Manipulator grabbing control system based on visual identification

Abstract

The invention discloses a manipulator grabbing control system based on visual identification, and the system comprises a visual identification module which is used for collecting the three-dimensional form data and surface feature information of a target object, and eliminating the noise interference in an original image through preprocessing; the multi-modal fusion processor is used for receiving the three-dimensional form data and the surface feature information output by the visual recognition module, executing point cloud structured conversion on the three-dimensional form data, extracting geometric features of an object through a neural network, and meanwhile, performing weight distribution on the surface feature information by adopting a feature attention mechanism to complete multi-feature fusion processing; according to the multi-axis cooperative manipulator provided with the end executor, the end executor integrates a touch sensing assembly used for touch sensing and an adsorption execution device used for assisting grabbing. By integrating the technologies of visual identification, multi-modal fusion, intelligent optimization, dynamic feedback and the like, the sensing precision, the environmental adaptability and the grabbing reliability of the system to object characteristics are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of control systems, in particular to a mechanical hand grabbing control system based on visual recognition. BACKGROUND

[0002] Mechanical hand grabbing control refers to the technical process of realizing stable and efficient grabbing of target objects by accurately regulating key parameters such as the motion trajectory of the mechanical hand, grabbing force, and contact posture. The core goal is to ensure that the mechanical hand can accurately plan the grabbing path and adjust the clamping force according to the characteristics of the object's shape, material, and other characteristics, avoiding object slipping, damage, or grabbing failure during the grabbing process. It is the core link of the mechanical hand to complete the operation task, and the mechanical hand grabbing control system is a complete technical system to realize the grabbing control function, achieving accurate and stable control of grabbing actions in different objects and environments.

[0003] In actual application, the existing technology has the problems of single function and insufficient adaptability, resulting in poor effect in complex scenes and difficulty in meeting diversified and high-precision grabbing or protection requirements. Therefore, a mechanical hand grabbing control system based on visual recognition is proposed. SUMMARY

[0004] The present application solves the above technical problems by the following technical solutions. The present application comprises:

[0005] The visual recognition module is used to collect three-dimensional shape data and surface feature information of the target object, and to eliminate noise interference in the original image through preprocessing;

[0006] The multi-modal fusion processor receives the three-dimensional shape data and surface feature information output by the visual recognition module, performs point cloud structure conversion on the three-dimensional shape data, extracts object geometric features through a neural network, and simultaneously performs weight distribution on the surface feature information using a feature attention mechanism to complete multi-feature fusion processing;

[0007] The multi-axis collaborative mechanical hand is provided with an end effector integrated with a tactile perception component for contact perception and an adsorption execution device for assisting grabbing;

[0008] The intelligent control unit constructs a grabbing evaluation model based on the fused feature information, generates multiple sets of initial grabbing pose candidate schemes and configures a reinforcement learning framework, and determines the optimal grabbing path through an iterative optimization algorithm;

[0009] The dynamic feedback regulation closed loop collects the tactile perception data of the end effector and the working parameters of the adsorption execution device in real time, and dynamically corrects the grabbing force and posture parameters.

[0010] Further, the visual recognition module comprises:

[0011] a structured light projection assembly configured to project a feature recognition pattern to a target object;

[0012] a binocular image acquisition assembly configured to acquire an object image containing the feature pattern through synchronous shooting by double view angles;

[0013] an ambient light adaptation assembly configured to automatically adjust a light compensation mode according to an ambient light intensity;

[0014] an image preprocessing unit configured to sequentially perform distortion correction, size normalization and gray scale equalization processing on the double view angle images, and to calculate and generate object three-dimensional morphological data through a stereo matching algorithm.

[0015] Further, the multi-modal fusion processor adopts a hierarchical processing architecture, which includes:

[0016] a geometric feature extraction layer configured to perform point cloud gridding processing on the three-dimensional morphological data, and to extract object contour, curvature and spatial distribution features through a multi-layer perception neural network;

[0017] a surface feature analysis layer configured to perform texture segmentation and feature point detection on the surface feature information, and to identify object surface material attribute related features;

[0018] a cross-modal fusion layer configured to convert the geometric features and the surface features to the same feature space through feature mapping, and to realize feature enhancement fusion by using an attention weight matrix.

