A method and system for generating acupuncture hand posture based on energy optimization

By constructing a three-dimensional geometric model of the acupuncture hand-pressing robot and using energy optimization methods, a stable and comfortable hand-pressing posture is generated, which solves the problem that the acupuncture robot hand-pressing robot cannot adapt to different patient body shapes and acupoint locations, and improves the automation and comfort of the hand-pressing posture.

CN123103641APending Publication Date: 2026-11-17GUANGZHOU HOSPITAL OF TRADITIONAL CHINESE MEDICINE +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611295883.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-25
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing acupuncture robot hand control methods cannot adapt to the differences in body shape and acupoint location among different patients, and cannot adjust posture in real time, causing patient discomfort. Furthermore, existing energy optimization methods have not effectively solved the problem of comfortable hand pressure.

Method used

A three-dimensional geometric model of an acupuncture hand robot is constructed, and the hand posture representation is defined. By combining contact energy, comfort energy, stability energy, and accessibility energy, the optimal hand posture is generated through a gradient descent optimization algorithm. Joint limit and joint stress energy terms are introduced to ensure the stability and comfort of the hand posture.

Benefits of technology

It enables the automatic generation of comfortable and stable hand-pressing postures based on the patient's acupoint location and body shape characteristics, improving the patient's comfort and adaptability of the hand-pressing posture, avoiding the one-size-fits-all problem in traditional methods, and reducing discomfort to the human body.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123103641A_ABST
    Figure CN123103641A_ABST
Patent Text Reader

Abstract

This invention discloses an energy-optimized method and system for generating acupuncture hand postures. The method includes: constructing a three-dimensional geometric model of an acupuncture hand robot and defining a hand posture representation; constructing a multi-dimensional objective hand posture energy optimization function by combining contact energy, comfort energy, stability energy, and reachability energy; converting the multi-dimensional objective hand posture energy optimization function into an energy minimization extremum constraint problem; and solving the problem using a gradient descent optimization algorithm, combining the three-dimensional geometric model of the acupuncture hand robot and the hand posture representation, to obtain the optimal acupuncture hand robot posture. This invention can automatically generate stable hand postures based on the patient's acupoint location and body shape characteristics. As an energy-optimized method and system for generating acupuncture hand postures, this invention can be widely applied in the field of robot posture planning technology.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of robot posture planning technology, and in particular to a method and system for generating acupuncture hand postures based on energy optimization. Background Technology

[0002] Acupuncture is an important component of traditional Chinese medicine, and "pressing hand" is one of the key techniques in acupuncture operation. The pressing hand refers to the hand used to press the skin around the acupoint, assisting in positioning and needle insertion during the needle insertion process. However, in traditional acupuncture, the pressing hand is usually operated manually by experienced acupuncturists, requiring extensive training to master. With the development of acupuncture robots, how to enable robots to automatically generate reasonable pressing hand postures has become a pressing technical problem. Existing acupuncture robot research mainly focuses on the control and planning of the needle-holding hand, with relatively little research on the pressing hand. Existing pressing hand control methods have the following shortcomings, for example: 1) Pre-set fixed postures cannot adapt to the differences in body shape and acupoint location among different patients; 2) The pressure and distribution of the hand pressure were not taken into account, which may easily cause discomfort to the patient; 3) Unable to adjust the hand's posture based on real-time feedback; 4) Training methods designed for specific body types are difficult to transfer to other patients.

[0003] Furthermore, significant progress has been made in energy-optimized grasping generation methods within the field of robotic dexterous hand grasping. These methods assess grasping quality by designing an energy function and search for the optimal grasping posture using gradient optimization methods. However, directly applying such methods to acupuncture hand manipulation presents several problems, including: 1) The goal of grasping is to stably grasp the object, while the goal of pressing is to comfortably press on the skin; 2) Grabbing usually requires wrapping the object, while the hand only needs to make point contact or area contact; 3) Grabbing requires resisting external force, and the pressing hand needs to provide appropriate pressure. Summary of the Invention

[0004] To address one of the aforementioned technical problems, the present invention aims to provide a method and system for generating acupuncture hand postures based on energy optimization, which can automatically generate stable hand postures according to the patient's acupoint location and body shape characteristics.

