Fine teleoperation mapping control method for five-finger dexterous hand

Abstract

A kind of five-finger dexterous hand fine teleoperation mapping control method is provided based on Meta Quest 3VR head-mounted display and multiple algorithm optimization mapping control will search section, the key point data of human hand is collected by VR equipment;Adaptive SG filter is used to denoise the collected instruction signal, and physiological tremor and other high-frequency noise are filtered out;The kinematic mapping model from human hand key point to dexterous hand joint angle is established, and the Cartesian space pose is converted into joint space instruction;Based on Gaussian mixture model, the probability model of robot working space is constructed, and the collision detection and safety check are carried out combined with GJK algorithm.

Description

Technical Field

[0001] This invention relates to a method for precise teleoperation mapping control of a five-finger dexterous hand, belonging to the field of robot control technology. Background Technology

[0002] Teleoperated robots play a crucial role in high-risk environments such as space exploration and disaster relief. However, existing teleoperation methods (such as joysticks and keyboards) often lack intuitiveness and struggle to perform complex and delicate tasks. While motion-sensing control methods based on data gloves or exoskeletons improve intuitiveness, physiological tremors of the operator's hands (typically in the 6-15Hz range) are directly transmitted to the robot, affecting operational accuracy. Furthermore, due to the differences between robot and human anatomy, direct mapping can easily lead to the robot exceeding its workspace or colliding with other parts of the body, and there is a lack of effective safety verification mechanisms. Summary of the Invention

[0003] The technical problem solved by this invention is: addressing the shortcomings of existing technologies, such as poor operational intuitiveness, low operational accuracy due to human physiological tremors, and lack of real-time safety verification for the nonlinear workspace of heterogeneous robotic arms, a fine teleoperation mapping control method for five-finger dexterity hand is proposed.

[0004] The present invention solves the above-mentioned technical problem through the following technical solution: A method for fine teleoperation mapping control of a five-finger dexterous hand includes: Real-time capture of the operator's key hand position and posture information, and preprocessing of the collected position and posture information; The initial control command is preset, and the preprocessed hand key point position and posture information is smoothed. Establish a kinematic mapping model from capturing the key points of the operator's hand to the joint angles of the dexterous hand and the robotic arm; The preset verification threshold is used to perform a dual verification process for the dexterous hand and robotic arm, including workspace probability verification and collision detection verification, using initial control commands. Once both checks pass, the kinematic mapping model is used to predict the joint angles of the dexterous hand and robotic arm at the next moment, and real-time control commands are generated to the robot controller to realize remote operation control of the dexterous hand and robotic arm.

[0005] The real-time capture of the key hand position and posture information is obtained by visual tracking using a VR interactive device. The VR interactive device captures the operator's key hand position and posture information in real time at a fixed frequency. The preprocessing method is as follows: the collected hand key point positions and posture information are serialized into Cartesian space pose data and sent to the server endpoint of the robot control network in real time through the preset TCP / IP protocol. The server endpoint converts the received serialized data message into a standard format message of the robot control network, which is used as the preprocessed hand key point positions and posture information.

[0006] The method for smoothing the hand key point positions and posture information obtained after preprocessing is as follows: An SG filter is used to denoise the initial control command, which is achieved through the filter calculation formula. The initial control command is a noise-contaminated command signal, represented as follows: ,in For noise; The filtering calculation formula is based on fitting an nth-order polynomial to a length of using the least squares method. The input sample set within the sliding window is derived, and its form is a linear combination of the input signals:

[0007] In the formula, The initial control command is given at the current time. Sampled values ​​within the center window , To fix the impulse response coefficient, the optimization problem is... Select the optimal window length so that the response bandwidth of the SG filter is greater than a certain value, so as to retain the signal waveform characteristics while filtering out jitter.

[0008] The objective of the SG filter is set as minimizing the cost function used to solve for the fixed impulse response coefficients. The cost function is:

[0009] In the formula, The coefficients of the fitted polynomial, Let be the order of the polynomial. Through... Differentiate and set it to zero, solve the least squares problem, and obtain the optimal polynomial coefficients. Take the center point of the window Fitted values And substitute them into the filtering formula to obtain the fixed impulse response coefficient. .