[0019] Further, the end effector of the multi-axis cooperative manipulator includes a plurality of independently driven flexible finger structures, and each finger realizes multi-degree-of-freedom movement through a joint linkage mechanism.

[0020] The tactile perception assembly adopts a distributed sensing array for detecting the pressure distribution state of the grasping contact area.

[0021] The adsorption execution device realizes negative pressure generation and release through an airflow control assembly, and forms a composite grasping mode in cooperation with the finger grasping.

[0022] Further, the iterative optimization process of the intelligent control unit is executed according to the following process:

[0023] First, a state evaluation space is constructed, which contains physical attribute parameters of the target object, geometric constraint conditions and environmental interference factors, to provide a basic framework for grasping scheme evaluation;

[0024] Then, based on the fusion feature information output by the multi-modal fusion processor, the constraint parameters of the state evaluation space are combined to call an initial grasping pose candidate scheme set, and grasping stability, operation convenience and energy consumption indicators are defined as core evaluation parameters;

[0025] Finally, the candidate poses are simulated for grasping tests through a reinforcement learning framework, the evaluation results are quantitatively evaluated by using preset evaluation parameters, and the grasping path parameters are iteratively optimized by using a greedy algorithm until the optimal grasping path is determined.

[0026] Further, the working process of the dynamic feedback adjustment closed loop is:

[0027] In the state perception stage, the pressure distribution data of the tactile perception component and the working parameters of the adsorption execution device are collected in real time.

[0028] In the deviation analysis stage, the actual perception data is compared with the preset grasping model parameters, and the posture and force deviation values are calculated.

[0029] In the control correction stage, the joint driving parameters of the multi-axis collaborative manipulator and the force parameters of the adsorption execution device are adjusted based on the deviation analysis results through a PID control algorithm.

[0030] Further, it further includes an environmental adaptability module, which includes:

[0031] An environmental perception component is used to detect the obstacle distribution and lighting conditions in the working space.

[0032] A path planning unit generates a collision-free motion path through a space search algorithm based on the path optimization logic of the intelligent control unit.

[0033] A light self-adaptive unit is used to dynamically adjust the exposure parameters and light supplement intensity of the image acquisition component in the visual recognition module according to the environmental light detection results.

[0034] Further, it further includes an intelligent learning and interaction module, which includes:

[0035] A digital twin mapping unit is used to construct a virtual simulation model of the physical system in real time, and to simulate the motion state and force condition of the multi-axis collaborative manipulator synchronously.

[0036] A human-computer interaction component is used to realize operation instruction input and state feedback through gesture recognition and visual guidance.

[0037] An experience learning database stores historical grasping case data, and periodically optimizes the grasping strategy model of the intelligent control unit through clustering analysis and pattern mining based on the iterative optimization results.

[0038] Compared with the prior art, the mechanical hand grabbing control system based on visual recognition has the following advantages: the visual recognition module generates three-dimensional shape data through structured light projection, binocular collection and preprocessing (distortion correction, gray balance, etc.), combines the layered architecture of the multi-modal fusion processor (geometric feature extraction, surface feature analysis and cross-modal fusion), and realizes enhanced fusion of geometric and surface features by using the attention mechanism, thereby providing comprehensive data support for accurate grabbing.

[0039] The intelligent control unit constructs an evaluation model based on the fused features, simulates the test of multiple candidate poses through a reinforcement learning framework, takes the grabbing stability, operation convenience and energy consumption as the core indicators, iteratively optimizes the path by using a greedy algorithm, reduces the grabbing deviation, and realizes the optimal path planning with high efficiency and low energy consumption.