[0005] The first technical solution adopted in this invention is: a method for generating acupuncture hand posture based on energy optimization, comprising the following steps: Construct a three-dimensional geometric model of the acupuncture hand robot and define the hand posture representation; By combining contact energy, comfort energy, stability energy, and accessibility energy, a multi-dimensional target hand gesture energy optimization function is constructed. The multi-dimensional objective hand pose energy optimization function is transformed into an energy minimization extreme value constraint problem. Combining the three-dimensional geometric model of the acupuncture hand robot with the hand pose representation, the optimal acupuncture hand robot pose is obtained by solving the problem using a gradient descent optimization algorithm.

[0006] Furthermore, the three-dimensional geometric model of the acupuncture hand-pressing robot has a multi-joint linkage mechanism. The hand-pressing posture representation includes the palm pose, joint angles, and the total parameter dimensions of the acupuncture hand-pressing robot. The multi-joint linkage mechanism specifically includes palm joints, finger joints, and joint angle vectors, wherein: The palm joint serves as a reference rigid body, defining the center position and orientation of the palm. Each of the five finger joints includes a metacarpophalangeal joint, a proximal interphalangeal joint, and a distal interphalangeal joint.

[0007] Furthermore, the multi-dimensional target hand gesture energy optimization function specifically includes contact energy, stability energy, comfort energy, accessibility energy, and effectiveness energy, wherein: The contact energy is used to measure the degree of contact between the pressing hand and the target acupoint area. The contact energy includes positional contact energy and force contact energy. The stability energy is used to evaluate the stability and anti-interference ability of the presser's posture. The stability energy includes force containment energy and contact force balance energy. The force containment energy is used to evaluate whether the presser can resist external force interference in any direction. The contact force balance energy is used to ensure that the forces at each contact point are in a balanced state. The comfort energy is used to assess the impact of hand posture on the human body. The comfort energy includes joint limit energy, joint stress energy, and finger spread energy. The joint limit energy is used to penalize postures that exceed the joint range of motion. The joint stress energy is used to penalize postures that exceed the preset requirement for joint flexion. The finger spread energy is used to penalize postures that exceed the preset requirement for finger spread. The reachability energy is used to ensure that the generated posture is within the range of motion of the hand. The reachability energy includes joint range energy and palm reachability energy. The joint range energy is used to ensure that all joint angles are within the reachable range, and the palm reachability energy is used to ensure that the palm can reach the target position. The effective energy is used to eliminate physically infeasible or uncomfortable postures. The effective energy includes self-collision energy and finger gap energy. The self-collision energy is used to punish self-collision of the pressing hand, and the finger gap energy is used to ensure that an appropriate gap is maintained between adjacent fingers.

[0008] Furthermore, the step of transforming the multi-dimensional target hand posture energy optimization function into an energy minimization extremum constraint problem, and then solving it using a gradient descent optimization algorithm based on the three-dimensional geometric model and hand posture representation of the acupuncture hand robot to obtain the optimal acupuncture hand robot posture, specifically includes: Combining the three-dimensional geometric model and hand posture representation of the acupuncture hand robot, the multi-dimensional objective hand posture energy optimization function is transformed into an energy minimization extreme value constraint problem; Based on the gradient calculation method of the automatic differentiation framework, the energy minimization extreme value constraint problem is optimized and solved to obtain the optimal posture of the acupuncture hand robot.

[0009] Furthermore, the step of combining the three-dimensional geometric model of the acupuncture hand-pressing robot with the hand-pressing posture representation to transform the multi-dimensional objective hand-pressing posture energy optimization function into an energy minimization extremum constraint problem specifically includes: The three-dimensional geometric model and hand posture representation of the acupuncture hand-pressing robot are converted into palm pose parameters and finger joint angle vectors, and parameterized decision variables are defined. The physical and physiological hard constraints in the multi-dimensional target hand posture energy optimization function are transformed into differentiable penalty terms and weighted summation is performed to construct the total energy function in the form of soft constraints. Substituting the parameterized decision variables into the total energy function in the form of soft constraints yields the energy minimization extreme value constraint problem.

