A general-purpose anthropomorphic prosthesis control method and system based on visual shared control, a terminal and a storage medium

Abstract

The application relates to the technical field of visual shared control, and discloses a general anthropomorphic prosthesis control method and system based on visual shared control, a terminal and a storage medium, wherein a pre-molding prior trajectory library is constructed through sample motion reorientation of hand grabbing, and a normalized phase driven by a hand-object space distance is used for query; a candidate object three-dimensional centroid is acquired, arm motion space geometric features are extracted, and a posteriori intention probability is updated in combination with a Bayesian network to lock a target; three-dimensional Euclidean distances between a hand and the target are calculated and converted into a current phase, a trajectory library is queried to generate pre-molding motor instructions; fingertip contact force is read, and when the hand approaches the target object, a first-order admittance control is used to adjust an approaching speed, and when contact is generated, a global tactile reflex is triggered to reduce a speed and a torque, and after a state is met, a joint is locked to complete grabbing. The application realizes accurate intention estimation under multiple targets, guarantees stable and flexible grabbing of unknown objects, and improves the safety of prosthesis interaction and the naturalness of motion in a complex environment.

Description

Technical Field

[0001] This invention relates to the field of visual sharing control technology, and in particular to a general anthropomorphic prosthetic limb control method, system, terminal, and computer-readable storage medium based on visual sharing control. Background Technology

[0002] With the increasing number of upper limb amputees, patients urgently need intelligent prostheses to assist in achieving self-care. Traditional electromyography (EMG) or brain-computer interface (BCI) control methods have limitations such as poor generalization ability, excessive cognitive load, or trauma risk. Therefore, "visual shared control," which combines human intention decision-making with machine autonomous perception, has become a key path to solve the above problems. Typical visual shared control mimics human grasping behavior and mainly consists of three stages: grasping intention estimation, grasping motion planning, and closed grasping control.

[0003] Existing technologies and their limitations: Previous research has utilized head-mounted vision combined with spatial distance control of objects to control prosthetic hands, but these technologies suffer from the following key drawbacks in practical applications: (1) Intent estimation relies on rigid geometric assumptions, and multiple targets are prone to misjudgment: Existing solutions usually only collect wrist coordinates to fit a straight line in space and forcibly assume that the target object must be on a specific side of the line. This rigid assumption of pure geometry ignores the nonlinear characteristics of human three-dimensional motion, and intent misjudgment is very likely to occur when there are multiple cluttered objects on the table.

[0004] (2) Stiff gesture trajectory modeling and lack of spatiotemporal generalization ability: In motion planning, the existing technology uses a fixed-order polynomial function to rigidly fit the distance between the hand and the object and the joint angle. This deterministic model cannot be compatible with the natural variance when the human body grasps, resulting in stiff prosthetic movements. Moreover, it fails to achieve spatiotemporal decoupling, is extremely sensitive to sensor measurement noise, has poor robustness, and is strongly bound to the hardware structure of the dexterous hand, lacking the universality to be deployed to other models of dexterous hands.

[0005] (3) Lack of smooth control in closed grasping within the visual blind zone: When the hand approaches the object, visual obstruction will occur. Existing systems mostly adopt open-loop "blind grasping" at this stage, or rely solely on tactile sensation or current threshold to execute a "hard stop" strategy (i.e., locking the joint instantly when the threshold is reached). This method cannot absorb impact force at the moment of contact, cannot adapt to the contour of the object, and is extremely easy to damage fragile items.

[0006] (4) The hardware architecture has a concentrated load and lacks a secure human-computer interaction closed loop: The existing high-computing modules are mostly concentrated at the stump, which increases the risk of patients bearing weight and thermal injury. In addition, the existing head-mounted camera only serves as a one-way data input. Before the patient grasps the device, he / she cannot intuitively know the system's predicted intention. The lack of a two-way "intention confirmation mechanism" increases the patient's psychological burden and safety risks.

[0007] Therefore, existing technologies still need to be improved and developed. Summary of the Invention

[0008] The main objective of this invention is to provide a universal anthropomorphic prosthetic limb control method, system, terminal, and computer-readable storage medium based on visual shared control, aiming to solve the problems of existing visual intelligent prosthetic limbs, such as insufficient intuitiveness in intent recognition, lack of generalization ability in trajectory planning, and insufficient smoothness in closure grasping.

[0009] To achieve the above objectives, the present invention provides a universal anthropomorphic prosthetic limb control method based on visual sharing control, the universal anthropomorphic prosthetic limb control method based on visual sharing control comprising the following steps: The human hand grasping sample is acquired and its motion is redirected to obtain the optimal joint angle sequence. The hand-object spatial distance is converted into a normalized phase, and the normalized phase is used to drive the construction of a pre-shaping prior trajectory library of probabilistic motion primitives. Obtain the set of candidate objects within the field of view and the three-dimensional centroid coordinates of each candidate object, extract the spatial geometric features of the arm movement trajectory, and combine a Bayesian network to update the posterior intent probability of each candidate object to lock the target object. Calculate the three-dimensional Euclidean distance between the hand and the target object, convert the three-dimensional Euclidean distance into the current normalized phase, query the pre-shaping prior trajectory library based on the current normalized phase, generate and issue pre-shaping motor commands to control the user's arm movement; The system reads the contact force of the prosthetic fingertip. When the hand and the target object are within a preset range, the system adjusts the motor approach speed based on the first-order admittance control law. When the finger is detected to be in contact with the target object, the system triggers a global tactile reflex to reduce the speed and torque of the entire palm. When the preset state is met, the joint is locked to complete the grasp.