[0040] The dynamic feedback regulation closed loop collects the tactile pressure distribution of the end effector and the adsorption device parameters in real time, compares the actual data with the preset model through deviation analysis, dynamically adjusts the joint driving parameters and adsorption strength based on the PID control algorithm, effectively responds to the differences in physical properties of objects and grabbing disturbances, and avoids object damage caused by slipping or excessive clamping.

[0041] The environmental adaptability module detects obstacles and light through the environment perception component, the path planning unit generates a collision-free motion path, and the light self-adaptive unit dynamically adjusts the image acquisition exposure and light intensity, ensuring that the system operates stably in a complex light and dynamic interference workspace, and reducing the influence of environmental factors on the grabbing effect.

[0042] The end effector adopts a multi-degree-of-freedom flexible finger structure and an adsorption execution device to form a composite grabbing mode, which can adapt to target objects of different shapes and materials, significantly expanding the application range of the system.

[0043] The intelligent learning and interaction module constructs a virtual simulation model through digital twin mapping, realizes instruction input and feedback through human-computer interaction components, stores historical cases in the experience learning database, periodically optimizes the grabbing strategy model through clustering analysis and pattern mining, so that the system has self-learning ability and continuously improves performance to adapt to diversified task requirements. BRIEF DESCRIPTION OF DRAWINGS

[0044] Figure 1 is a system block diagram of the present application. DETAILED DESCRIPTION

[0045] The embodiments of the present application will be described in detail below. The embodiments are implemented on the premise of the technical solution of the present application, and detailed implementation modes and specific operation processes are given, but the protection scope of the present application is not limited to the following embodiments.

[0046] As Figure 1As shown, the embodiment provides a technical solution: a mechanical hand grabbing control system based on visual recognition, comprising:

[0047] A visual recognition module is configured to collect three-dimensional shape data and surface feature information of a target object, and eliminate noise interference in the original image through preprocessing;

[0048] A multi-modal fusion processor is configured to receive the three-dimensional shape data and surface feature information output by the visual recognition module, perform point cloud structured conversion on the three-dimensional shape data, extract object geometric features through a neural network, and complete multi-feature fusion processing by using a feature attention mechanism to assign weights to the surface feature information;

[0049] A multi-axis collaborative manipulator with an end effector is provided, and the end effector integrates a tactile perception component for contact perception and a suction execution device for assisting grabbing;

[0050] An intelligent control unit is configured to construct a grabbing evaluation model based on the fused feature information, generate multiple sets of initial grabbing pose candidate schemes and configure a reinforcement learning framework, and determine an optimal grabbing path through an iterative optimization algorithm;

[0051] A dynamic feedback regulation closed loop is configured to collect tactile perception data of the end effector and working parameters of the suction execution device in real time, and dynamically correct grabbing force and posture parameters.

[0052] The visual recognition module comprises:

[0053] A structured light projection component is configured to project a feature recognition pattern to a target object;

[0054] A binocular image acquisition component is configured to obtain an object image containing a feature pattern through synchronous shooting from two visual angles;

[0055] An ambient light adaptation component is configured to automatically adjust a light compensation mode according to ambient light intensity;

[0056] An image preprocessing unit is configured to sequentially perform distortion correction, size normalization and gray scale equalization processing on the binocular images, and calculate and generate three-dimensional shape data of the object through a stereo matching algorithm;

[0057] The feature recognition pattern is projected to the target object by the structured light projection assembly, and the object image containing rich parallax information is obtained through the synchronous shooting of the binocular image acquisition assembly, thereby providing accurate feature reference for the three-dimensional shape data generated by the stereo matching algorithm, and effectively improving the spatial positioning accuracy of the three-dimensional shape data; the ambient light adaptation assembly can automatically adjust the light compensation mode according to the ambient light intensity, thereby avoiding the problems of image blurring, low contrast and the like caused by insufficient, excessive or uneven light, and ensuring that clear and effective object images can be acquired under complex lighting conditions, thereby ensuring the stability of data acquisition; the image preprocessing unit sequentially performs distortion correction, size normalization and gray equalization on the binocular images, thereby effectively eliminating noise interference and distortion influence in the original images, providing high-quality image input for the stereo matching algorithm, and further improving the reliability of the three-dimensional shape data.

