Underactuated manipulator self-adaptive grabbing force balance control method for special-shaped parts

Abstract

The invention discloses a self-adaptive grabbing force balance control method of an under-actuated manipulator for special-shaped parts. The method comprises the steps that the appearance and the geometric center position of the special-shaped part are obtained through a visual sensor, the gravity center position of the special-shaped part is calculated, and a mechanical arm makes contact with the special-shaped part from the heavier side; gradually increasing the grabbing force of the manipulator, and continuously monitoring the stress condition of each finger of the manipulator; if the stress of at least one finger of the manipulator exceeds a stress upper limit threshold value or the stress balance value of the whole manipulator exceeds a preset stress balance threshold value, the grabbing force of the manipulator returns to the grabbing force level before the last adjustment; a stress compensation strategy of the manipulator is triggered, and grabbing force is increased to try to grab the special-shaped part again; and if the manipulator triggers the stress compensation strategy and cannot grab the special-shaped part successfully, the manipulator is released, and the whole grabbing process is restarted. According to the method, the grabbing safety and stability are improved, meanwhile, the finger layout is optimized, and the method is suitable for complex-surface and special-shaped objects.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of robot technology and automation control, and particularly relates to an adaptive grasping force balance control method for an underactuated manipulator for irregular-shaped parts. BACKGROUND

[0002] With the development of intelligent manufacturing and automation control, manipulators are widely used in medicine, food, electronics and other industries when handling irregular-shaped parts. However, the existing manipulators usually adopt a fixed finger position or an experience-based grasping strategy when grasping irregular-shaped parts.

[0003] In the related art, the traditional scheme manipulator usually adopts a fixed finger layout or an experience-based grasping strategy. Although it can complete the basic grasping task, when the irregular-shaped part has uneven mass distribution, complex shape or fragile surface, the existing method is prone to cause single finger overload, unbalanced overall grasping force, and thus object tilting, sliding or damage, insufficient grasping stability and safety. SUMMARY

[0004] In order to solve the above technical problems, the present application provides an adaptive grasping force balance control method for an underactuated manipulator for irregular-shaped parts.

[0005] The method obtains three-dimensional point cloud data of the irregular-shaped part through a vision sensor installed on the manipulator, pre-processes to generate a three-dimensional model, and estimates the center of gravity position of the irregular-shaped part based on the model space distribution and surface characteristic information, so that the manipulator can contact the object from the heavy side. During the grasping process, the system gradually increases the grasping force of the manipulator, and monitors the force of each finger in real time. When the force of a single finger exceeds the preset upper limit or the overall force balance value exceeds the threshold value, it automatically reverts to the last grasping force level and triggers the force compensation strategy to ensure the mechanical balance and safety during the grasping process. In addition, the present application uses a gray level co-occurrence matrix to calculate the texture, extracts texture features such as contrast, uniformity, entropy and correlation on the object surface, and determines the safety of the candidate grasping area, effectively avoiding unstable grasping areas. For the newly added finger contact point, the present application calculates the geometric center, evaluates the spatial symmetry and standard deviation of the candidate point and the original finger contact point, and combines the distance from the point to the center of gravity to realize the optimal selection of the grasping point, so as to ensure the uniformity and strong symmetry of the multi-finger grasping layout, and enhance the overall grasping stability. It can significantly improve the grasping safety and stability, and avoid sliding or tilting caused by single finger overload and unbalanced overall force.