[0010] Optionally, the general anthropomorphic prosthetic limb control method based on vision-shared control, wherein acquiring hand grasping samples and performing motion redirection to obtain the optimal joint angle sequence, converting the hand-object spatial distance into a normalized phase, and using the normalized phase to drive the construction of a pre-shaping prior trajectory library of probabilistic motion primitives, specifically includes: A hand-grabbing video was recorded using an RGB-D depth camera. Based on the pinhole projection model of the RGB-D depth camera, the pixel coordinates were... With depth value Mapped to 3D camera coordinates : ; in, and These represent the optical centers of the RGB-D depth camera on the image plane. shaft and Axis coordinates and These represent the RGB-D depth camera at... shaft and Focal length along the axial direction; Combining depth robust sampling, calculate the human wrist With respect to the centroid of the target object 3D Euclidean distance : ; Construct a local coordinate system for the human hand and align it with the coordinate system of the target prosthetic hand. Utilize the vector field optimization redirection paradigm to calculate the target direction vectors from the wrist to each fingertip. Under the URDF kinematic constraints of a specific prosthetic hand after fine-tuning, the optimal joint angle vector at the current moment is solved. The quadratic optimization objective function is: ; in, Let be the optimal joint angle vector of the prosthetic hand that needs to be solved at the current moment. To control the number of fingertips. To accommodate the size differences between human hands and robotic arms, a vector scaling factor is applied. The first one extracted by hand The target direction vector of the fingertip Let be the positive kinematics function of the robot arm. To smooth out the penalty weights, This is the optimal joint angle at the previous moment. and These represent the minimum and maximum physical limits of motion of the prosthetic hand joint; Throughout the entire lifecycle of a single offline capture demonstration, each sampling moment is recorded and accumulated in real time at a fixed sampling frequency. The optimal joint angle vector obtained by solving ,in, , This represents the total number of sampling steps for the current action. The continuous joint angle vectors within the complete lifecycle are sequentially concatenated according to time sequence to construct the complete motion trajectory of the current capture demonstration, and this complete motion trajectory is defined as the redirected sample trajectory.

[0011] ; The complete motion trajectory is transformed into a continuous function driven by spatial distance, and an effective pre-shaping interval is defined. Calculate the current 3D Euclidean distance Normalized distance phase : ; in, This is the starting distance for pre-shaping. This is the distance at the end of the pre-shaping process. This is a truncation function; Constructing Gaussian radial basis functions : ; in, For the first The center of a Gaussian basis is the width of the Gaussian basis; Forming eigenvectors ; Sample trajectory obtained by redirection To achieve the goal, ridge regression is used to solve for the weight vector of a single crawled sample. : ; in, The design matrix is ​​constructed based on basis functions. The regularization coefficient is . Represents the identity matrix. Indicates transpose; Weight distribution of samples from multiple users Take the mean value to generate a multi-sample fusion pre-shaping prior trajectory that is decoupled from time and driven purely by spatial distance phase: ; in, This is the final sequence of continuous joint angle commands. Indicates a Gaussian distribution; This is the mean vector of the weight distribution of multiple samples; It is the covariance matrix of the weight distribution of multiple samples.

[0012] Optionally, in the aforementioned general anthropomorphic prosthetic limb control method based on visual sharing control, the normalized distance phase... The range is between 0 and 1. When the time is right, it indicates the start of pre-shaping. This indicates that the destination has been reached.

[0013] Optionally, the general anthropomorphic prosthetic limb control method based on visual shared control, wherein acquiring the candidate object set within the field of view and the three-dimensional centroid coordinates of each candidate object, extracting the spatial geometric features of the arm movement trajectory, and updating the posterior intent probability of each candidate object using a Bayesian network to lock onto the target object, specifically includes: The YOLO26-seg object detection is performed asynchronously using the edge computing module of the large arm to obtain a set of candidate objects within the field of view and the 3D centroid coordinates of each object. The set of candidate objects is represented as follows: ,in, Indicates the first within the field of vision One candidate object, The total number of candidate objects. Each candidate object Each corresponds to a three-dimensional centroid coordinate; Feature extraction based on the minimum jerkness model minimizes the jerkness integral during arm movement planning by the human brain. ; in, It represents the total jerkness during the entire movement. This indicates the total time it takes for the arm to complete the movement. Represents a continuous time variable. These represent the arm's movement in a three-dimensional Cartesian coordinate system. axis, axis, Real-time position coordinates of the axis; Three spatial geometric features reflecting the direction of the straight line position and the direction of the bell velocity are extracted, including the orthogonal deviation penalty term, the effective projection reward term, and the instantaneous tangent alignment. Among them, the orthogonal deviation penalty term Represented as: ; Among them, the effective projection reward item Represented as: ; Among them, instantaneous tangent alignment Represented as: ; in, From the starting point of the hand to the... Candidate objects The linear vector of the centroid, This is the actual displacement vector of the hand from the starting point to the current position. This represents the current three-dimensional instantaneous velocity vector of the hand. The hand is pointing to the current position of the first... Candidate objects The instantaneous direction vector; By combining directional likelihood and probabilistic likelihood, a log-observational likelihood function is constructed: ; in, In order to achieve the target hypothesis Under these conditions, the current hand trajectory was observed. The likelihood probability, All of these are weight hyperparameters obtained through pre-learning using real datasets. A smoothing function to prevent logarithmic calculations from going out of bounds; Posterior probability smoothing using exponential moving average: ; in, For the present Given a known trajectory at all times, the target assumption is... The posterior probability of its validity. The instantaneous probability calculated based on the observations of the current frame. Let be the posterior probability of the previous time step. To smoothly update the coefficients; Calculate the posterior probability of all candidate objects, given a certain target hypothesis. If the posterior probability of a hypothesis ranks first for a consecutive preset number of frames, and this probability value exceeds a preset locking threshold, then the candidate object corresponding to that hypothesis is confirmed as the final capture target, and is denoted as the target. ,in This is the index number for the final target.

[0014] Optionally, the general anthropomorphic prosthetic limb control method based on vision-shared control, wherein calculating the three-dimensional Euclidean distance between the hand and the target object, converting the three-dimensional Euclidean distance into a current normalized phase, querying the pre-shaping prior trajectory library based on the current normalized phase, generating and issuing pre-shaping motor commands to control the user's arm movement, specifically includes: Lock target Then, real-time calculation of the hand and target The three-dimensional Euclidean distance between the centroids is converted into a normalized phase. ; Normalized phase As input, in the final generated sequence of continuous joint angle commands The pre-shaping motor command can be directly retrieved to control the user's arm movement; When the user's arm advances at a slower speed or hovers, control the normalized phase. The changes are slowed down or stopped to control the prosthesis pre-shaping movements to automatically adapt to the user's spatiotemporal rhythm.