[0058] The multi-modal fusion processor adopts a hierarchical processing architecture, and the hierarchical processing architecture comprises:

[0059] The geometric feature extraction layer performs point cloud grid processing on the three-dimensional shape data, and extracts object contour, curvature and spatial distribution features through a multi-layer perception neural network;

[0060] The surface feature analysis layer performs texture segmentation and feature point detection on the surface feature information, and identifies object surface material attribute related features;

[0061] The cross-modal fusion layer converts the geometric features and the surface features to the same feature space through feature mapping, and realizes feature enhancement fusion through an attention weight matrix;

[0062] The geometric feature extraction layer can extract core geometric features such as object contour, curvature and spatial distribution through point cloud grid processing and a multi-layer perception neural network, thereby accurately depicting the three-dimensional structure of the object and providing geometric basis for determining the grasping position; the surface feature analysis layer can effectively identify object surface material attribute related features through texture segmentation and feature point detection, thereby providing surface property support for subsequent matching grasping force; the cross-modal fusion layer converts the geometric features and the surface features to the same feature space through feature mapping, eliminates the dimensional difference between different modal features, and then performs weight distribution and enhancement fusion on the key features through an attention weight matrix, thereby avoiding feature redundancy or key information loss, making the fused feature information more comprehensive and focused, and providing high-quality feature input for the intelligent control unit to build a grasping evaluation model and generate a candidate scheme; the hierarchical processing architecture ensures comprehensive perception of the object shape structure and surface properties through fine extraction and enhancement fusion of geometric and surface features, solves the grasping decision deviation problem caused by incomplete single feature information, and provides accurate and reliable feature basis for subsequent intelligent control unit to select the optimal grasping path and dynamically adjust the grasping force, thereby improving the scientificity and accuracy of grasping control from the bottom feature processing level.

[0063] The end effector of the multi-axis cooperative manipulator comprises a plurality of independently driven flexible finger structures, each finger being capable of multi-degree-of-freedom movement through a joint linkage mechanism;

[0064] The tactile perception component employs a distributed sensing array to detect the pressure distribution state of the grasping contact area;

[0065] The adsorption execution device generates and releases negative pressure through the airflow control component, forming a composite grasping mode in cooperation with the finger grasping;

[0066] The plurality of independently driven flexible finger structures of the end effector, in combination with the joint linkage mechanism, realize multi-degree-of-freedom movement, and can flexibly adjust the finger posture to fit the contour of target objects of different shapes and sizes, greatly improving the physical form adaptation capability for diversified objects. The tactile perception component employs a distributed sensing array to detect the pressure distribution state of the grasping contact area in real time, accurately feedback the uniformity of the clamping force and the contact stability, and provide data support for subsequent dynamic adjustment, avoiding object slipping or damage caused by excessive clamping due to unbalanced force. The adsorption execution device generates and releases negative pressure through the airflow control component, forming a composite grasping mode with the finger grasping. For objects with smooth surface, easy to slip or irregular shape, negative pressure adsorption can provide additional fixing force, making up for the limitations of pure mechanical clamping, significantly expanding the range of objects that can be stably grasped by the system, and enhancing the grasping reliability in complex scenarios.

[0067] The iterative optimization process of the intelligent control unit is performed according to the following flow:

[0068] First, a state evaluation space is constructed, which contains the physical attribute parameters of the target object, geometric constraint conditions and environmental interference factors, providing a basic framework for grasping scheme evaluation;

[0069] Then, based on the fusion feature information output by the multi-modal fusion processor, the initial grasping pose candidate scheme set is called in combination with the constraint parameters of the state evaluation space, and the grasping stability, operation convenience and energy consumption indicators are defined as the core evaluation parameters;