[0006] By the above technical solution, the mechanical hand gripping force is gradually increased, and the stress of each finger is monitored in real time, so that damage of the irregular part caused by one-time excessive clamping force can be effectively avoided. If the stress of a finger exceeds the upper limit or the stress balance value exceeds the threshold, the system triggers a rollback mechanism to roll back the gripping force to the level before the last adjustment, thereby protecting fragile, soft or irregular objects from overpressure damage. This method not only reduces the damage probability of the object surface during the gripping process, but also prolongs the service life of the driving components of the mechanical hand itself. When the stress is unbalanced and causes unstable gripping, the method can trigger a stress compensation priority strategy to preferentially increase the fingers according to the direction of the weight deviation, or use a symmetric compensation strategy to form a balanced support on the side of the object weight deviation and the opposite side. This can significantly reduce the risk of overturning caused by the deviation of the center of gravity, and ensure that the object center of gravity projection falls within the support area to achieve balanced distribution of the overall stress. Thus, the stability of the irregular part during lifting and carrying is improved, and the failure rate caused by eccentricity is reduced. Combined with historical gripping data and the physical characteristics of the irregular part (including size, shape, friction coefficient and center of gravity position distribution characteristics), the method of the present application can dynamically determine a reasonable stress balance threshold. For known categories of objects, the system can directly call the experience threshold stored in the database; for new types of objects, an initial estimated value is given based on a geometric and center of gravity distribution calculation model. This adaptive threshold setting method avoids misjudgment due to natural eccentricity or complex shape of the object, and improves the generalization ability and adaptability of the gripping strategy. In the case of multiple attempts that still fail to grip, the method sets a safe release and re-grasping mechanism to avoid the mechanical hand maintaining high force output or unstable support for a long time, preventing the object from being damaged by continuous stress. Through this closed-loop process of "grasping - monitoring - rollback - compensation - release", the system can ensure the stability of the gripping while shortening the number of invalid attempts, thereby reducing the overall energy consumption and improving the task efficiency. The method automatically evaluates the candidate gripping points and optimizes the finger layout to improve the gripping success rate and operation efficiency; it takes into account the center of gravity position and spatial symmetry to achieve adaptive gripping of mechanical balance; it is suitable for irregular, uneven mass distribution or complex surface irregular part gripping, and has wide applicability; and it reduces the dependence on operation experience. BRIEF DESCRIPTION OF DRAWINGS

[0007] Figure 1 is a flow chart of an adaptive gripping force balance control method for an underactuated mechanical hand of an irregular part provided by an embodiment of the present application; Figure 2 is a structural block diagram of an adaptive gripping force balance control method for an underactuated mechanical hand of an irregular part provided by an embodiment of the present application; Figure 3 is a flow chart of determining the position of the added finger provided by an embodiment of the present application.

[0008] Figure 4is a safety area determination flowchart provided by an embodiment of the present application DETAILED DESCRIPTION

[0009] The terms used in the following embodiments of the present application are only for the purpose of describing the specific embodiments and are not intended to be limiting of the present application. As used in the specification and the appended claims of the present application, the singular forms "a," "an" and "the" are intended to include both singular and plural forms, unless the context clearly indicates otherwise. It will be further understood that the terms "and / or," as used in the present application, refers to any or all possible combinations of one or more of the associated listed items.

[0010] Hereinafter, the terms "first" and "second" are only for the purpose of description, and cannot be understood as implying or suggesting relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first" and "second" can explicitly or implicitly include one or more of the features, and in the description of the embodiments of the present application, the meaning of "a plurality of" is two or more, unless otherwise specified.

[0011] The following will be described in conjunction with the accompanying Figures 1-4 The embodiments of the present application are described.

[0012] The present application provides an adaptive grasping force balance control method for an underactuated manipulator for irregular-shaped parts, Figure 1 is a flowchart of an adaptive grasping force balance control method for an underactuated manipulator for irregular-shaped parts provided by an embodiment of the present application, the method comprising: Step S101, acquiring the appearance and geometric center position of the irregular-shaped part through a visual sensor, calculating the center of gravity position of the irregular-shaped part, and making the manipulator contact the irregular-shaped part from the heavy side; Step S102, gradually increasing the grasping force of the manipulator, while continuously monitoring the force on each finger of the manipulator; Step S103, if the force on at least one finger of the manipulator exceeds the upper limit threshold of the force or the force balance value of the whole manipulator exceeds the preset force balance threshold, the grasping force of the manipulator is rolled back to the grasping force level before the last adjustment; Step S104, triggering the force compensation strategy of the manipulator and increasing the grasping force to try to grasp the irregular-shaped part again; Step S105, if the manipulator fails to successfully grasp the irregular-shaped part after triggering the force compensation strategy, the manipulator is released and the whole grasping process is started again.