[0010] Furthermore, the step of transforming the physical and physiological hard constraints in the multi-dimensional target bettor posture energy optimization function into differentiable penalty terms and performing weighted summation to construct the total energy function in the form of soft constraints specifically includes: Based on the contact energy in the multi-dimensional target pressure hand posture energy optimization function, the sum of squared distances between the fingertip contact point and the target acupoint is minimized to drive the fingertip to align with the acupoint, thus constructing a contact energy constraint term; Based on the stability energy in the multi-dimensional target presser posture energy optimization function, and based on a differentiable force closed approximation estimator, the pressing stability of the posture is quantified, and a stability energy constraint term is constructed. Based on the comfort energy in the multi-dimensional target hand posture energy optimization function, the hard boundary of the joint angle is transformed into a funnel-shaped penalty function to construct the joint limit energy constraint term; Based on the reachability energy and effectiveness energy in the multi-dimensional target bettor posture energy optimization function, differentiable self-collision and skin penetration detection are realized by point cloud distance calculation. When penetration occurs, a penalty is applied, and a penetration penalty energy constraint term is constructed. By combining the contact energy constraint, stability energy constraint, joint limit energy constraint, and penetration penalty energy constraint, a weighted sum is performed to construct the total energy function in the form of soft constraints.

[0011] Furthermore, the gradient calculation method based on the automatic differentiation framework optimizes the solution of the energy minimization extremum constraint problem to obtain the optimal posture of the acupuncture hand robot. This step specifically includes: By combining parameterized decision variables with the total energy function in the form of soft constraints, a differentiable computational graph is constructed. The framework's automatic differentiation interface is invoked, and the algorithm propagates backward along the differentiable computation graph and uses the chain rule to calculate the precise gradient vector of the total energy with respect to the bettor's posture parameters. By processing the precise gradient vector through non-differentiable manipulation and engineering optimization, the optimal posture of the acupuncture hand robot is obtained.

[0012] Furthermore, the step of constructing a differentiable computational graph by combining parameterized decision variables with the total energy function in the form of soft constraints specifically includes: Convert parameterized decision variables into differentiable tensors; By mapping differentiable tensors to the spatial poses of each link in the finger and the three-dimensional coordinates of the contact points through forward kinematics, a forward kinematics computation chain is constructed. The total energy function in the form of soft constraints is transformed into a differentiable computation function, which is then substituted into the forward kinematics computation chain for weighted summation to construct a differentiable computation graph.

[0013] Furthermore, the step of processing the precise gradient vector for non-differentiable manipulation and engineering optimization to obtain the optimal acupuncture hand robot posture specifically includes: Discrete collision detection was performed on the acupuncture hand-pressing robot. Differentiable approximation was performed using the signed distance calculation of point clouds in the Kaolin library to obtain the discrete collision detection results that are differentiable approximations. By leveraging the batch parallel capabilities of GPUs, a computational graph of multiple candidate poses is constructed. Combined with the discrete collision detection results that are differentiable approximations, energy and gradients are calculated in parallel. Starting from the initial posture, the posture parameters are iteratively updated along the negative gradient direction until the total energy converges, and the optimal acupuncture hand robot posture is output.

[0014] The second technical solution adopted in this invention is: an energy-optimized acupuncture hand posture generation system, comprising: The first module is used to construct the three-dimensional geometric model of the acupuncture hand robot and define the hand posture representation; The second module is used to combine contact energy, comfort energy, stability energy and accessibility energy to construct a multi-dimensional target hand gesture energy optimization function; The third module is used to transform the multi-dimensional target hand posture energy optimization function into an energy minimization extreme value constraint problem. Combining the three-dimensional geometric model of the acupuncture hand robot with the hand posture representation, the optimal acupuncture hand robot posture is obtained by solving the problem through a gradient descent optimization algorithm.

[0015] The beneficial effects of the method and system of this invention are as follows: This invention constructs a three-dimensional geometric model of an acupuncture hand-pressing robot and defines the hand-pressing posture representation; further, it combines contact energy, comfort energy, stability energy, and accessibility energy to construct a multi-dimensional objective hand-pressing posture energy optimization function. Through force closure and contact force balance terms in the energy function, it ensures that the hand-pressing robot provides appropriate and evenly distributed pressing pressure, improving patient comfort. It integrates contact accuracy, mechanical safety, ergonomic comfort, and physical feasibility into the optimization framework, fully demonstrating the leap from simple physical operation to natural human-computer interaction. Finally, the multi-dimensional objective hand-pressing posture energy optimization function is transformed into an energy minimization extreme value constraint problem. Combining the three-dimensional geometric model of the acupuncture hand-pressing robot and the hand-pressing posture representation, it is solved using a gradient descent optimization algorithm to obtain the optimal acupuncture hand-pressing robot posture. Ergonomic energy terms such as joint limits and joint stress are introduced to avoid discomfort caused to the human body by the hand-pressing posture. It can automatically generate a comfortable and stable hand-pressing posture based on the patient's acupoint location and body shape characteristics. Attached Figure Description