[0015] Optionally, the general anthropomorphic prosthetic limb control method based on vision-shared control, wherein reading the contact force of the prosthetic fingertip, adjusting the motor approach speed based on a first-order admittance control law when the hand and the target object are within a preset range, triggering a global tactile reflex to reduce the speed and torque of the entire palm when the finger is detected to be in contact with the target object, and locking the joint to complete the grasp when a preset state is met, specifically includes: An independent background haptic sensing thread is started, using a single-finger rotation architecture to read the pressure matrix of each fingertip. After reading a single finger, the bus mutex is briefly released, and the current real-time total contact force of the fingertip is updated in the cache. ; For lightweight prosthetic systems, the mass and spring terms of traditional second-order admittance are eliminated, and a simplified first-order admittance velocity control law is adopted: ; in, The speed command sent to the motor. To set the maximum approach initial velocity, The admittance gain coefficient, This represents the total real-time contact force at the fingertips. Calculated Confined to a safe area between; If any finger makes initial contact with an object, triggering a global reflex: the maximum speed of the entire palm motor immediately drops to a safe low speed. And the full palm torque switches to a low torque smooth mode; After the preset conditions are triggered during the contact phase, the joint position is forcibly locked. When all fingers are locked, the grasping action is successfully completed.

[0016] Optionally, in the aforementioned general anthropomorphic prosthetic limb control method based on vision sharing control, the preset conditions and corresponding locking actions include: Force lock: The set target grip strength threshold; Positional stability lock: Actual change in joint position Set a threshold and duration. Set safety time ; Timeout fallback lockout: The loop time reaches the set maximum timeout limit.

[0017] Furthermore, to achieve the above objectives, the present invention also provides a universal anthropomorphic prosthetic limb control system based on visual sharing control, wherein the universal anthropomorphic prosthetic limb control system based on visual sharing control includes: The gesture mapping and prior construction module is used to acquire human hand grasping samples and perform motion redirection to obtain the optimal joint angle sequence, convert the hand-object spatial distance into a normalized phase, and use the normalized phase to drive the construction of a pre-shaping prior trajectory library of probabilistic motion primitives. The target object locking module is used to obtain the set of candidate objects within the field of view and the three-dimensional centroid coordinates of each candidate object, extract the spatial geometric features of the arm movement trajectory, and combine a Bayesian network to update the posterior intent probability of each candidate object to lock the target object. The adaptive pre-shaping control module is used to calculate the three-dimensional Euclidean distance between the hand and the target object, convert the three-dimensional Euclidean distance into the current normalized phase, query the pre-shaping prior trajectory library according to the current normalized phase, generate and issue pre-shaping motor commands to control the user's arm movement. The asynchronous sensing and compliant control module is used to read the contact force of the prosthetic fingertip. When the hand and the target object are within a preset range, the motor approach speed is adjusted based on the first-order admittance control law. When the finger is detected to be in contact with the target object, a global tactile reflex is triggered to reduce the speed and torque of the whole palm. When the preset state is met, the joint is locked to complete the grasp.

[0018] Furthermore, to achieve the above objectives, the present invention also provides a terminal, wherein the terminal includes: a memory, a processor, and a universal anthropomorphic prosthetic limb control program based on visual sharing control stored in the memory and executable on the processor, wherein when the universal anthropomorphic prosthetic limb control program based on visual sharing control is executed by the processor, it implements the steps of the universal anthropomorphic prosthetic limb control method based on visual sharing control as described above.

[0019] Furthermore, to achieve the above objectives, the present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a general anthropomorphic prosthesis control program based on visual sharing control, and when the general anthropomorphic prosthesis control program based on visual sharing control is executed by a processor, it implements the steps of the general anthropomorphic prosthesis control method based on visual sharing control as described above.

[0020] In this invention, a human hand grasping sample is acquired and its motion is redirected to obtain the optimal joint angle sequence. The spatial distance between the hand and the object is converted into a normalized phase, and the normalized phase is used to drive the construction of a pre-shaping prior trajectory library of probabilistic motion primitives. A set of candidate objects within the field of view and the three-dimensional centroid coordinates of each candidate object are acquired. The spatial geometric features of the arm movement trajectory are extracted, and the posterior intent probability of each candidate object is updated using a Bayesian network to lock the target object. The three-dimensional Euclidean distance between the hand and the target object is calculated and converted into the current normalized phase. The pre-shaping prior trajectory library is queried based on the current normalized phase to generate and issue pre-shaping motor commands to control the user's arm movement. The contact force of the prosthetic fingertip is read. When the hand and the target object are within a preset range, the motor approach speed is adjusted based on the first-order admittance control law. When the finger is detected to be in contact with the target object, a global tactile reflex is triggered to reduce the speed and torque of the entire palm. When a preset state is met, the joint is locked to complete the grasping. This invention achieves accurate estimation of intuitive intent under multi-target conditions, enables stable and compliant grasping of unknown objects, and greatly improves the interactive safety and naturalness of the prosthetic system in complex environments. Attached Figure Description

[0021] Figure 1 This is a flowchart of a preferred embodiment of the universal anthropomorphic prosthetic limb control method based on visual sharing control of the present invention; Figure 2 This is a schematic diagram of the offline stage and the real-time shared control stage in a preferred embodiment of the universal anthropomorphic prosthetic control system based on visual shared control of the present invention; Figure 3 This is a schematic diagram of the MJM spatial geometric constraint scenario of the intent estimation module in a preferred embodiment of the universal anthropomorphic prosthetic control system based on visual shared control of the present invention. Figure 4 This is a schematic diagram of the entire execution process in a preferred embodiment of the universal anthropomorphic prosthetic control system based on visual sharing control of the present invention; Figure 5 This is a structural diagram of a preferred embodiment of the universal anthropomorphic prosthetic limb control system based on visual sharing control of the present invention; Figure 6 This is a structural diagram of a preferred embodiment of the terminal of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0023] To address the shortcomings of existing visual intelligent prosthetics, such as insufficient intuitiveness in intent recognition, lack of generalization ability in trajectory planning, insufficient smoothness in closed grasping, and concentrated wear weight, this invention adopts a distributed lightweight wearable architecture of "head visual perception + upper arm edge computing + residual limb execution," which significantly reduces the residual limb load and ensures high real-time control. At the control algorithm level, this invention innovatively combines the Minimum Jerk Model (MJM) with Bayesian networks to achieve intuitive and accurate estimation of intent under multi-objective conditions. Simultaneously, a general motion mapping pipeline decoupled from specific hardware is constructed to transform the spatial distance between the hand and object into normalized phase to drive probabilistic motion primitives (ProMP), generating an adaptive anthropomorphic pre-shaping trajectory decoupled from time. Finally, a first-order force admittance control law and a global tactile reflex mechanism of "single-finger triggering - full palm coordination" are designed for visual blind spots to achieve stable and smooth grasping of unknown objects. This invention not only possesses strong versatility, seamlessly adapting to various dexterous hand configurations, but also greatly improves the interactive safety and naturalness of movements of prosthetic systems in complex environments.