[0070] Finally, the reinforcement learning framework is used to simulate grasping test for each candidate pose, the evaluation results are quantitatively evaluated using the preset evaluation parameters, and the greedy algorithm is used to iteratively optimize the grasping path parameters until the optimal grasping path is determined;

[0071] By constructing a state evaluation space containing the physical attribute parameters of the target object, geometric constraints and environmental interference factors, a complete constraint framework is provided for the evaluation of the grasping scheme, ensuring that the evaluation process fully considers the object characteristics, spatial limitations and environmental interference, avoiding defects in the scheme due to the omission of key factors, and providing a reliable foundation for subsequent optimization; focusing on core indicators to improve the evaluation pertinence: based on the fusion of feature information, the initial candidate scheme is called, and the grasping stability, operation convenience and energy consumption are defined as the core evaluation parameters, so that the evaluation process focuses on the key requirements of the grasping task, ensuring reliable grasping without slipping, efficient operation and lower energy consumption, avoiding invalid index interference, and improving the accuracy of scheme selection; using the reinforcement learning framework to simulate the grasping test of the candidate pose, combining the preset evaluation parameter quantization results, and then iteratively optimizing the path parameters through the greedy algorithm, the optimal solution can be efficiently selected from multiple candidate schemes, reducing the path deviation caused by experience decision or single scheme test, and finally realizing the accurate planning of the grasping path, taking into account stability, convenience and low energy consumption, and improving the overall grasping efficiency and reliability.

[0072] The working process of the dynamic feedback regulation closed loop is:

[0073] In the state perception stage, the pressure distribution data of the tactile perception component and the working parameters of the adsorption execution device are collected in real time;

[0074] In the deviation analysis stage, the actual perception data is compared with the preset grasping model parameters to calculate the posture and force deviation values;

[0075] In the control correction stage, based on the deviation analysis results, the joint driving parameters of the multi-axis collaborative manipulator and the force parameters of the adsorption execution device are adjusted through the PID control algorithm;

[0076] In the state perception stage, the pressure distribution data of the tactile perception component and the working parameters of the adsorption execution device are collected in real time, which can accurately capture the uniformity of the contact force, the stability of the clamping and the adsorption effect and other key state information in the grasping process, providing immediate and comprehensive feedback data for subsequent adjustment, avoiding grasping errors caused by state information lag, in the deviation analysis stage, the actual perception data is compared with the preset grasping model parameters to calculate the posture and force deviation values, which can accurately identify problems such as insufficient force, posture deviation and adsorption failure that may occur during grasping, providing a clear target and basis for control correction, avoiding blind adjustment, in the control correction stage, based on the deviation analysis results, the joint driving parameters of the multi-axis collaborative manipulator and the force parameters of the adsorption execution device are adjusted in real time through the PID control algorithm, which can quickly respond to dynamic changes in the grasping process (such as slight object sliding and uneven force), dynamically optimize the grasping force and posture, effectively avoid object damage caused by excessive force or slipping caused by insufficient force, and significantly improve the stability and reliability of grasping in complex scenarios.

[0077] The environmental adaptability module further comprises:

[0078] An environmental perception component for detecting the distribution of obstacles and lighting conditions within the workspace;

[0079] A path planning unit for generating a collision-free motion path through a space search algorithm based on the path optimization logic of the intelligent control unit;

[0080] A lighting adaptation unit for dynamically adjusting the exposure parameters and light supplement intensity of the image acquisition component in the visual recognition module according to the environmental light detection results;

[0081] The environmental perception component detects the distribution of obstacles within the workspace in real time, and the path planning unit generates a collision-free motion path through a space search algorithm based on this, ensuring that the multi-axis collaborative robot can avoid static or dynamic obstacles during motion, effectively reducing the risk of collision between the robot and environmental objects, and ensuring the safety of the device operation and the integrity of the workspace. The lighting adaptation unit dynamically adjusts the exposure parameters and light supplement intensity of the image acquisition component in the visual recognition module according to the environmental light detection results. This can effectively deal with complex lighting scenarios such as excessive, weak or uneven lighting, avoid image blurring, feature loss or insufficient contrast due to lighting problems, and ensure that the visual recognition module can always acquire clear and effective object images, providing high-quality raw data support for subsequent three-dimensional shape data generation and surface feature extraction. Through the cooperative action of obstacle detection and collision-free path planning, and lighting dynamic adaptation, the environmental adaptability module makes the system no longer limited to a single or ideal working environment, and can operate stably in complex scenarios with obstacles and variable lighting conditions, significantly improving the system's adaptability to different working environments and expanding its range of practical application scenarios.