[0016] Figure 1 This is a flowchart of the steps of an energy-optimized acupuncture hand posture generation method according to the present invention; Figure 2 This is a structural block diagram of an acupuncture hand posture generation system based on energy optimization according to the present invention; Figure 3 This is a schematic diagram of the acupuncture hand posture generation process provided in a specific embodiment of the present invention. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. The step numbers in the following embodiments are only for ease of explanation and do not limit the order of the steps. The execution order of each step in the embodiments can be adapted according to the understanding of those skilled in the art.

[0018] Reference Figure 1 This invention provides a method for generating acupuncture hand gestures based on energy optimization, the method comprising the following steps: S100. Construct a three-dimensional geometric model of the acupuncture hand robot and define the hand posture representation; Specifically, the three-dimensional geometric model of the acupuncture hand-pressing robot has a multi-joint linkage mechanism. The hand-pressing posture representation includes the palm pose, joint angles, and the total parameter dimensions of the acupuncture hand-pressing robot. The multi-joint linkage mechanism specifically includes palm joints, five finger joints, and joint angle vectors. The palm joint serves as a reference rigid body, defining the center position and palm orientation of the palm. Each finger joint in the five finger joints includes a metacarpophalangeal joint, a proximal interphalangeal joint, and a distal interphalangeal joint.

[0019] In this embodiment, the press handle structure is parameterized, and the press handle is modeled as a multi-joint linkage mechanism, specifically including: 1) Hand: As a reference rigid body, define the center position of the hand. and the direction of the palm (using quaternions) express).

[0020] 2) Five fingers: Each finger contains three joints: MCP (metacarpophalangeal joint), PIP (proximal interphalangeal joint), and DIP (distal interphalangeal joint), for a total of 15 joint angles.

[0021] 3) Joint angle vector: .

[0022] Next is the hand gesture representation, which is determined by the following parameters, specifically including: 1) Hand position: ( , ), a total of 7 dimensions.

[0023] 2) Joint angle: There are 15 dimensions in total.

[0024] 3) Total parameter dimensions: 22 dimensions.

[0025] S200, combining contact energy, comfort energy, stability energy and accessibility energy, constructs a multi-dimensional target hand gesture energy optimization function; First, it needs to be clarified that the human-computer interaction task of acupuncture hand pressure requires "comfortable and safe pressing of acupoints." This shift in objectives directly determines the design logic of the energy function, requiring the palm and fingers to adhere to the human body surface, applying effective pressure while avoiding tissue damage or pain reflexes. Therefore, the hand posture energy function designed in this invention... It consists of a weighted average of multiple energy terms. Contact energy needs to be considered. ) and reachability energy ( To ensure that the fingertips accurately reach acupoints while remaining within the robotic arm's workspace, this invention also innovatively introduces several energy items for safe human-machine interaction. Comfort energy ( The system consists of three parts: joint limit energy, joint stress energy, and finger extension energy. Joint limit energy establishes a threshold penalty based on the physiological range of motion of each finger joint to prevent posture from exceeding anatomical limits. Joint stress energy applies increasing weights according to the degree of joint flexion to inhibit tissue compression caused by excessive flexion. Finger extension energy, based on a natural relaxed state, penalizes excessive finger abduction to avoid unnecessary traction on the skin around acupoints. Stability energy (…) This system draws upon the concept of force closure and combines it with pressure balance conditions. It assesses the presser's ability to resist external disturbances through force closure energy, and then constrains the sum of the resultant force and external force at each contact point to approach zero through contact force balance energy. This ensures uniform pressing pressure, stable force application, and avoids excessive local pressure. Finally, the effective energy ( By employing self-collision detection and finger gap penalty, invalid postures that are physically impossible or may pinch the skin are eliminated from the ground up. This energy function system integrates contact accuracy, mechanical safety, ergonomic comfort, and physical feasibility into an optimization framework, fully demonstrating the leap from simple physical operation to natural human-computer interaction.

[0026] The multi-dimensional target hand gesture energy optimization function specifically includes contact energy, stability energy, comfort energy, accessibility energy, and effectiveness energy.