[0024] The system of this invention mainly consists of two parts: physical hardware architecture and core algorithm module.

[0025] The physical hardware architecture includes: The head-mounted visual perception module includes an RGB-D (Red Green Blue – Depth, visual data containing color and depth information) depth camera or AR (Augmented Reality) glasses worn on the head, used to acquire global 3D environment point clouds in real time from a first-person perspective; if AR glasses are used, the intent estimation results can also be projected into the user's field of vision in augmented reality form, forming an intent confirmation loop.

[0026] The wearable edge computing module for the upper arm has a built-in lightweight microprocessor and power supply unit, and is fixed to the outside of the patient's upper arm. This distributed design avoids overload of the stump and shortens the communication cable distance with the end effector, ensuring the real-time performance of the control system.

[0027] The general-purpose end-effector bionic actuator module is a multi-free prosthetic dexterous hand with fingertip force sensors. Hardware decoupling is achieved at the control layer, which is not limited to a specific model. It can be seamlessly adapted to prosthetic hands of different configurations simply by replacing the corresponding URDF (Unified Robot Description Format) file.

[0028] The core algorithm module includes: The offline general prior construction module uses deep networks to extract high-fidelity hand-object interaction 3D poses, performs general motion redirection mapping based on a fine-tuned URDF model, and constructs an original probabilistic motion primitive (ProMP) prior library based on distance and phase driving.

[0029] The online intent estimation module combines object detection to obtain the 3D centroid of the object, extracts geometric constraint features based on the Minimum Jerk Model (MJM), and uses Bayesian networks and smoothing algorithms to infer the user's grasping target in real time.

[0030] The online pre-shaping and compliant closure module utilizes the spatial distance between the hand and the object to transform the normalized phase, driving the ProMP primitive to issue spatiotemporally decoupled motor commands. During the closure contact phase, it adopts background asynchronous tactile perception and completes compliant grasping through a first-order admittance control law, a global tactile reflex mechanism, and a triple-lock state machine.

[0031] The preferred embodiment of the present invention describes a general anthropomorphic prosthetic limb control method based on vision-shared control, such as... Figure 1 , Figure 2 and Figure 4 As shown, the general anthropomorphic prosthetic limb control method based on visual sharing control includes the following steps: Step S10: Obtain the human hand grasping sample and perform motion redirection to obtain the optimal joint angle sequence. Convert the hand-object spatial distance into a normalized phase and use the normalized phase to drive the construction of a pre-shaping prior trajectory library of probabilistic motion primitives.

[0032] Specifically, step S10 includes: Step S11: High-fidelity pose extraction and spatial positioning.

[0033] A hand-grabbing video was recorded using an RGB-D depth camera. Based on the pinhole projection model of the RGB-D depth camera, the pixel coordinates were... With depth value Mapped to 3D camera coordinates : ; in, and These represent the optical centers (or principal points) of the RGB-D depth camera on the image plane. shaft and Axis coordinates and These represent the RGB-D depth camera at... shaft and Focal length along the axial direction (in pixels).

[0034] Combined with depth robust sampling (median filtering), calculate the human wrist area With respect to the centroid of the target object 3D Euclidean distance : .

[0035] Step S12: Universal motion redirection based on URDF fine-tuning.

[0036] A local coordinate system for the human hand is constructed and aligned with the coordinate system of the target prosthetic hand. To achieve universal control with hardware decoupling, a vector field optimization reorientation paradigm is used to calculate the target direction vectors from the wrist to each fingertip. Under the URDF kinematic constraints of a specific prosthetic hand after fine-tuning, the optimal joint angle vector at the current moment is solved. ,right The quadratic optimization objective function for the solution process is: ; in, Let be the optimal joint angle vector of the prosthetic hand that needs to be solved at the current moment. The number of fingertips to control (e.g., 5 fingers). To accommodate the size differences between human hands and robotic arms, a vector scaling factor is applied. The first one extracted by hand The target direction vector of the fingertip This is the positive kinematics function of the robot (the fingertip position is calculated based on the URDF model at the current joint angle). To smooth out the penalty weights, This is the optimal joint angle at the previous moment. and These represent the minimum and maximum physical range of motion of the prosthetic hand joint, respectively.

[0037] Throughout the entire lifecycle of a single offline capture demonstration, each sampling moment is recorded and accumulated in real time at a fixed sampling frequency. The optimal joint angle vector obtained by solving ,in, , This represents the total number of sampled steps for the current action. Continuous joint angle vectors within the complete lifecycle are sequentially concatenated according to their chronological order to construct the complete motion trajectory of the current grabbing demonstration. This complete motion trajectory is defined as the redirected sample trajectory.

[0038] ; This provides baseline target time series data for probabilistic motion primitive (ProMP) modeling in step S13.

[0039] Step S13: Spatiotemporally decoupled probabilistic motion primitive (ProMP) modeling.

[0040] Traditional motion trajectories heavily rely on absolute time. The generalization ability is low. This invention proposes to transform the complete motion trajectory into a continuous function driven by spatial distance, and to define an effective pre-shaping interval. Calculate the current three-dimensional Euclidean distance Normalized distance phase : ; Wherein, the normalized distance phase The range is between 0 and 1. When the time is right, it indicates the start of pre-shaping. Indicates arrival at the destination. This is the starting distance for pre-shaping (e.g., 0.50m). This is the pre-shaping end distance (e.g., 0.20m). This is a truncation function to ensure that the phase does not go out of bounds.

[0041] Constructing Gaussian radial basis functions : ; in, For the first The center of a Gaussian basis is the width of the Gaussian basis; Forming eigenvectors .