[0082] The intelligent learning and interaction module further comprises:

[0083] A digital twin mapping unit for real-time construction of a virtual simulation model of the physical system, simulating the motion state and force condition of the multi-axis collaborative robot in synchronization;

[0084] A human-machine interaction component for realizing operation instruction input and state feedback through gesture recognition and visual guidance;

[0085] An experience learning database for storing historical grasping case data, periodically optimizing the grasping strategy model of the intelligent control unit through clustering analysis and pattern mining based on the iterative optimization results;

[0086] The digital twin mapping unit constructs a virtual simulation model of the physical system in real time, synchronously simulating the motion state and force condition of the multi-axis collaborative manipulator. This allows testing of the grasping strategy in a virtual environment before actual physical operation, discovering potential problems such as motion interference and improper force in advance, reducing physical trial and error costs, reducing equipment wear and operation risk, improving debugging and optimization efficiency, and simplifying the traditional complex operation process. The human-computer interaction component realizes operation instruction input and state feedback through gesture recognition and visual guidance, simplifying the traditional complex operation process. Users can intuitively interact with the system through gestures or visual signals, reducing the operation threshold, while obtaining real-time system operation status, enhancing the convenience and accuracy of human-computer collaboration, and experience learning database stores historical grasping case data, periodically optimizes the grasping strategy model of the intelligent control unit based on iterative optimization results through clustering analysis and pattern mining. This enables the system to accumulate experience from past successful or failed cases, continuously iterate and upgrade the grasping strategy, gradually improve the adaptability to diversified objects and complex scenes, and realize continuous optimization of performance, expanding the task application range of the system.

[0087] In addition, the terms "first", "second", "third", etc. are used herein only to describe various circumstances, and cannot be interpreted or deduced to indicate or imply relative importance or imply the number of the technical features indicated. Therefore, the features defined as "first", "second", etc. can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "a plurality of" is at least two, such as two, three, etc., unless otherwise specifically limited.

[0088] In the description of the present application, the description of 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 the present application, the illustrative description of the above terms is not necessarily directed to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or more embodiments or examples. In addition, different embodiments or examples described in the present application and the features of different embodiments or examples can be combined and combined by those skilled in the art without contradiction.

[0089] Although the embodiments of the present application have been shown and described above, it should be understood that the above embodiments are exemplary and cannot be construed as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above embodiments within the scope of the present application.