[0010] The kinematic mapping model establishes a forward kinematic model based on the relevant parameters of the dexterous hand and the robotic arm, and uses the Levenberg-Marquardt iterative algorithm to solve the inverse kinematic model. The iterative algorithm is as follows:

[0011] In the formula, This is the joint angle vector for the next moment. This is the current joint angle vector. For Jacobian matrices, Let be the end-effector pose error function. is the damping coefficient.

[0012] The end pose error function In conjunction with the design of an adaptive damping adjustment strategy, when hour, To ensure rapid convergence; when the error hour, Halving for convergence control, end-effector pose error function By employing an adaptive damping adjustment strategy, a kinematic mapping model, and an LM iterative algorithm, the relative positions of key hand points in a dexterous hand are directly mapped to the driving angles of each finger joint to achieve five-finger dexterous hand reproduction control.

[0013] In the kinematic mapping model, the Denavit-Hartenberg method is used to determine the homogeneous transformation matrix of adjacent links in the robotic arm, establishing the forward kinematic model of the robotic arm, and the LM iterative algorithm is used to solve the inverse kinematic model. The link to the first Homogeneous transformation matrix of the links for:

[0014] In the formula, Let be the rotation angle about the z-axis, used to describe the first... The link to the first The rotation of each link around the z-axis is represented by the joint angle in the DH parameters; Let be the translation distance around the z-axis, and be the th . The link to the first The amount of movement of each link along the z-axis is the link offset in the DH parameters. Let be the translation distance along the x-axis, and be the th . The link to the first The length of each link along the x-axis is the link length in the DH parameters; Let be the rotation angle about the x-axis, used to describe the first... The link to the first The rotation of the link about the x-axis is the link twist angle in the DH parameters; Indicates rotation about the z-axis The rotation transformation matrix of the angle. Indicates translation about the z-axis The translation matrix of the distance. Indicates translation along the x-axis The translation matrix of the distance. Indicates rotation about the x-axis The rotation transformation matrix of the angle.

[0015] The workspace probability verification and collision detection verification are both implemented through initial control commands before the real-time control commands are issued. The method of the dual verification process is as follows:

[0016] In the formula, For workspace, The end-effector pose vector. and Let be the mean and covariance of the k-th Gaussian distribution, respectively. In the workspace probability verification process, when the calculated probability density value of the current instruction pose... Below the verification threshold If so, the initial control command is determined to be outside the safe working space and is truncated; The collision detection verification process is used to analyze the minimum distance between the convex hulls of each link of the robot body. By calculating the difference set of the convex hulls of adjacent links, the collision detection is converted into a distance from the origin using the calculated difference set to determine whether the minimum distance is included in the difference set. If the minimum distance between two link convex hulls is 0, a collision is determined to have occurred, and the collision detection verification is considered to have failed, triggering the robot's emergency stop protection or obstacle avoidance strategy.

[0017] Once both checks pass, the real-time control command is updated according to the preset control strategy, and the joint angle vector for the next moment obtained through the kinematic mapping model is sent to the robot controller for execution via the communication interface; if either check fails, the current control command is immediately blocked and a safety warning is triggered to prevent mechanical damage to the robot. The dexterous hand and robotic arm control motion parameters, planning filters, and safety boundaries are all visualized through a 3D display user interface.

[0018] The advantages of this invention compared to the prior art are: (1) The present invention provides a fine teleoperation mapping control method for a five-finger dexterous hand. By integrating SG filtering denoising, kinematic precision mapping, GMM probability space verification and GJK real-time collision detection technology, it realizes high-precision, low-latency and safe and reliable fine teleoperation of robots. By introducing the SG filtering algorithm, the signal is smoothed in the time domain by using polynomial fitting, which effectively solves the interference of human physiological tremor of 6-15Hz on fine operation. Compared with the traditional filtering method, it has a smaller phase delay and the measured response frequency can reach 60Hz, which ensures the real-time operation. (2) This invention abandons the traditional simple joint limiting method and uses the GMM probability model to accurately describe the highly nonlinear workspace of the robotic arm, realizing real-time reachability analysis of any complex pose command; combined with GJK collision detection, it forms a complete active safety protection system of "soft constraint + hard constraint"; (3) The present invention utilizes VR technology to achieve natural human-computer interaction. Operators can complete fine actions such as screwing, unscrewing and plugging without complex command input, which greatly reduces the operation threshold. Attached Figure Description