[0042] Sample trajectory obtained by redirection To achieve the goal, ridge regression is used to solve for the weight vector of a single crawled sample. : ; in, The design matrix is ​​constructed based on basis functions. The regularization coefficient is used to prevent overfitting. Represents the identity matrix. This indicates transpose.

[0043] Finally, the weight distribution of samples from multiple users is calculated. Take the mean value to generate a multi-sample fusion pre-shaping prior trajectory that is decoupled from time and driven purely by spatial distance phase: ; in, The final generated sequence of continuous joint angle commands is entirely composed of normalized distance and phase. (i.e., hand-object distance) driven, Indicates a Gaussian distribution; This is the mean vector of the weight distribution of multiple samples; It is the covariance matrix of the weight distribution of multiple samples.

[0044] Step S20: Obtain the set of candidate objects within the field of view and the three-dimensional centroid coordinates of each candidate object, extract the spatial geometric features of the arm movement trajectory, and update the posterior intent probability of each candidate object using a Bayesian network to lock the target object.

[0045] Specifically, step S20 includes: Step S21, Global Awareness.

[0046] The edge computing module of the large arm asynchronously runs YOLO26-seg (You Only Look Once - Segmentation) object detection to obtain a set of candidate objects within the field of view and the three-dimensional centroid coordinates of each object. The set of candidate objects is represented as follows: ,in, Indicates the first within the field of vision One candidate object, The total number of candidate objects. Each candidate object Each corresponds to a three-dimensional centroid coordinate.

[0047] Step S22: Feature extraction based on the Minimum Jerk Model (MJM).

[0048] When the human brain plans arm movements, it tends to minimize the integral of the jerk (rate of change of acceleration) during the movement: ; in, It represents the total jerkiness (or jerkiness cost function) during the entire motion process. This indicates the total time it takes for the arm to complete the movement. Represents a continuous time variable. These represent the arm's movement in a three-dimensional Cartesian coordinate system. axis, axis, Real-time position coordinates of the axis.

[0049] Based on this human intuitive physiological prior, as shown in Figure 3, this invention extracts three spatial geometric features that reflect the direction of the straight line position and the direction of the bell-shaped velocity, including the orthogonal deviation penalty term, the effective projection reward term, and the instantaneous tangent alignment.

[0050] Among them, the orthogonal deviation penalty term Represented as: ; Among them, the effective projection reward item Represented as: ; Among them, instantaneous tangent alignment Represented as: ; in, From the starting point of the hand to the... Candidate objects The linear vector of the centroid, This is the actual displacement vector of the hand from the starting point to the current position. This represents the current three-dimensional instantaneous velocity vector of the hand. The hand is pointing to the current position of the first... Candidate objects The instantaneous direction vector.

[0051] Step S23: Bayesian intent probability update.

[0052] By combining directional likelihood and probabilistic likelihood, a log-observational likelihood function is constructed: ; in, In order to achieve the target hypothesis (i.e., user intent to crawl the first) Candidate objects Under the condition of ), the current hand trajectory is observed. The likelihood probability, All of these are weight hyperparameters obtained through pre-learning using real datasets. A smoothing function to prevent logarithmic calculations from going out of bounds.

[0053] Posterior probability smoothing is achieved by combining exponential moving average (EMA): ; in, For the present Given a known trajectory at all times, the target assumption is... The posterior probability of its validity. The instantaneous probability is calculated based on the observations of the current frame. Let be the posterior probability of the previous time step. For smoothing the update coefficients (range 0 to 1).

[0054] Calculate the posterior probability of all candidate objects, given a certain target hypothesis. If the posterior probability of a hypothesis ranks first for a consecutive preset number of frames, and this probability value exceeds a preset locking threshold, then the candidate object corresponding to that hypothesis is confirmed as the final capture target, and is denoted as the target. ,in This is the index number for the final target.

[0055] Step S30: Calculate the three-dimensional Euclidean distance between the hand and the target object, convert the three-dimensional Euclidean distance into the current normalized phase, query the pre-shaping prior trajectory library according to the current normalized phase, generate and issue pre-shaping motor commands to control the user's arm movement.

[0056] Specifically, lock onto the target Then, real-time calculation of the hand and target The three-dimensional Euclidean distance between the centroids is converted into a normalized phase. Normalize the phase As input, in the final generated sequence of continuous joint angle commands The pre-shaping motor command is directly retrieved to control the user's arm movement; when the user's arm advances at a slower speed or hovers, the normalized phase is controlled. The changes are slowed down or stopped to control the prosthesis pre-shaping movements to automatically adapt to the user's spatiotemporal rhythm.

[0057] Step S40: Read the contact force of the prosthetic fingertip. When the hand and the target object are within a preset range, adjust the motor approach speed based on the first-order admittance control law. When the finger is detected to be in contact with the target object, trigger a global tactile reflex to reduce the speed and torque of the whole palm. Lock the joint to complete the grasp when the preset state is met.

[0058] Specifically, step S40 specifically includes: Step S41: Decoupling of sensor read / write operations.

[0059] An independent background haptic sensing thread is started, using a single-finger rotation architecture to read the pressure matrix of each fingertip. After reading a single finger, the bus mutex is briefly released, and the current real-time total contact force of the fingertip is updated in the cache. This avoids blocking the main control thread from issuing motion commands.

[0060] Step S42: Tactile state machine and first-order admittance force control.

[0061] For lightweight prosthetic systems, the mass and spring terms of traditional second-order admittance are eliminated, and a simplified first-order admittance velocity control law is adopted: ; in, The speed command sent to the motor. To set the maximum approach initial velocity, It is the admittance gain coefficient (reflecting the compliance of the system). This represents the total real-time contact force at the fingertips; the calculated value is... Confined to a safe area between.

[0062] Step S43, Global tactile reflex coordination.

[0063] If any finger makes initial contact with an object ( (Contact threshold), triggering global reflection: immediately reduces the maximum speed of the entire palm motor to a safe low speed. Furthermore, the full palm torque is switched to a low torque smooth mode.

[0064] Step S44: Triple lockout protection.

[0065] After a preset condition (any of the following preset conditions) is triggered during the contact phase, the joint position is forcibly locked: (1) Force lock: The set target gripping force threshold.

[0066] (2) Positional stability lock: The actual change in joint position Set a threshold and duration. Set safety time .

[0067] (3) Timeout fallback lockout: The loop time reaches the set maximum timeout limit.

