

[0062] Step 3.2: As Figure 3 As shown, in this embodiment, two face-to-face tactile sensors are installed on the working surface of the robotic arm gripper. To facilitate the description of contact information of the same object, the coordinate systems of the two contact images are redefined. The right image coordinate system o-xz is established with the imaging of the right sensor as the reference and the center of the 320×240 imaging area as the coordinate system center. The left image coordinate system o-xz is established with the center of the left image as the center of the o-xyz coordinate system mirrored by the zox plane of the o-xyz coordinate system. The contact areas generated by the left and right images are marked as A1 and A2, respectively, and the corresponding centers are p1(x1,z1) and p2(x2,z2), respectively. The coordinate values ​​of p(x,z) are the mapping values ​​from the tactile image space to the actual contact surface size.

[0063] Step 4: Perform a trial grasp of the object to induce slippage, calculate the object's tactile perception slippage vector, and infer the object's pose changes during the slippage. This includes:

[0064] Step 4.1: This invention uniformly describes the initial state as a deviation; the initial deviation of the object includes the initial position deviation (Δx, Δy, Δz) and the initial angle deviation (Δr). x ,Δr y ,Δr z ), where Δy is obtained from the camera's field of view, and its magnitude represents the distance of the object from the zox plane of the grasping coordinate system o-xyz; the pose deviation at the initial moment is shown in the following formula:

[0065]

[0066] Step 4.2: By reading the video stream of the tactile image, obtain the motion features of the contact ellipse and the sliding features between two frames, such as... Figure 4 As shown;

[0067] Step 4.3: Due to the inherent errors in ellipse fitting methods, directly determining the ellipse's state vector using the first and last frames of the video can lead to significant errors. To reduce these errors, the ellipse parameters are first filtered before describing the ellipse's state vector, ultimately determining the object's pose change during the sliding process.

[0068]

[0069] Step 5: As Figure 5 As shown, corresponding optimizations and adjustments are made to address the varying degrees of slippage that occur during the initial grasping process, thereby reconstructing the robotic arm's re-grasping pose. Specifically, this includes:

[0070] Step 5.1: Optimize and adjust the initial slippage that occurs during the trial grasping process, specifically as follows:

[0071] Based on the convergent and bounded characteristics of the initial slip, the following two adjustment strategies are proposed:

[0072] (1) The center of the desired contact area should be within the sensor’s sensing range. If the center of the fitted ellipse of the contact detection is outside the sensor’s sensing range when the contact is made, global slippage is very likely to occur. When adjusting, the center of the contact area should be moved along the direction close to the center of the sensor’s sensing range.

[0073] (2) For small initial slip movements, the adjustment strategy only requires adjusting the pose of the end effector in the opposite direction of the object's pose change. Specifically, due to gravity, objects naturally tend to move along the negative z-axis. Scaling is used for changes in the z-axis direction. This invention uses a scaling factor of 0.6, which means capturing the configuration map G(x,y,z,r). x ,r y ,r z ,W) is adjusted to:

[0074] G′(x-dx,y-dy,z-dz×0.6,r x -dr x ,r y -dr y ,r z -dr z ,W).

[0075] Step 5.2: Optimize and adjust the local slippage that occurs during the trial grasping process, specifically as follows:

[0076] Referring to the current object's grasping position map, a search is performed along the x-axis of the current grasping coordinate system towards the object's center of gravity at non-zero positions on the grasping map. The grasping configuration is reconstructed based on the points with the highest confidence near these areas, and a new grasping configuration is obtained through the spatial mapping function f. The search direction is determined by two tactile sensors; if the object's dr... y If the value is positive, then search along the negative x-axis; if dr y If the value is negative, then the search proceeds along the positive x-axis.

[0077] Taking a rod-shaped object in three-dimensional space as an example, such as Figure 6 As shown, force analysis reveals that the critical condition for stable gripping is that the frictional torque of the robotic arm's end finger equals the weight multiplied by the lever arm. The distance of the lever arm can be obtained from the following equation:

[0078]

[0079] Where L is the gravitational lever arm, which can be moved by this distance along the x-axis in the current gripping coordinate to correct the local slippage into a stable gripping action; T is the torque of the friction pair; and θ is the angle with the horizontal direction.

[0080] Since the type, shape, mass, distribution characteristics, and surface contact properties of the object being grasped are all unknown, the above formula needs to be estimated. In this embodiment, the mass of the object being grasped does not exceed 1 kg, so G can be set to 9.8, and the frictional torque T can be obtained from the following formula.