[0019] Figure 1 A system functional module partitioning diagram for the five-finger dexterity hand fine teleoperation mapping algorithm provided by the present invention; Figure 2 The flowchart of the filtering and security instruction conversion module provided by this invention; Figure 3 A schematic diagram of GMM workspace modeling provided by the present invention; Figure 4 The schematic diagram of GJK collision detection provided by this invention. Detailed Implementation

[0020] A method for precise teleoperation mapping control of a five-finger dexterous hand is proposed. This method utilizes the Meta Quest 3VR headset and a multi-algorithm optimized mapping control segment. Key point data of the human hand is collected via VR equipment. An adaptive SG filter is used to denoise the collected command signals, filtering out high-frequency noise such as physiological tremors. A kinematic mapping model from key points of the human hand to joint angles of the dexterous hand is established, converting Cartesian space pose into joint space commands. A probabilistic model of the robotic arm's workspace is constructed based on a Gaussian mixture model, and collision detection and safety verification are performed using the GJK algorithm.

[0021] A method for precise telescopic mapping control of a five-finger dexterity hand, the steps of which include: Real-time capture of the operator's key hand position and posture information, and preprocessing of the collected position and posture information; The initial control command is preset, and the preprocessed hand key point position and posture information is smoothed. Establish a kinematic mapping model from capturing the key points of the operator's hand to the joint angles of the dexterous hand and the robotic arm; The preset verification threshold is used to perform a dual verification process for the dexterous hand and robotic arm, including workspace probability verification and collision detection verification, using initial control commands. Once both checks pass, the kinematic mapping model is used to predict the joint angles of the dexterous hand and robotic arm at the next moment, and real-time control commands are generated to the robot controller to realize remote operation control of the dexterous hand and robotic arm.

[0022] The real-time capture of key hand position and posture information is obtained through visual tracking using VR interactive devices. The VR interactive devices capture the operator's key hand position and posture information in real time at a fixed frequency. The preprocessing method is as follows: the collected hand key point positions and posture information are serialized into Cartesian space pose data and sent to the server endpoint of the robot control network in real time through the preset TCP / IP protocol. The server endpoint converts the received serialized data message into a standard format message of the robot control network, which is used as the preprocessed hand key point positions and posture information.

[0023] The method for smoothing the hand key point positions and pose information obtained after preprocessing is as follows: An SG filter is used to denoise the initial control command, which is achieved through a filtering calculation formula. The initial control command is a command signal contaminated by noise, represented as follows: ,in For noise; The filtering calculation formula is based on fitting an nth-order polynomial to a length of using the least squares method. The input sample set within the sliding window is derived, and its form is a linear combination of the input signals:

[0024] In the formula, The initial control command is given at the current time. Sampled values ​​within the center window , To fix the impulse response coefficient, the optimization problem is... Select the optimal window length so that the response bandwidth of the SG filter is greater than a certain value, so as to retain the signal waveform characteristics while filtering out jitter.

[0025] The objective of the SG filter is set as minimizing the cost function used to solve for the fixed impulse response coefficients. The cost function is:

[0026] In the formula, The coefficients of the fitted polynomial, Let be the order of the polynomial. Through... Differentiate and set it to zero, solve the least squares problem, and obtain the optimal polynomial coefficients. Take the center point of the window Fitted values Substituting these values ​​into the filtering formula yields the fixed impulse response coefficients. .

[0027] The kinematic mapping model establishes a forward kinematic model based on the relevant parameters of the dexterous hand and the robotic arm, and then uses the LM iterative algorithm to solve the inverse kinematic model. The iterative algorithm is as follows:

[0028] In the formula, This is the joint angle vector for the next moment. This is the current joint angle vector. For Jacobian matrices, Let be the end-effector pose error function. is the damping coefficient.

[0029] End-effector pose error function In conjunction with the design of an adaptive damping adjustment strategy, when hour, To ensure rapid convergence; when the error hour, Halving for convergence control, end-effector pose error function By employing an adaptive damping adjustment strategy, a kinematic mapping model, and an LM iterative algorithm, the relative positions of key hand points in a dexterous hand are directly mapped to the driving angles of each finger joint to achieve five-finger dexterous hand reproduction control.

[0030] In the kinematic mapping model, the homogeneous transformation matrix of adjacent links in the robotic arm is determined using the DH method, the forward kinematic model of the robotic arm is established, and the inverse kinematic model is solved in conjunction with the LM iterative algorithm.