[0013] The execution subject of the above steps can be a system, or a platform, or a backend process, etc., but is not limited thereto.

[0014] In an optional embodiment, the appearance and geometric center position of the irregular part are acquired by a visual sensor, and the center of gravity of the irregular part is calculated so that the robot contacts the irregular part from the side with the greater weight, specifically comprising: acquiring multi-angle point cloud data of the irregular part by a visual sensor installed on the robot, and generating a three-dimensional model by processing the point cloud data, and estimating the center of gravity of the irregular part based on the spatial distribution and surface characteristic information of the three-dimensional model.

[0015] In the above embodiment, the visual sensor is preferably a depth camera installed on the robot, which can continuously shoot the irregular part from multiple angles to obtain three-dimensional point cloud data containing the contour and depth information of the object In the specific implementation process, first, the point cloud is denoised to remove environmental noise and invalid background points, and the original point cloud is voxel down-sampled to reduce the data volume and improve the subsequent operation efficiency.

[0016] After the point cloud preprocessing is completed, the three-dimensional contour model of the object is generated by surface reconstruction or geometric fitting method. Then, the system approximates the quality distribution of the three-dimensional model by a geometric center estimation method. Specifically, the geometric center of the whole object can be calculated based on the point cloud density distribution or voxel division result, and the surface color, texture, roughness, and light reflection characteristic information of the irregular part are collected by the RGB camera and the depth camera in combination with the local thickness and material characteristics of the object, and the material category is roughly guessed by matching with the database of known materials, and the center of gravity of the object is approximately estimated.

[0017] After the center of gravity information is calculated, it is input to the grabbing path planning module. The path planning module first determines the direction of the weight deviation according to the deviation between the center of gravity and the geometric center. Then, the approaching path of the robot is adjusted according to the weight deviation direction, so that the robot contacts the irregular part from the side with the greater weight. During the execution, the control system limits the movement speed of the robot, so that it completes the initial contact at a lower speed, avoiding the instantaneous sliding or dumping of the object due to the weight deviation.

[0018] By using the depth camera and point cloud processing technology, the shape and center of gravity information of the irregular part can be quickly and accurately acquired; by guiding the contact strategy of the robot through the center of gravity direction, the application of the clamping force is more in line with the physical characteristics of the object. This method not only reduces the dependence on high-precision force sensors, but also reduces the energy consumption and trial-and-error times in the grabbing process, and improves the overall stability and success rate of the system.

[0019] In an optional embodiment, if the force on at least one finger of the robot exceeds the upper limit threshold or the force balance value of the entire robot exceeds the preset force balance threshold, the gripping force of the robot is returned to the gripping force level before the last adjustment, specifically including: the upper limit threshold of the force is the inherent characteristic of each finger of the robot. The force balance value is obtained by measuring the real-time force of each finger of the robot, calculating the average value of the forces of all fingers, calculating the average value of the deviation squares of the forces of each finger from the average value, and finally taking the square root to obtain the force balance value.

[0020] In the above embodiment, the return process of the gripping force can be automatically completed by the control module of the robot. Specifically, each finger is provided with a fixed upper limit threshold of the force, which is determined according to the finger structure, material strength and driving capacity in the design stage of the robot, to ensure that the finger will not be damaged or fail in function due to overload during gripping. During real-time gripping, the controller continuously monitors the force sensor data of each finger, and calculates the force currently borne by each finger. If it is found that the force on at least one finger exceeds the corresponding fixed upper limit threshold of the force, the system will immediately trigger the gripping force return operation. The return operation will reduce the gripping force of the robot to the last safe gripping force level, to prevent the finger from being damaged by excessive mechanical stress. To ensure smooth return process and enhance the refinement of control, the return action adopts gradual adjustment, and a plurality of decreasing steps are set to gradually reduce the gripping force, rather than suddenly reducing it, to ensure smooth return process.

[0021] By setting the fixed upper limit threshold of the force and monitoring it during real-time gripping, each finger can work within a safe range, thereby protecting the structural integrity of the robot and improving the safety and reliability of the gripping action.