Claims

1. A vision recognition based robot gripper control system, characterized in that, The application relates to a multi-axis collaborative manipulator and an intelligent control method thereof. The application comprises: a visual recognition module for collecting three-dimensional shape data and surface feature information of a target object, and eliminating noise interference in original images through preprocessing; a multi-modal fusion processor for receiving the three-dimensional shape data and surface feature information output by the visual recognition module, performing point cloud structured conversion on the three-dimensional shape data, extracting object geometric features through a neural network, and simultaneously performing weight distribution on the surface feature information through a feature attention mechanism to complete multi-feature fusion processing; a multi-axis collaborative manipulator provided with an end effector, wherein the end effector is integrated with a tactile perception component for contact perception and an adsorption execution device for assisting grasping; an intelligent control unit for constructing a grasping evaluation model based on the fused feature information, generating a plurality of initial grasping pose candidate schemes, and configuring a reinforcement learning framework to determine an optimal grasping path through an iterative optimization algorithm; 2.The vision-identification-based mechanical hand grasping control system according to claim 1, wherein: a dynamic feedback regulation closed loop for collecting tactile perception data of the end effector and working parameters of the adsorption execution device in real time, and dynamically correcting grasping force and posture parameters. The visual recognition module comprises: a structured light projection component for projecting a feature recognition pattern to a target object; a binocular image acquisition component for acquiring an object image containing the feature pattern through synchronous shooting from two visual angles; an ambient light adaptation component for automatically adjusting a light compensation mode according to ambient light intensity; 3. The vision-based robotic gripper control system of claim 2, wherein: an image preprocessing unit for sequentially performing distortion correction, size normalization and gray scale equalization processing on the binocular images, and calculating and generating three-dimensional shape data of the object through a stereo matching algorithm. The multi-modal fusion processor adopts a hierarchical processing architecture, which comprises: a geometric feature extraction layer for performing point cloud grid processing on the three-dimensional shape data, and extracting object contour, curvature and spatial distribution features through a multi-layer perception neural network; a surface feature analysis layer for performing texture segmentation and feature point detection on the surface feature information, and identifying object surface material attribute related features; 4. The visual recognition based robot gripper control system of claim 3, wherein: a cross-modal fusion layer for converting the geometric features and the surface features to the same feature space through feature mapping, and realizing feature enhancement fusion through an attention weight matrix. The end effector of the multi-axis collaborative manipulator comprises a plurality of independently driven flexible finger structures, and each finger realizes multi-degree-of-freedom movement through a joint linkage mechanism; the tactile perception component adopts a distributed sensing array for detecting pressure distribution states of a grasping contact area; 5. The vision-based robotic gripper control system of claim 4, wherein: the adsorption execution device realizes negative pressure generation and release through an airflow control component, and forms a composite grasping mode in cooperation with the finger grasping. The iterative optimization process of the intelligent control unit is performed according to the following procedure: firstly, a state evaluation space is constructed, which contains physical attribute parameters of the target object, geometric constraint conditions and environmental interference factors, thereby providing a basic framework for grasping scheme evaluation; then, based on the fused feature information output by the multi-modal fusion processor, the initial grasping pose candidate scheme set is called in combination with the constraint parameters of the state evaluation space, and grasping stability, operation convenience and energy consumption indicators are defined as core evaluation parameters; finally, the reinforcement learning framework is used to simulate grasping tests on each candidate pose, the evaluation results are quantitatively evaluated by using preset evaluation parameters, and the grasping path parameters are iteratively optimized by using a greedy algorithm until the optimal grasping path is determined.

6. The vision-based robotic gripper control system of claim 5, wherein: The working process of the dynamic feedback adjustment closed loop is: State perception stage, real-time acquisition of pressure distribution data of tactile perception component and working parameters of adsorption execution device; Deviation analysis stage, compare the actual perception data with the preset grabbing model parameters, calculate the posture and force deviation value; Control correction stage, based on the deviation analysis result, adjust the joint driving parameters of the multi-axis collaborative manipulator and the force parameters of the adsorption execution device through PID control algorithm.

7. The vision-based robotic gripper control system of claim 6, wherein: It also includes an environmental adaptability module, which includes: Environment perception component, for detecting the obstacle distribution and lighting conditions in the working space; Path planning unit, based on the path optimization logic of the intelligent control unit, generate collision-free motion path through space search algorithm; Light self-adaptive unit, for dynamically adjusting the exposure parameters and light supplement intensity of the image acquisition component in the visual recognition module according to the environmental light detection result.

8. The vision-based robotic gripper control system of claim 7, wherein: It also includes an intelligent learning and interaction module, which includes: Digital twin mapping unit, for real-time construction of virtual simulation model of physical system, synchronous simulation of motion state and stress condition of multi-axis collaborative manipulator; Human-computer interaction component, for realizing operation instruction input and state feedback through gesture recognition and visual guidance; Experience learning database, store historical grabbing case data, based on iterative optimization result, periodically optimize the grabbing strategy model of intelligent control unit through clustering analysis and pattern mining.

Images