[0027] In this embodiment, the presser posture energy function designed according to the present invention... It consists of a weighted average of multiple energy terms, and its expression is:

[0028] In the above formula, In order to access energy, For stable energy, For comfort energy, For reachable energy, For effective energy.

[0029] More specifically, the contact energy is used to measure the degree of contact between the pressing hand and the target acupoint area, and the contact energy includes positional contact energy and force contact energy; In this embodiment, contact energy is used to measure the degree of contact between the pressing hand and the target acupoint area, and its expression is:

[0030] in For position contact energy, For force to contact energy.

[0031] For position contact energy, its expression is:

[0032] in Let i be the position of the i-th contact point. The target acupoint location, This represents the number of contact points.

[0033] The expression for contact energy is:

[0034] in The actual force applied (which can be estimated using a force sensor or mechanical model). The pressure applied to the target.

[0035] The stability energy is used to evaluate the stability and anti-interference ability of the presser's posture. The stability energy includes force containment energy and contact force balance energy. The force containment energy is used to evaluate whether the presser can resist external force interference in any direction. The contact force balance energy is used to ensure that the forces at each contact point are in a balanced state. In this embodiment, stability energy is used to evaluate the stability and anti-interference capability of the presser's posture, referencing the closed-loop concept, and its expression is:

[0036] Among them, the expression for the force-closed energy is:

[0037] Assess whether the bettor can resist external forces from any direction.

[0038] The expression for the contact force equilibrium energy is:

[0039] Ensure that the forces at each contact point are in a balanced state.

[0040] The comfort energy is used to assess the impact of hand posture on the human body. The comfort energy includes joint limit energy, joint stress energy, and finger spread energy. The joint limit energy is used to penalize postures that exceed the joint range of motion. The joint stress energy is used to penalize postures that exceed the preset requirement for joint flexion. The finger spread energy is used to penalize postures that exceed the preset requirement for finger spread. In this embodiment, comfort energy is used to evaluate the impact of the hand posture on the human body. Referring to ergonomic principles, its expression is:

[0041] Joint limit energy is used to penalize postures that approach or exceed the joint's range of motion, and its expression is:

[0042] Joint stress energy is used to punish excessive bending of the joint, and its expression is:

[0043] in These are stress-weighted coefficients; the greater the degree of bending, the higher the weight. Finger extension energy is used to punish over-extension of the fingers (the pressing hand usually does not need to be over-extensioned), and its expression is:

[0044] In the above formula, It represents the energy of the fingers spreading out.

[0045] The reachability energy is used to ensure that the generated posture is within the range of motion of the hand. The reachability energy includes joint range energy and palm reachability energy. The joint range energy is used to ensure that all joint angles are within the reachable range, and the palm reachability energy is used to ensure that the palm can reach the target position. In this embodiment, reachability energy is used to ensure that the generated posture is within the range of motion of the hand, and its expression is:

[0046] Joint range energy is used to ensure that all joint angles are within reach, and its expression is:

[0047] The reachable energy of the palm is used to ensure that the palm can reach the target location, and its expression is:

[0048] in This is the maximum distance that the bettor can reach.

[0049] The effective energy is used to eliminate physically infeasible or uncomfortable postures. The effective energy includes self-collision energy and finger gap energy. The self-collision energy is used to punish self-collision of the pressing hand, and the finger gap energy is used to ensure that an appropriate gap is maintained between adjacent fingers.

[0050] In this embodiment, effective energy is used to exclude postures that are physically infeasible or uncomfortable for the human body, and its expression is:

[0051] The self-collision energy is used to penalize the bettor's self-collision, and its expression is:

[0052] Interdigital energy is used to ensure proper spacing between adjacent fingers, and its expression is:

[0053] In the above formula, This represents the energy in the interfinite space.

[0054] S300. The multi-dimensional target hand posture energy optimization function is transformed into an energy minimization extreme value constraint problem. Combining the three-dimensional geometric model of the acupuncture hand robot with the hand posture representation, the optimal acupuncture hand robot posture is obtained by solving the problem through a gradient descent optimization algorithm.

[0055] First, it's important to clarify that formalizing the acupuncture hand posture generation problem into an unconstrained optimization problem involves transforming the search process for the hand posture into finding posture parameters that minimize the preset energy function value. The key to this formalization lies in cleverly designing the energy function to "soften" hard constraints such as fingers not penetrating the skin, joints not exceeding their range of motion, and the palm reaching the target acupoint into differentiable penalty terms. This transforms the originally complex constrained optimization problem into an unconstrained form that is easily solved using methods such as gradient descent.