[0081] T=μF∫ A rdr

[0082] Where μ is the coefficient of friction between the finger and the object, which is approximately taken as 0.7 in this embodiment; F is the clamping force, the magnitude of which can be read from the state of the gripper; A is the contact area between the object and the tactile sensor. Because the surface curvature distribution of the object is inconsistent, for the convenience of subsequent processing, this embodiment assumes that the force distribution in the deformation area of ​​the elastic body meets the requirement of uniform distribution. In summary, the magnitude of the frictional torque T can be estimated. Finally, the estimated L is taken as the exploration distance, and it is mapped to the image space to obtain the exploration distance L. i Vector This refers to the iteration direction and step size of the search in the image space. The grasp point is changed at each iteration, and the new grasp coordinates are expressed as follows:

[0083]

[0084] Step 5.3: Optimize and adjust the overall slippage that occurs during the test grasping process, specifically as follows:

[0085] If the current crawl configuration experiences a complete slippage, it indicates that the current crawling posture is unsuitable for the current crawling task. The adjustment strategy should discard the current crawling area, replace the crawling point on the crawl configuration map, discard the peak confidence level on the current crawling location map, find another local peak exceeding the threshold, generate a new coordinate system from the crawling map, and infer the new crawling configuration to obtain the updated crawling configuration. Finally, through multiple crawling quality analyses and local adjustments, the complete slippage problem can be transformed into a local slippage or initial slippage problem. Thus, the crawling posture can be adjusted near the new crawl configuration according to the above strategy until a stable crawling state is achieved.

[0086] The present invention will become more apparent from the following description of the embodiments.

[0087] Performance testing of this invention: The object used in this embodiment is from EGAD. Before the grasping experiment, its collision model was first constructed and scaled to a size suitable for gripping. The uncertainty of the grasping object's properties was simulated by randomly changing its center of gravity position. Ten random center of gravity changes were performed on each grasping object, and ten grasping experiments were conducted on each object with a changed center of gravity. The object model used in the experiment is as follows: Figure 7 As shown. This embodiment is based on Pybullet for simulation experiments, using the Rethink sawyer robot, WSG-50 parallel gripper, and DIGIT tactile sensor. The experimental process is as follows. Figure 8 As shown, the crawling configuration results before and after adjustment are as follows: Figure 9 As shown.

[0088] Example 1:

[0089] To simulate the uncertainty of the pose of the grasped object, this embodiment randomly samples the uncertain space of the grasped object pose as shown in the following formula; to simulate the influence of robot execution error and other uncertain environmental factors, random perturbations are added to the predicted grasping posture during the grasping process.

[0090] S={<x,y,θ>∣x∈[-20,20],y∈[-20,20],θ∈[0,2π]}

[0091] In this embodiment, the mass of the object being grasped is 1 kg, the gripping force of the gripper is 20 N, the maximum number of adjustments is five, and the positional change of the object before and after grasping is obtained by the getBasePositionAndOrientation method of Pybullet. If the change reaches 80% to 100% of the ideal value, it is considered a successful grasp; otherwise, it is considered a failed grasp. If the change reaches 90% to 100% of the ideal value, it is considered a stable grasp.

[0092] The experimental results are shown in the table below:

[0093]

[0094] As shown in the table, with the application of re-grasping posture adjustment, the average grasping success rate increased from 87.6% to 92.2%, of which the stable grasping rate was 88.4%. This result indicates that the reactive re-grasping optimization strategy improves the stability of the grasping process under uncertain object and posture conditions.

[0095] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention in any way. Although the implementation process of the present invention has been described in detail above, those skilled in the art can still modify the technical solutions described in the foregoing examples or make equivalent substitutions for some of the technical features. All modifications and equivalent substitutions made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A robot regrasp optimization method based on tactile primitive slip feature feedback, characterized in that, The method comprises the following steps: Step 1: obtaining tactile modal information by using an optical tactile sensor arranged on a mechanical arm, i.e. perceiving to obtain a tactile element of a contact area; Step 2: performing ellipse parameter fitting based on the obtained tactile modal information; Step 3: constructing a trial grasping coordinate system of the mechanical arm; Step 4: performing trial grasping on the object to cause the object to slip, calculating a tactile perception sliding vector of the object, and inferring a pose change of the object during the slipping; Step 5: performing corresponding optimization adjustment on different degrees of slipping of the object during the trial grasping, and reconstructing a re-grasping pose of the mechanical arm.

2. The robot regrasp optimization method based on tactile primitive slip feature feedback according to claim 1, wherein, The step 1 specifically comprises: judging whether the mechanical arm gripper contacts the object by the change of the tactile image of the tactile sensor based on vision, obtaining the sensor tactile image if yes, and obtaining a no-contact tactile initial image if no; obtaining the area, position and direction of the contact of the object; and obtaining the motion state of the contact part on the sensor surface through continuous video images; The tactile sensor estimates the contact area of the object through single-point and multi-point features, and realizes the perception of the contact area through the extraction and matching of geometric features, so as to obtain a difference image with the initial image; The characteristics of the elastomer material and the surface micro-convex configuration are expressed as elliptical contact areas or circular contact areas on the imaging image. The contact areas are extracted by the tactile perception modal ellipse element, and the contact areas are uniformly expressed in the format of ; wherein, is the center coordinate of the contact ellipse in the tactile sensor image coordinate system, is the semi-major axis length of the contact ellipse, is the semi-minor axis length of the contact ellipse, is the deflection angle of the contact ellipse, with the major axis of the ellipse as the reference, the value is the angle of the major axis deflection in the horizontal direction, and the value range is ; when the shape of the contact area is circular, the major axis length is equal to the minor axis length, i.e. , is any value in.