[0022] During gripping of irregular-shaped objects, the robot needs to pay attention not only to the upper limit of the force on a single finger, but also to the force balance among the fingers. The force balance value can be obtained by calculating the standard deviation or variance of the forces of all fingers, reflecting the difference degree of the forces applied by the fingers. If some fingers bear too much force while other fingers bear less force, it means that the gripping force is not evenly distributed, and the object may tilt or overturn during gripping. In order to prevent the irregular-shaped object from overturning, a preset force balance threshold is set, and when the real-time calculated force balance value exceeds the threshold, it means that there is potential instability in gripping. At this time, the robot will immediately perform the gripping force return operation to restore the overall gripping force to the safe level before the last adjustment. By monitoring the force balance value and setting the threshold, the risk of overturning of the irregular-shaped object during gripping can be effectively avoided, the safety and reliability of the gripping operation are improved, and the irregular-shaped object itself is protected from damage.

[0023] The force balance value calculation method is: first, calculate the arithmetic mean of the force of all fingers, then calculate the difference between the force of each finger and the average value in turn, and square sum the difference, finally take the average of the square sum and take the square root, so as to obtain the numerical value of the force balance value. If the force balance value exceeds the set force balance threshold, the controller determines that the current grasping state is not balanced, which will cause the grasped irregular part to overturn.

[0024] Through the implementation of the above scheme, the manipulator can actively avoid the potential risks caused by excessive force on a single finger or excessive overall force balance value during object grasping, thereby effectively avoiding object damage, manipulator structure wear or grasping failure. In addition, the grasping force rollback mechanism enables the system to have "self-correction" capability, thereby not only improving the operation safety of the manipulator when handling irregular, center-of-gravity shifted objects, but also significantly enhancing the stability and reliability of the system.

[0025] In an optional embodiment, if the force on at least one finger of the manipulator exceeds the upper force threshold or the overall force balance value of the manipulator exceeds the preset force balance threshold, the grasping force of the manipulator is rolled back to the grasping force level before the last adjustment, which specifically includes: the preset force balance threshold of the manipulator is determined according to the physical characteristics of the irregular part, the center-of-gravity position distribution characteristics and the historical grasping data, the physical characteristics include the size, shape and friction coefficient of the irregular part, the center-of-gravity position distribution characteristics are the distribution state of the mass of the irregular part in space, and the historical grasping data are the state information recorded by the manipulator when grasping the same or similar type of irregular part in the past.

[0026] In the above embodiment, the determination method of the preset force balance threshold can combine the physical characteristics of the irregular part, the historical grasping data and the center-of-gravity position distribution characteristics. Specifically, in the system initialization stage, the control module first performs appearance recognition and parameter extraction on the target irregular part through the vision sensor and the depth camera, obtains key physical characteristics including size, geometric shape, surface roughness and friction coefficient with the material of the manipulator finger end, and calculates the center-of-gravity position of the irregular part and its distribution state in space. Then, combined with the historical grasping records and the center-of-gravity position distribution characteristics of this type of irregular part, the most suitable force balance threshold for this type of object is selected, thereby improving the stability and success rate of grasping.

[0027] In its implementation, the controller maintains a mapping table between object characteristics and thresholds in the database. For any irregular object to be captured, it first categorizes it using an object classification model, then retrieves the corresponding threshold parameter from the mapping table. If it's a new object type appearing for the first time and there's no suitable category in the database, the system provides an initial estimated threshold based on its physical characteristics and center-of-gravity distribution. The system then searches the database for the historical capture record most similar to the current object in terms of characteristics and center-of-gravity distribution, using the corresponding capture threshold parameter as a reference value.

[0028] Surface roughness is obtained indirectly through a visual sensor or a depth camera. The texture information collected by the visual sensor is used to analyze the grayscale changes and light reflection characteristics of the object's surface to distinguish between smooth and rough surfaces. The three-dimensional point cloud data obtained by the depth camera is used to roughly determine the roughness level of the surface by the magnitude of the residual of the local plane fitting of the point cloud.

[0029] By incorporating both physical characteristics and historical data into the force balance threshold setting process, personalized adaptive adjustment for different objects is achieved. This means that for objects with low friction coefficients and a tendency to slip, the system automatically sets a stricter force balance threshold to ensure a more uniform force distribution; while for objects with high friction coefficients or regular shapes, the system allows for greater tolerance to force differences, improving gripping efficiency.