[0031] No. The link to the first Homogeneous transformation matrix of the links for:

[0032] In the formula, Let be the rotation angle about the z-axis, which describes the first... The link to the first The rotation of a link around the z-axis is the joint angle in the DH parameters. Let be the translation distance around the z-axis, representing the th . The link to the first The amount of movement of a link along the z-axis is the link offset in the DH parameters. Let be the translation distance along the x-axis, representing the th . The link to the first The length of each link along the x-axis is the link length in the DH parameters. Let be the rotation angle about the x-axis, describing the first... The link to the first The rotation of a link about the x-axis is the link twist angle in the DH parameters. Indicates rotation about the z-axis The rotation transformation matrix of the angle. Indicates translation about the z-axis The translation matrix of the distance. Indicates translation along the x-axis The translation matrix of the distance. Indicates rotation about the x-axis The rotation transformation matrix of the angle.

[0033] Both workspace probability verification and collision detection verification are performed before real-time control commands are issued, through initial control commands. The method of the dual verification process is as follows:

[0034] In the formula, For workspace, The end-effector pose vector. and Let be the mean and covariance of the k-th Gaussian distribution, respectively. In the workspace probability verification process, when the calculated probability density value of the current instruction pose... Below the verification threshold If so, the initial control command is determined to be outside the safe working space and is truncated; The collision detection verification process is used to analyze the minimum distance between the convex hulls of each link of the robot body. By calculating the difference set of the convex hulls of adjacent links, the collision detection is converted into a distance from the origin using the calculated difference set to determine whether the minimum distance is included in the difference set. If the minimum distance between two link convex hulls is 0, a collision is determined to have occurred, and the collision detection verification is considered to have failed, triggering the robot's emergency stop protection or obstacle avoidance strategy.

[0035] Once both checks pass, the real-time control command is updated according to the preset control strategy, and the joint angle vector for the next moment obtained through the kinematic mapping model is sent to the robot controller for execution via the communication interface; if either check fails, the current control command is immediately blocked and a safety warning is triggered to prevent mechanical damage to the robot. The dexterous hand and robotic arm control motion parameters, planning filters, and safety boundaries are all visualized through a 3D display user interface.

[0036] The following description, in conjunction with the accompanying drawings and preferred embodiments, provides further details: In the current embodiment, the specific steps of the five-finger dexterity hand fine teleoperation mapping control algorithm are as follows: (1) Step S1: Data Acquisition and Preprocessing. Using the visual tracking function of a VR interactive device (such as the Meta Quest 3 headset), the operator's hand key point positions and pose information are captured in real time at a frequency of 90Hz. The system uses the Unity development environment to communicate with the ROS2 robot operating system. Through the ROS TCP Connector component, the acquired hand Cartesian space pose data is serialized and sent in real time to the server endpoint in the ROS2 network via the TCP / IP protocol. The server endpoint deserializes the received TCP messages into the ROS2 standard message format (such as geometry_msgs / PoseStamped), completing the data transmission from the virtual environment to the robot control system.

[0037] (2) Step S2: Command denoising based on SG filtering. To address the 6-15Hz physiological tremors and sensor noise generated by the human hand during fine manipulation, a Savitzky-Golay (SG) filter is used to smooth the acquired raw pose commands. The core idea of ​​SG filtering is to use a polynomial to perform least-squares fitting on the data within the sliding window in the time domain. Assume the noise-contaminated command signal is... ,in For noise. Consider an nth-order polynomial and a length of For a symmetric window, the filter aims to minimize the following cost function:

[0038] In the formula, The coefficients of the fitted polynomial, Let be the order of the polynomial. Through... Differentiate and set it to zero, solve the least squares problem, and obtain the optimal polynomial coefficients. .

[0039] Filter output Equivalent to a weighted average of samples within the window:

[0040] in, The fixed impulse response coefficients of the filter depend only on the order n and the window length. Find the center point of the window. Fitted values The fixed impulse response coefficient can then be obtained. This invention addresses the problem by optimizing... The optimal window length is determined so that the filter's response bandwidth is greater than 20Hz. Compared with traditional low-pass filters, this method effectively filters out high-frequency jitter while preserving the signal's waveform characteristics (such as peak times) and significantly reducing phase delay.