[0068] When all fingers are locked, the grasping action is successfully completed.

[0069] The main innovations and corresponding technical effects of this invention are as follows: (1) Intent estimation introduces the three-dimensional spatial geometric features of the Human Minimum Jerk Model (MJM): It abandons the rigid assumption of "spatial linear regression + fixed side" in the existing technology, and innovatively extracts geometric features (orthogonal deviation, projection progress, instantaneous tangent) based on MJM, and combines them with Bayesian network for posterior probability update. It completely solves the pain point of existing systems being prone to misjudgment in cluttered multi-object scenes, and achieves high-accuracy target locking that conforms to human intuition without the need for users to deliberately track their eyes.

[0070] (2) The trajectory generation adopts the probabilistic motion primitive (ProMP) driven by "distance-phase": Instead of using the trajectory function that strongly depends on absolute time, the spatial distance between the hand and the object is transformed into a normalized phase, which drives the ProMP to generate action commands; at the same time, a general redirection algorithm with fine-tuned URDF model is adopted. Perfect spatiotemporal adaptation is achieved (when the user's arm advances slowly or hovers, the prosthetic hand gesture will decelerate or stop synchronously), which greatly improves the generalization ability; at the same time, it is decoupled from specific hardware and can be easily adapted to various multi-finger dexterous hands.

[0071] (3) The closed-loop control adopts an asynchronous first-order admittance and a "global tactile reflex" mechanism: In response to the visual blind spot caused by hand occlusion, a background single-finger rotation asynchronous reading architecture and a simplified first-order force admittance control law are designed. A global reflex is introduced, which "triggers full-palm deceleration and smoothness upon the first contact of any finger" and a triple lock-up guarantee (force, position stability, and timeout) are also introduced. This makes up for the hardware defects of low polling frequency and high communication delay of wearable prosthetic sensor, eliminates the hard contact impact caused by traditional open-loop blind grasping, and ensures safe and smooth grasping of unknown objects.

[0072] (4) The hardware adopts a distributed wearable architecture of "edge computing on the upper arm + visual intent confirmation": This changes the previous practice of concentrating computing power on the stump or relying on an external host, and adopts a design that separates wearable computing on the upper arm from visual interaction with AR on the head. Physically, the computing load and the weight of the end effector are decoupled, which greatly reduces the weight-bearing and thermal damage to amputees; at the same time, AR can be used to provide visual intent feedback, forming a safe human-computer interaction closed loop.

[0073] Furthermore, such as Figure 5 As shown, based on the above-mentioned universal anthropomorphic prosthetic limb control method based on visual sharing control, the present invention also provides a universal anthropomorphic prosthetic limb control system based on visual sharing control, wherein the universal anthropomorphic prosthetic limb control system based on visual sharing control includes: The gesture mapping and prior construction module 51 is used to acquire human hand grasping samples and perform motion redirection to obtain the optimal joint angle sequence, convert the hand-object spatial distance into a normalized phase, and use the normalized phase to drive the construction of a pre-shaping prior trajectory library of probabilistic motion primitives. The target object locking module 52 is used to obtain the candidate object set within the field of view and the three-dimensional centroid coordinates of each candidate object, extract the spatial geometric features of the arm movement trajectory, and combine the Bayesian network to update the posterior intent probability of each candidate object to lock the target object. The adaptive pre-shaping control module 53 is used to calculate the three-dimensional Euclidean distance between the hand and the target object, convert the three-dimensional Euclidean distance into the current normalized phase, query the pre-shaping prior trajectory library according to the current normalized phase, generate and issue pre-shaping motor commands to control the user's arm movement. The asynchronous sensing and compliant control module 54 is used to read the contact force of the prosthetic fingertip. When the hand and the target object are within a preset range, the motor approach speed is adjusted based on the first-order admittance control law. When the finger is detected to be in contact with the target object, a global tactile reflex is triggered to reduce the speed and torque of the whole palm. When the preset state is met, the joint is locked to complete the grasping.

[0074] Furthermore, such as Figure 6 As shown, based on the above-mentioned general anthropomorphic prosthetic limb control method and system based on visual sharing control, the present invention also provides a terminal, which includes a processor 10, a memory 20 and a display 30. Figure 6 Only some of the terminal components are shown; however, it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0075] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as a hard disk or memory. In other embodiments, the memory 20 may be an external storage device of the terminal, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc. Further, the memory 20 may include both internal and external storage devices. The memory 20 is used to store application software and various types of data installed on the terminal, such as program code installed on the terminal. The memory 20 may also be used to temporarily store data that has been output or will be output. In one embodiment, the memory 20 stores a general anthropomorphic prosthetic limb control program 40 based on vision sharing control. This general anthropomorphic prosthetic limb control program 40 based on vision sharing control can be executed by the processor 10, thereby implementing the general anthropomorphic prosthetic limb control method based on vision sharing control in this application.

[0076] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor, or other data processing chip, used to run program code stored in the memory 20 or process data, such as executing the general anthropomorphic prosthetic control method based on vision sharing control.

[0077] In some embodiments, the display 30 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. The display 30 is used to display information on the terminal and to display a visual user interface. The terminal's processor 10, memory 20, and display 30 communicate with each other via a system bus.

[0078] In one embodiment, when the processor 10 executes the general anthropomorphic prosthetic control program 40 based on vision-sharing control in the memory 20, it implements the steps of the general anthropomorphic prosthetic control method based on vision-sharing control as described above.

[0079] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a general anthropomorphic prosthesis control program based on visual sharing control, and the general anthropomorphic prosthesis control program based on visual sharing control, when executed by a processor, implements the steps of the general anthropomorphic prosthesis control method based on visual sharing control as described above.