[0030] In an optional embodiment, the priority strategy for force compensation of the robotic arm specifically includes: the force compensation strategy of the robotic arm: collecting a three-dimensional point set on the surface of the irregular part through sensors; filtering the three-dimensional point set for safe areas and removing invalid points to obtain a candidate point set; evaluating the symmetry of the spatial distribution relationship between each point in the candidate point set and the existing finger contact point; and determining the optimal position of the new finger contact point.

[0031] Figure 3 This is a flowchart of determining the position of adding a finger provided in an embodiment of this application. In the above embodiment, the force compensation strategy is to add the finger to a suitable position. The specific implementation method is as follows: The first step involves obtaining discrete 3D coordinates of the object's surface using a sensor, forming a 3D point set. A depth camera is preferred because it provides depth information (i.e., the distance from the camera to the object's surface) for every pixel within the entire field of view in a single acquisition. The pixel coordinates (u, v) are then combined with the depth value D, and the result is obtained using the camera's intrinsic parameter matrix (focal length). Main point This allows you to map pixels to three-dimensional coordinates (x, y, z): ; Second, determine whether each point in the three-dimensional point set is in the safety area. Traverse all points in the three-dimensional point set, and determine whether the point is in the safety area according to the coordinates of the point Determine whether the point is in the safety area: find the window containing the point, and check whether the window is determined as the safety area. All points in the safety area are regarded as candidate points.

[0032] Figure 4 The safety area determination flowchart provided by the embodiment of the present application is shown in FIG. 1. A specific implementation of determining the safety area is as follows: an object surface image is collected by using a visual sensor, and the visual sensor is preferably a high-resolution RGB camera, which has the advantage of providing clear image information and making material analysis more accurate. Image graying: first, the color image is converted into a gray image, and a pair of adjacent pixels is selected: for each pixel point, a pair of horizontally adjacent pixels is selected. Matrix construction: the frequency of each pair of gray values in the image is counted to obtain a two-dimensional matrix. Each element of the matrix represents the number of occurrences of a specific pair of gray values in the image. Each number in the co-occurrence matrix is divided by the sum of all numbers in the matrix in order to perform feature calculation and subsequent analysis.

[0033] Texture analysis is performed using the gray level co-occurrence matrix, which can sensitively reflect subtle texture changes in the image and has good performance for judging whether the surface is smooth or rough, soft or hard. The gray level co-occurrence matrix mainly depends on the gray value relationship of adjacent pixel pairs. Assuming that there are p gray levels in the image, the size of the gray level co-occurrence matrix is p x p, and each element G(i, j) represents the frequency of occurrence of a pixel pair with gray values i and j.

[0034] Co-occurrence matrix calculation: for a gray image, a window (such as 3x3 pixels or 5x5 pixels) is selected, and a matrix is constructed by calculating the frequency of adjacent pixel gray pairs in the window.

[0035] Texture feature extraction: once the gray level co-occurrence matrix is obtained, multiple texture features can be calculated to describe the surface properties (such as roughness, smoothness) of the object, including: Contrast: measures the degree of difference in pixel gray levels in the image. Higher contrast indicates greater texture difference in the image, corresponding to a rougher surface.

[0036] ; Homogeneity: measures the smoothness of the gray level pairs in the image. Areas with high homogeneity indicate a smoother surface.

[0037] ; Entropy: the complexity of the image gray distribution, the higher the entropy value, the more complex the texture of the image surface.

[0038] ; Correlation: measures the relationship and gray consistency between pixels, the higher the correlation, the more consistent the image texture.

[0039] ; G(i,j) is an element in the gray level co-occurrence matrix, i and j are two gray levels, and is the mean value of the gray distribution.

[0040] A specific implementation method for judging whether it is a safe area using the above texture feature values: divide the value of each feature into different intervals, and set a suitable score for each interval, then add up the scores of the four features to finally judge whether it is a safe area.