[0041] (3) Step S3: Kinematic mapping. Establish a kinematic mapping model from key points of the human hand to the joint angles of the dexterous hand and the robotic arm. First, establish the forward kinematic model of the robotic arm using the DH parameter method. The homogeneous transformation matrix of adjacent links is:

[0042] Then, the LM iterative algorithm is used to solve the inverse kinematics equations and calculate the angles of each joint of the robotic arm that satisfy the target pose. The iterative formula is as follows:

[0043] in, This is the current joint angle vector. For Jacobian matrices, Let be the end-effector pose error function. Let be the damping coefficient. This invention designs an adaptive damping adjustment strategy: when the error... hour, To ensure rapid convergence; when the error hour, Halving is used to achieve fine convergence. For five-finger dexterity hands, the relative positions of key points on the hand are directly mapped to the driving angles of each finger joint, enabling accurate reproduction of gestures.

[0044] (4) Step S4: Safety verification. Before the instruction is issued, a dual safety assessment of workspace probability verification and collision detection is performed.

[0045]

[0046] in, For workspace, The end-effector pose vector. and These are the mean and covariance of the k-th Gaussian distribution, respectively. During online verification, the probability density value of the current command pose is calculated; if it is lower than a preset threshold... (For example If the command exceeds the safe working space, it is determined to be interrupted. Real-time collision detection based on GJK: The minimum distance between the convex hulls of each link in the robot body is calculated using the GJK (Gilbert-Johnson-Keerthi) algorithm. The algorithm calculates the Minkowski difference between two convex bodies A and B. This transforms collision detection into determining whether the origin is contained within C. If the minimum distance between two convex bodies... If a collision is detected, the system will immediately trigger an emergency stop protection or obstacle avoidance strategy.

[0047] (5) Step S5: Instruction execution. The control strategy is executed according to the verification result of step S4: if the instruction passes both the workspace verification and collision detection, the calculated joint angle instruction is sent to the robot controller for execution through the communication interface; if either verification fails, the instruction is immediately blocked and a safety warning signal is triggered to prevent mechanical damage to the robot.

[0048] The data acquisition and preprocessing process is as follows: Figure 2 The motion capture portion shown in this embodiment uses the MetaQuest 3 VR headset as the input device. The system establishes communication with the ROS2 system through the ROS TCP Connector component in the Unity development environment. The Unity client acquires the Cartesian space pose (position) of the hand key points in real time. and quaternion pose After serialization, it is sent to the server endpoint via TCP / IP protocol, with the sampling frequency set to 90Hz to ensure high dynamic response of the operation.

[0049] Instruction denoising based on SG filtering, such as Figure 2 As shown, corresponding to the "SG filtering" section, this embodiment uses a Savitzky-Golay (SG) filter to smooth the original command, specifically addressing the 6-15Hz physiological tremor present in the human hand. Figure 2 As shown, the SG filtering module receives the original command and uses the least squares method to fit an nth-order polynomial to a length of... The input sample set within the sliding window. The specific filtering calculation formula is:

[0050] The initial control command is given at the current time. Sampled values ​​within the center window , To fix the impulse response coefficient, in this embodiment, an optimization problem is solved. The optimal window length is selected so that the filter's response bandwidth is greater than 20Hz (actually measured to reach 60Hz). This step filters out jitter while preserving the signal's waveform characteristics to the greatest extent, avoiding the phase delay caused by traditional low-pass filtering. In the kinematic mapping model design process, such as... Figure 2 The "accessible pose sampling and mapping" section of the robotic arm, as shown, establishes a forward kinematics model based on the robotic arm's Denavit-Hartenberg (DH) parameters and solves the inverse kinematics using the Levenberg-Marquardt (LM) iterative algorithm. The iterative formula is as follows:

[0051] in, This is the current joint angle vector. For Jacobian matrices, Let be the end-effector pose error function. Let be the damping coefficient. This invention designs an adaptive damping adjustment strategy: when the error... hour, To ensure rapid convergence; when the error hour, Halving is used to achieve fine convergence. For five-finger dexterity hands, the relative positions of key points on the hand are directly mapped to the driving angles of each finger joint, enabling accurate reproduction of gestures.