[0080] In summary, this invention provides a universal anthropomorphic prosthetic limb control method, system, terminal, and computer-readable storage medium based on vision-shared control. The method includes: acquiring hand grasping samples and performing motion redirection to obtain an optimal joint angle sequence; converting the hand-object spatial distance into a normalized phase; and using the normalized phase to drive the construction of a pre-shaping prior trajectory library of probabilistic motion primitives; acquiring a set of candidate objects within the field of view and the three-dimensional centroid coordinates of each candidate object; extracting the spatial geometric features of the arm's motion trajectory; and updating the posterior intent probability of each candidate object using a Bayesian network. The system identifies a target object; calculates the three-dimensional Euclidean distance between the hand and the target object, converts this distance into a current normalized phase, queries the pre-shaping prior trajectory library based on the current normalized phase, generates and issues pre-shaping motor commands to control the user's arm movement; reads the contact force of the prosthetic fingertips, and adjusts the motor approach speed based on a first-order admittance control law when the hand and the target object are within a preset range. When finger contact with the target object is detected, a global tactile reflex is triggered to reduce the speed and torque of the entire palm, and the joint is locked to complete the grasp when a preset state is met. This invention achieves accurate estimation of intuitive intent under multi-target conditions, enables stable and compliant grasping of unknown objects, and greatly improves the interactive safety and naturalness of the prosthetic system in complex environments.

[0081] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal that includes that element.

[0082] Of course, those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.). The program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The computer-readable storage medium can be a memory, magnetic disk, optical disk, etc.

[0083] It should be understood that the application of the present invention is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A general anthropomorphic prosthetic limb control method based on visual sharing control, characterized in that, The general anthropomorphic prosthetic limb control method based on vision sharing includes: The human hand grasping sample is acquired and its motion is redirected to obtain the optimal joint angle sequence. The hand-object spatial distance is converted into a normalized phase, and the normalized phase is used to drive the construction of a pre-shaping prior trajectory library of probabilistic motion primitives. Obtain the set of candidate objects within the field of view and the three-dimensional centroid coordinates of each candidate object, extract the spatial geometric features of the arm movement trajectory, and combine a Bayesian network to update the posterior intent probability of each candidate object to lock the target object. Calculate the three-dimensional Euclidean distance between the hand and the target object, convert the three-dimensional Euclidean distance into the current normalized phase, query the pre-shaping prior trajectory library based on the current normalized phase, generate and issue pre-shaping motor commands to control the user's arm movement; The system reads the contact force of the prosthetic fingertip. When the hand and the target object are within a preset range, the system adjusts the motor approach speed based on the first-order admittance control law. When the finger is detected to be in contact with the target object, the system triggers a global tactile reflex to reduce the speed and torque of the entire palm. When the preset state is met, the joint is locked to complete the grasp.

2. The universal anthropomorphic prosthetic limb control method based on visual sharing control according to claim 1, characterized in that, The process of acquiring hand-grabbing samples and redirecting their motion to obtain the optimal joint angle sequence, converting the hand-object spatial distance into a normalized phase, and using the normalized phase to drive the construction of a pre-shaping prior trajectory library of probabilistic motion primitives specifically includes: A hand-grabbing video was recorded using an RGB-D depth camera. Based on the pinhole projection model of the RGB-D depth camera, the pixel coordinates were... With depth value Mapped to 3D camera coordinates : ； in, and These represent the optical centers of the RGB-D depth camera on the image plane. shaft and Axis coordinates and These represent the RGB-D depth camera at... shaft and Focal length along the axial direction; Combining depth robust sampling, calculate the human wrist With respect to the centroid of the target object 3D Euclidean distance : ； Construct a local coordinate system for the human hand and align it with the coordinate system of the target prosthetic hand. Utilize the vector field optimization redirection paradigm to calculate the target direction vectors from the wrist to each fingertip. Under the URDF kinematic constraints of a specific prosthetic hand after fine-tuning, the optimal joint angle vector at the current moment is solved. The quadratic optimization objective function is: ； in, Let be the optimal joint angle vector of the prosthetic hand that needs to be solved at the current moment. To control the number of fingertips. To accommodate the size differences between human hands and robotic arms, a vector scaling factor is applied. The first one extracted by hand The target direction vector of the fingertip Let be the positive kinematics function of the robot arm. To smooth out the penalty weights, This is the optimal joint angle at the previous moment. and These represent the minimum and maximum physical limits of motion of the prosthetic hand joint; Throughout the entire lifecycle of a single offline capture demonstration, each sampling moment is recorded and accumulated in real time at a fixed sampling frequency. The optimal joint angle vector obtained by solving ,in, , This represents the total number of sampling steps for the current action. The continuous joint angle vectors within the complete lifecycle are sequentially concatenated according to time sequence to construct the complete motion trajectory of the current capture demonstration, and this complete motion trajectory is defined as the redirected sample trajectory. ； The complete motion trajectory is transformed into a continuous function driven by spatial distance, and an effective pre-shaping interval is defined. Calculate the current three-dimensional Euclidean distance Normalized distance phase : ； in, This is the starting distance for pre-shaping. This is the distance at the end of the pre-shaping process. This is a truncation function; Constructing Gaussian radial basis functions : ； in, For the first The center of a Gaussian basis is the width of the Gaussian basis; Forming eigenvectors ; Sample trajectory obtained by redirection To achieve the goal, ridge regression is used to solve for the weight vector of a single crawled sample. : ； in, The design matrix is ​​constructed based on basis functions. The regularization coefficient is . Represents the identity matrix. Indicates transpose; Weight distribution of samples from multiple users Take the mean value to generate a multi-sample fusion pre-shaping prior trajectory that is decoupled from time and driven purely by spatial distance phase: ； in, This is the final sequence of continuous joint angle commands. Indicates a Gaussian distribution; This is the mean vector of the weight distribution of multiple samples; It is the covariance matrix of the weight distribution of multiple samples.

3. The universal anthropomorphic prosthetic limb control method based on visual sharing control according to claim 2, characterized in that, The normalized distance phase The range is between 0 and 1. When the time is right, it indicates the start of pre-shaping. This indicates that the destination has been reached.