[0041] Set intervals and scores for each feature: Contrast: 0-30 (low), 30-70 (medium), 70-100 (high), corresponding scores are 1, 2, and 3 respectively.

[0042] Uniformity: 0-0.5 (low), 0.5-0.8 (medium), 0.8-1.0 (high), corresponding scores are 3, 2, and 1 respectively.

[0043] Entropy: 0-3 (low), 3-5 (medium), 5-7 (high), corresponding scores are 1, 2, and 3 respectively.

[0044] Correlation: 0-0.5 (low), 0.5-0.8 (medium), 0.8-1.0 (high), corresponding scores are 3, 2, and 1 respectively.

[0045] Add up the scores of each feature to get the total score: total score = contrast score + uniformity score + entropy score + correlation score. If the total score is greater than or equal to 8, it is considered suitable for grabbing and is a safe area; if the total score is less than 8, it is not suitable for grabbing and is not a safe area. The advantage of this scheme is that threshold division can simplify the decision-making process and be efficient.

[0046] Step 3: Calculate the coordinates of the center of gravity according to the three-dimensional point set: the center of gravity is the average value of all point coordinates in the three-dimensional point set, and the coordinates of each point in the three-dimensional point set are , then the coordinates of the center of gravity are which can be calculated by the following formula: ; where N is the number of points in the point set, is the coordinate of the i-th point in the point cloud.

[0047] Once the coordinates of the barycenter are calculated , the Euclidean distance formula can be used to calculate the distance of each point in all candidate points to the barycenter : ; Fourth step, evaluate the symmetry of the spatial position relationship formed by all candidate point positions and the original finger contact positions. The purpose of symmetry evaluation is to consider not only the position of a single finger but also the layout of all fingers in space when the robot is grasping an irregular object, so as to form a stable support structure and reduce the risk of overturning or sliding.

[0048] The specific implementation of symmetry evaluation: assuming that the robot originally has three fingers in contact with the surface of the object, their contact points are , , , and the contact point coordinates of the planned added finger are .

[0049] Calculate the geometric center C of the four contact points: ; Calculate the distance from each force point to the geometric center: ; Calculate the standard deviation of the distance ; The more uniform the distance from the geometric center (the smaller the standard deviation), the more "symmetric" the distribution of the contact points, and the more stable the grasp. If the addition of a finger results in a decrease in the standard deviation, it means that the new contact position makes the distribution more uniform and the symmetry stronger.

[0050] Fifth step, combine the distance of the point to the barycenter ( ) and the symmetry (reflected by the standard deviation ) to find the optimal solution, i.e., the contact point of the finger to be added. One implementation method is to use the Pareto front method (non-dominated sorting) algorithm.

[0051] Prepare candidate points: the coordinates of each point of the above candidate points are , and each point is represented as a two-dimensional target point:

[0052] Determine the non-dominated relationship: for any two points A and B,

[0053] if both conditions are met: and

[0054] And at least one strictly smaller: or

[0055] Then A is said to dominate B; otherwise, A and B do not dominate each other. Calculate the number of times each candidate point is dominated. For example, a candidate point C initially Iterate through all the candidate points mentioned above. If a candidate point dominates point C, then point C's... Add 1, all The candidate points belong to the first-level Pareto front, and the final point is selected from the first-level Pareto front.

[0056] If there is only one candidate point in the first-layer Pareto front, then that point is the optimal solution and is used as the final point. If there are multiple candidate points in the first-layer Pareto front, then the distance from the selected point to the centroid ( The smallest candidate point is taken as the final point.

[0057] The aforementioned final point is the most suitable position to place the added finger.

[0058] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0059] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and inventive constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0060] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0061] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0062] Finally: the above only for the preferred embodiments of the present application, and not for limiting the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application, should be included in the scope of protection of the present application.