[0052] Security verification such as Figure 2 The "Feasibility Analysis" section shows that before the instruction is issued, the system performs a dual security check: workspace probability check: such as... Figure 3 The diagram shows a Gaussian Mixture Model (GMM) probabilistic model of the robotic arm's workspace. It illustrates the reachability probability distribution of the robotic arm's end effector in three-dimensional space (warmer colors represent higher probability density). In this embodiment, the robotic arm was pre-sampled offline (approximately 1 million samples), and a Gaussian Mixture Model (GMM) was trained using the EM algorithm. During online execution, the current command pose was calculated. probability density:

[0053] like Below the preset threshold (In this embodiment, it is set to) If the instruction exceeds the safe workspace, it is determined that the instruction is outside the safe workspace. Figure 3 The system will refuse to execute the instruction within the area indicated by the outer contour. Real-time collision detection: such as... Figure 4The diagram shown illustrates the principle of collision detection based on the GJK algorithm. It demonstrates the iterative search using a simplex algorithm to find the minimum distance between two convex bodies (robot links). This embodiment calculates the Minkowski difference set of the convex hulls of each robot component. Determine whether the origin is contained within In the middle. For example Figure 4 As shown in (f), if the minimum distance calculated iteratively... If the value is zero, it indicates a collision has occurred. This algorithm can monitor its components in real time with extremely high efficiency (milliseconds) to prevent self-collisions.

[0054] In the command execution and display design, after verification, joint commands are sent to the robot controller. Simultaneously, if... Figure 1 The "3D command display module" shown will render the robot's motion status, planned path and safety boundary (red and blue areas) in real time on the user interface, realizing a "what you see is what you get" teleoperation experience.

[0055] The robot body consists of a dexterous hand and a robotic arm. The inertial positioning system calculation module is used for the acquisition and calculation of pose data. The instruction 3D display module is used to realize user control and display. The instruction verification and protection module is used to perform dual verification. The instruction fine calculation module is used to realize the joint execution path planning of the robotic arm and dexterous hand. The filtering and safety instruction conversion module is used to realize the remote operation motion capture control by real-time control instructions that determine the initial control instructions and the joint angle vector at the next moment.

[0056] Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make possible changes and modifications to the technical solutions of the present invention by utilizing the methods and techniques disclosed above without departing from the spirit and scope of the present invention. Therefore, any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solutions of the present invention shall fall within the protection scope of the technical solutions of the present invention.

[0057] The contents not described in detail in this specification are common knowledge to those skilled in the art.

Claims

1. A method for precise teleoperation mapping control of a five-finger dexterous hand, characterized in that... include: Real-time capture of the operator's key hand position and posture information, and preprocessing of the collected position and posture information; The initial control command is preset, and the preprocessed hand key point position and posture information is smoothed. Establish a kinematic mapping model from capturing the key points of the operator's hand to the joint angles of the dexterous hand and the robotic arm; The preset verification threshold is used to perform a dual verification process for the dexterous hand and robotic arm, including workspace probability verification and collision detection verification, using initial control commands. Once both checks pass, the kinematic mapping model is used to predict the joint angles of the dexterous hand and robotic arm at the next moment, and real-time control commands are generated to the robot controller to realize remote operation control of the dexterous hand and robotic arm.

2. The method for fine teleoperation mapping control of a five-finger dexterous hand according to claim 1, characterized in that: The real-time capture of the key hand position and posture information is obtained by visual tracking using a VR interactive device. The VR interactive device captures the operator's key hand position and posture information in real time at a fixed frequency. The preprocessing method is as follows: the collected hand key point positions and posture information are serialized into Cartesian space pose data and sent to the server endpoint of the robot control network in real time through the preset TCP / IP protocol. The server endpoint converts the received serialized data message into a standard format message of the robot control network, which is used as the preprocessed hand key point positions and posture information.

3. The method for fine teleoperation mapping control of a five-finger dexterous hand according to claim 1, characterized in that: The method for smoothing the hand key point positions and posture information obtained after preprocessing is as follows: An SG filter is used to denoise the initial control command, which is achieved through the filter calculation formula. The initial control command is a noise-contaminated command signal, represented as follows: ,in For noise; The filtering calculation formula is based on fitting an nth-order polynomial to a length of using the least squares method. The input sample set within the sliding window is derived, and its form is a linear combination of the input signals: In the formula, The initial control command is given at the current time. Sampled values ​​within the center window , To fix the impulse response coefficient; by optimizing the problem Select the optimal window length so that the response bandwidth of the SG filter is greater than a certain value, so as to retain the signal waveform characteristics while filtering out jitter.