3. The robot regrasp optimization method based on tactile primitive slip feature feedback according to claim 2, wherein, The step 2 specifically comprises: Step 2.1: performing edge detection and extraction on the contact area after gray scale and morphological processing of the difference image using a Canny edge detection operator; Step 2.2: fitting the edge image detected and extracted using a least square method-based geometric ellipse fitting method.

4. The robot regrasp optimization method based on tactile primitive slip feature feedback according to claim 3, wherein, The step 3 specifically comprises: Step 3.1: with the current time instant grasp configuration map inferred by the grasp detection algorithm To establish the grasp coordinate system with the grasp center coordinates as the origin ; Step 3.2: the working surface of the mechanical arm gripper is equipped with two pairs of facing tactile sensors, then the coordinate system of the two tactile images obtained by the tactile sensors is redefined, taking the right sensor imaging as the reference, and the left sensor imaging as the mirror image of the right sensor imaging, and the left image coordinate system is established with the image center as the center The imaging area center is the center of the coordinate system to establish the right image coordinate system , the left sensor imaging is in accordance with the plane of the coordinate system is a mirror image, and the left image coordinate system is established with the image center as the center The contact areas generated by the left and right images are respectively marked as , , and the corresponding centers are respectively , ; wherein The coordinate value of is the mapping value of the tactile image space to the real contact surface size.

5. The robot regrasp optimization method based on tactile primitive slip feature feedback according to claim 4, characterized in that, The step 4 specifically comprises: Step 4.1: Describe the initial state in the form of deviation; the initial deviation of the object includes the initial position deviation. and initial angle deviation ,in Obtained from the camera's field of view, its size indicates the object's deviation from the grasping coordinate system. of Planar distance; the initial pose deviation is shown in the following formula: ； Step 4.2: obtaining the motion features of the contact ellipse by reading the video stream of the tactile image; Step 4.3: filtering the ellipse parameters, describing the ellipse state vector, and finally determining the pose change of the object during the slipping: 。 6. The robot regrasp optimization method based on tactile primitive slip feature feedback according to claim 5, wherein, The step 5 specifically comprises: Step 5.1: performing optimization adjustment on the initial slipping of the object during the trial grasping; Step 5.2: performing optimization adjustment on the local slipping of the object during the trial grasping; Step 5.3: performing optimization adjustment on the overall slipping of the object during the trial grasping.

7. The robot regrasp optimization method based on tactile primitive slip feature feedback according to claim 6, characterized in that, The step 5.1 specifically comprises: performing the following optimization adjustment according to the characteristics of the initial slipping convergence and boundedness: (1) the center position of the expected contact area is within the sensing range of the sensor, if the center of the fitted ellipse of the contact detection is located outside the sensing range of the sensor when just contacting, then overall slipping occurs, and the center of the contact area is adjusted along the direction close to the center of the sensing range of the sensor; (2) For small amplitude initial slip motion, that is, the fitting ellipse center is located within the sensor's sensing range, only the pose of the gripper needs to be adjusted in the opposite direction of the object pose change. In particular, due to the influence of gravity, the object naturally has a tendency to move in the negative direction of the z-axis. For the change in the z-axis direction, a scaling process is adopted, and the scaling factor is 0.6, that is, the grasp configuration map is adjusted to: 。 8. The robot regrasp optimization method based on tactile primitive slip feature feedback according to claim 6, wherein, The step 5.2, in particular: searching along the direction of the x-axis of the current grasping coordinate system close to the object's gravity center in the non-zero position of the grasping configuration map, reconstructing the grasping configuration according to the point of the highest confidence in the area, obtaining the new grasping configuration through the spatial mapping function f, and the searching direction is determined by the two tactile sensors. If the value of the object's is positive, search in the negative direction of the x-axis. If the value of the object's is negative, search in the positive direction of the x-axis.

9. The robot regrasp optimization method based on tactile primitive slip feature feedback of claim 6, wherein, The step 5.3, in particular: if the current grasping configuration has overall slip, discard the current grasping area, replace the grasping point on the grasping configuration map, discard the peak value of the confidence on the current grasping configuration map, find another The local peak value exceeding the threshold value, then generate a new coordinate from the grasping configuration map, and infer a new grasping configuration to obtain an updated grasping configuration; finally, through multiple grasping quality analysis and local adjustment, the overall slip problem is converted into a local slip or initial slip problem, so as to adjust the grasping posture near the new grasping configuration until the stable grasping state is reached.

Images