Claims

1. An adaptive grasp force balancing control method for an underactuated manipulator of a foreign object, characterized in that, The method comprises the following steps: Obtain the appearance and geometric center position of the irregular part through the visual sensor, calculate the center of gravity position of the irregular part, and make the robot hand contact the irregular part from the side with the weight bias; Gradually increase the gripping force of the robot hand while continuously monitoring the force on each finger of the robot hand; If the force on at least one finger of the robot hand exceeds the upper limit threshold of the force or the force balance value of the entire robot hand exceeds the preset force balance threshold, the gripping force of the robot hand is returned to the gripping force level before the last adjustment; Trigger the force compensation strategy of the robot hand and increase the gripping force to try to grab the irregular part again; If the robot hand fails to successfully grab the irregular part after triggering the force compensation strategy, the robot hand is released and the entire grabbing process is restarted. 2.The adaptive grasp force balancing control method for the underactuated manipulator of a foreign object according to claim 1, wherein, The method comprises the following steps: Obtain the appearance and geometric center position of the irregular part through the visual sensor, calculate the center of gravity position of the irregular part, and make the robot hand contact the irregular part from the side with the weight bias; 3.The adaptive grasp force balancing control method for the underactuated manipulator of a foreign object according to claim 1, wherein, Obtain multi-angle point cloud data of the irregular part through the visual sensor installed on the robot hand, generate a three-dimensional model through processing the point cloud data, estimate the center of gravity position of the irregular part based on the spatial distribution and surface characteristic information of the three-dimensional model. If the force on at least one finger of the robot hand exceeds the upper limit threshold of the force or the force balance value of the entire robot hand exceeds the preset force balance threshold, the gripping force of the robot hand is returned to the gripping force level before the last adjustment, which comprises the following steps:

4. The adaptive grasp force balancing control method for the underactuated manipulator of foreign matter according to claim 1, wherein, The upper limit threshold of the force is the inherent characteristic of each finger of the robot hand, the force balance value is calculated by measuring the real-time force of each finger of the robot hand, calculating the average value of the forces of all fingers, calculating the average value of the deviation squares of each finger force from the average value, and finally taking the square root to obtain the force balance value. If the force on at least one finger of the robot hand exceeds the upper limit threshold of the force or the force balance value of the entire robot hand exceeds the preset force balance threshold, the gripping force of the robot hand is returned to the gripping force level before the last adjustment, which comprises the following steps:

5. The adaptive grasp force balancing control method for the underactuated manipulator of foreign matter according to claim 1, wherein, The preset force balance threshold of the robot hand is determined according to the physical characteristics, center of gravity position distribution characteristics and historical grabbing data of the irregular part, the physical characteristics include the size, shape and friction coefficient of the irregular part, the center of gravity position distribution characteristics are the distribution state of the mass of the irregular part in space, and the historical grabbing data are the state information recorded when the robot hand grabs the same or similar type of irregular part in the past. The robot hand force compensation priority strategy comprises the following steps:

6. The adaptive grasp force balancing control method for the underactuated manipulator of foreign matter according to claim 5, wherein, The robot hand force compensation strategy: collecting a three-dimensional point set on the surface of the irregular part through a sensor; performing safety area screening on the three-dimensional point set to remove invalid points and obtain a candidate point set; evaluating the symmetry of the spatial distribution relationship between each point in the candidate point set and the existing finger contact point; and determining the optimal new finger contact point position. The safety area screening on the three-dimensional point set to remove invalid points and obtain a candidate point set comprises the following steps:

7. The adaptive grasp force balancing control method for the underactuated manipulator of foreign matter according to claim 5, wherein, Safety area determination method: using a gray level co-occurrence matrix texture calculation method to extract contrast, uniformity, entropy and correlation texture features, scoring and accumulating the correlation texture features according to a preset interval to obtain a total score, and determining the corresponding area as a safety area when the total score is greater than or equal to a preset score threshold. The symmetry evaluation comprises the following steps: The geometric center of the candidate point and the existing finger contact point is obtained, the distances from the candidate point and the existing finger contact point to the geometric center are calculated, and the symmetry is evaluated based on the standard deviation of the distances. 8.The adaptive grasp force balancing control method for the underactuated gripper of a foreign object according to claim 5, wherein, The optimal new finger contact point position is determined, specifically including: Based on the distance from the candidate point to the center of gravity, the symmetry evaluation of the candidate point, the stability of the candidate point is comprehensively evaluated, and the optimal new finger contact point position is determined.

Images