4. The method for fine teleoperation mapping control of a five-finger dexterous hand according to claim 3, characterized in that: The objective of the SG filter is set as minimizing the cost function used to solve for the fixed impulse response coefficients. The cost function is: In the formula, The coefficients of the fitted polynomial, The order of the polynomial; by... Differentiate and set it to zero, solve the least squares problem, and obtain the optimal polynomial coefficients. Take the center point of the window Fitted values And substitute them into the filtering formula to obtain the fixed impulse response coefficient. .

5. The method for fine teleoperation mapping control of a five-finger dexterous hand according to claim 3, characterized in that: The kinematic mapping model establishes a forward kinematic model based on the relevant parameters of the dexterous hand and the robotic arm, and uses the Levenberg-Marquardt iterative algorithm to solve the inverse kinematic model. The iterative algorithm is as follows: In the formula, This is the joint angle vector for the next moment. This is the current joint angle vector. For Jacobian matrices, Let be the end-effector pose error function. is the damping coefficient.

6. The method for fine teleoperation mapping control of a five-finger dexterous hand according to claim 5, characterized in that: The end pose error function In conjunction with the design of an adaptive damping adjustment strategy, when hour, To ensure rapid convergence; when the error hour, Halving for convergence control, end-effector pose error function By employing an adaptive damping adjustment strategy, a kinematic mapping model, and an LM iterative algorithm, the relative positions of key hand points in a dexterous hand are directly mapped to the driving angles of each finger joint to achieve five-finger dexterous hand reproduction control.

7. The method for fine teleoperation mapping control of a five-finger dexterous hand according to claim 5, characterized in that: In the kinematic mapping model, the Denavit-Hartenberg method is used to determine the homogeneous transformation matrix of adjacent links in the robotic arm, establishing the forward kinematic model of the robotic arm, and the LM iterative algorithm is used to solve the inverse kinematic model; The link to the first Homogeneous transformation matrix of the links for: In the formula, Let be the rotation angle about the z-axis, used to describe the first... The link to the first The rotation of each link around the z-axis is represented by the joint angle in the DH parameters; Let be the translation distance around the z-axis, and be the th . The link to the first The amount of movement of each link along the z-axis is the link offset in the DH parameters; Let be the translation distance along the x-axis, and be the th . The link to the first The length of each link along the x-axis is the link length in the DH parameters; Let be the rotation angle about the x-axis, used to describe the first... The link to the first The rotation of the link about the x-axis is the link twist angle in the DH parameters; Indicates rotation about the z-axis The rotation transformation matrix of the angle. Indicates translation about the z-axis The translation matrix of the distance. Indicates translation along the x-axis The translation matrix of the distance. Indicates rotation about the x-axis The rotation transformation matrix of the angle.

8. A method for fine teleoperation mapping control of a five-finger dexterous hand according to claim 5, characterized in that: The workspace probability verification and collision detection verification are both implemented through initial control commands before the real-time control commands are issued. The method of the dual verification process is as follows: In the formula, For workspace, The end-effector pose vector. and denoted as the mean and covariance of the k-th Gaussian distribution, respectively.

9. A method for fine teleoperation mapping control of a five-finger dexterous hand according to claim 8, characterized in that: In the workspace probability verification process, when the calculated probability density value of the current instruction pose... Below the verification threshold If so, the initial control command is determined to be outside the safe working space and is truncated; The collision detection verification process is used to analyze the minimum distance between the convex hulls of each link of the robot body. By calculating the difference set of the convex hulls of adjacent links, the collision detection is converted into a distance from the origin using the calculated difference set to determine whether the minimum distance is included in the difference set. If the minimum distance between two link convex hulls is 0, a collision is determined to have occurred, and the collision detection verification is considered to have failed, triggering the robot's emergency stop protection or obstacle avoidance strategy.

10. A method for fine teleoperation mapping control of a five-finger dexterous hand according to claim 9, characterized in that: Once both checks pass, the real-time control command is updated according to the preset control strategy, and the joint angle vector for the next moment obtained through the kinematic mapping model is sent to the robot controller for execution via the communication interface; if either check fails, the current control command is immediately blocked and a safety warning is triggered to prevent mechanical damage to the robot. The dexterous hand and robotic arm control motion parameters, planning filters, and safety boundaries are all visualized through a 3D display user interface.

Images