4. The universal anthropomorphic prosthetic limb control method based on visual sharing control according to claim 2, characterized in that, The process of acquiring a set of candidate objects within the field of view and the three-dimensional centroid coordinates of each candidate object, extracting the spatial geometric features of the arm's motion trajectory, and updating the posterior intent probability of each candidate object using a Bayesian network to lock onto the target object specifically includes: The YOLO26-seg object detection is asynchronously run using the edge computing module of the large arm to obtain a set of candidate objects within the field of view and the 3D centroid coordinates of each object. The set of candidate objects is represented as follows: ,in, Indicates the first within the field of vision One candidate object, The total number of candidate objects. Each candidate object Each corresponds to a three-dimensional centroid coordinate; Feature extraction based on the minimum jerkness model minimizes the jerkness integral during arm movement planning by the human brain. ； in, It represents the total jerkness during the entire movement. This indicates the total time it takes for the arm to complete the movement. Represents a continuous time variable. These represent the arm's movement in a three-dimensional Cartesian coordinate system. axis, axis, Real-time position coordinates of the axis; Three spatial geometric features reflecting the direction of the straight line position and the direction of the bell velocity are extracted, including the orthogonal deviation penalty term, the effective projection reward term, and the instantaneous tangent alignment. Among them, the orthogonal deviation penalty term Represented as: ； Among them, the effective projection reward item Represented as: ； Among them, instantaneous tangent alignment Represented as: ； in, From the starting point of the hand to the... Candidate objects The linear vector of the centroid, This is the actual displacement vector of the hand from the starting point to the current position. This represents the current three-dimensional instantaneous velocity vector of the hand. The hand is pointing to the current position of the first... Candidate objects The instantaneous direction vector; By combining directional likelihood and probabilistic likelihood, a log-observational likelihood function is constructed: ； in, In order to achieve the target hypothesis Under these conditions, the current hand trajectory was observed. The likelihood probability, All of these are weight hyperparameters obtained through pre-learning using real datasets. A smoothing function to prevent logarithmic calculations from going out of bounds; Posterior probability smoothing using exponential moving average: ； in, For the present Given a known trajectory at all times, the target assumption is... The posterior probability of its validity. The instantaneous probability calculated based on the observations of the current frame. Let be the posterior probability of the previous time step. To smoothly update the coefficients; Calculate the posterior probability of all candidate objects, given a certain target hypothesis. If the posterior probability of a hypothesis ranks first for a consecutive preset number of frames, and this probability value exceeds a preset locking threshold, then the candidate object corresponding to that hypothesis is confirmed as the final capture target, and is denoted as the target. ,in This is the index number for the final target.

5. The universal anthropomorphic prosthetic limb control method based on visual sharing control according to claim 4, characterized in that, The process of calculating the three-dimensional Euclidean distance between the hand and the target object, converting the three-dimensional Euclidean distance into a current normalized phase, querying the pre-shaping prior trajectory library based on the current normalized phase, generating and issuing pre-shaping motor commands to control the user's arm movement, specifically includes: Lock target Then, real-time calculation of the hand and target The three-dimensional Euclidean distance between the centroids is converted into a normalized phase. ; Normalized phase As input, in the final generated sequence of continuous joint angle commands The pre-shaping motor command can be directly retrieved to control the user's arm movement; When the user's arm advances at a slower speed or hovers, control the normalized phase. The changes are slowed down or stopped to control the prosthesis pre-shaping movements to automatically adapt to the user's spatiotemporal rhythm.

6. The universal anthropomorphic prosthetic limb control method based on visual sharing control according to claim 5, characterized in that, The process of reading the contact force of the prosthetic fingertip involves adjusting the motor approach speed based on a first-order admittance control law when the hand and the target object are within a preset range. When the finger is detected to be in contact with the target object, a global tactile reflex is triggered to reduce the speed and torque of the entire palm. The joint is locked when a preset state is met to complete the grasping action. Specifically, this includes: An independent background haptic sensing thread is started, using a single-finger rotation architecture to read the pressure matrix of each fingertip. After reading a single finger, the bus mutex is briefly released, and the current real-time total contact force of the fingertip is updated in the cache. ; For lightweight prosthetic systems, the mass and spring terms of traditional second-order admittance are eliminated, and a simplified first-order admittance velocity control law is adopted: ； in, The speed command sent to the motor. To set the maximum approach initial velocity, The admittance gain coefficient, This represents the total real-time contact force at the fingertips. Calculated Confined to a safe area between; If any finger makes initial contact with an object, triggering a global reflex: the maximum speed of the entire palm motor immediately drops to a safe low speed. And the full palm torque switches to a low torque smooth mode; After the preset conditions are triggered during the contact phase, the joint position is forcibly locked. When all fingers are locked, the grasping action is successfully completed.

7. The universal anthropomorphic prosthetic limb control method based on visual sharing control according to claim 6, characterized in that, The preset conditions and corresponding locking actions include: Force lock: The set target grip strength threshold; Positional stability lock: Actual change in joint position Set a threshold and duration. Set safety time ; Timeout fallback lockout: The loop time reaches the set maximum timeout limit.

8. A universal anthropomorphic prosthetic limb control system based on vision-shared control, characterized in that, The general anthropomorphic prosthetic limb control system based on vision sharing control includes: The gesture mapping and prior construction module is used to acquire human hand grasping samples and perform motion redirection to obtain the optimal joint angle sequence, convert the hand-object spatial distance into a normalized phase, and use the normalized phase to drive the construction of a pre-shaping prior trajectory library of probabilistic motion primitives. The target object locking module is used to obtain the set of candidate objects within the field of view and the three-dimensional centroid coordinates of each candidate object, extract the spatial geometric features of the arm movement trajectory, and combine a Bayesian network to update the posterior intent probability of each candidate object to lock the target object. The adaptive pre-shaping control module is used to calculate the three-dimensional Euclidean distance between the hand and the target object, convert the three-dimensional Euclidean distance into the current normalized phase, query the pre-shaping prior trajectory library according to the current normalized phase, generate and issue pre-shaping motor commands to control the user's arm movement. The asynchronous sensing and compliant control module is used to read the contact force of the prosthetic fingertip. When the hand and the target object are within a preset range, the motor approach speed is adjusted based on the first-order admittance control law. When the finger is detected to be in contact with the target object, a global tactile reflex is triggered to reduce the speed and torque of the whole palm. When the preset state is met, the joint is locked to complete the grasp.

9. A terminal, characterized in that, The terminal includes: a memory, a processor, and a universal anthropomorphic prosthetic limb control program based on visual sharing control stored in the memory and executable on the processor. When the universal anthropomorphic prosthetic limb control program based on visual sharing control is executed by the processor, it implements the steps of the universal anthropomorphic prosthetic limb control method based on visual sharing control as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a general anthropomorphic prosthesis control program based on visual sharing control, which, when executed by a processor, implements the steps of the general anthropomorphic prosthesis control method based on visual sharing control as described in any one of claims 1-7.

Images