[0056] S310. Combining the three-dimensional geometric model and hand posture representation of the acupuncture hand robot, the multi-dimensional objective hand posture energy optimization function is transformed into an energy minimization extreme value constraint problem. Specifically, the three-dimensional geometric model and hand posture representation of the acupuncture hand-pressing robot are converted into palm pose parameters and finger joint angle vectors, and parameterized decision variables are defined. The physical and physiological hard constraints in the multi-dimensional objective hand posture energy optimization function are converted into differentiable penalty terms and weighted summation is performed to construct a soft-constraint form of the total energy function. The parameterized decision variables are substituted into the soft-constraint form of the total energy function to obtain the energy minimization extreme value constraint problem.

[0057] In this embodiment, parameterized decision variables are first defined. The "hand gripping posture" needs to be transformed into an explicit mathematical variable. The gripping posture is parameterized as a tuple. In this invention, the hand posture is parameterized into a 22-dimensional vector, including hand pose parameters (…). , ) and finger joint angle vector The goal of the optimization problem is to find an optimal set of pose parameters. .

[0058] Furthermore, an energy function in the form of "soft constraints" is constructed. This is a crucial step in the formalization process; the original physical and physiological constraints are designed as energy terms with penalties. Specifically, contact constraints are constructed as contact energy. By minimizing the sum of squares of the distances between the contact point and the target acupoint, the fingertip is driven to automatically move to the correct position; the joint limit constraint is constructed as the joint limit energy, which sets the hard boundary conditions. The function is transformed into a funnel-shaped penalty function, where the energy is zero when the angle is within the feasible range, and increases quadratically when it exceeds the range; the penetration constraint is constructed as the effectiveness energy. The self-collision penalty term in the algorithm applies a large penalty value once a finger is detected penalizing the skin or colliding with itself, forcing the optimizer to actively avoid infeasible poses. Furthermore, this embodiment uses a differentiable force closure approximation estimator, allowing force closure—a complex physical condition that would otherwise only be verified later—to be directly optimized by gradient descent. The differentiable force closure estimator is a mathematical method that transforms the discrete, non-differentiable "force closure" physical grasping index into a continuous, differentiable geometric energy term. It constructs a "grasping matrix" by sampling several contact points on the surface of the dexterous hand and combining it with the surface normal, and then calculates the sum of the resultant force and resultant torque generated on the object when the fingers collaboratively apply a unit internal force.

[0059] Scraping Matrix The expression is as follows:

[0060] in The identity matrix is ​​used to map the resultant force at each contact point. The coordinates of the contact point The antisymmetric matrix is ​​used to calculate the resultant torque generated by the contact force about the center of the object.

[0061] For the three-dimensional coordinates of a certain contact point in space Its antisymmetric matrix form is:

[0062] By crawling matrix With contact normal vector Multiplying the (unit normal force applied at each point) together, a differentiable closed-loop energy function is constructed. When this energy term approaches zero during optimization, it means that the force applied by all fingers can physically achieve perfect internal force balance, thus forming a stable, closed grip that can resist external forces from any direction. Similarly, the stability energy in the pressing hand energy function... This can be used as the optimization objective for the differentiable estimator.

[0063] Finally, it is formalized into an unconstrained optimization problem. When all constraints are integrated into the total energy function... The problem then transforms into an unconstrained minimization problem:

[0064] in This represents the feasible attitude space. It is a weighted sum of all energy terms, and is completely differentiable (or uses a differentiable approximation).

[0065] At this point, all complex initial constraints can be ignored, and we can start directly from the initial attitude and proceed along... Iterative update of negative gradient direction The process continues until it converges to the optimal betting posture that minimizes the total energy. The greatest advantage of this formalization is that it encapsulates complex physical and physiological compliance issues into a single mathematical function, enabling efficient and flexible solutions using modern automatic differentiation frameworks.

[0066] S320, a gradient calculation method based on an automatic differentiation framework is used to optimize and solve the energy minimization extreme value constraint problem, thereby obtaining the optimal posture of the acupuncture hand robot.

[0067] Specifically, a differentiable computational graph is constructed by combining parameterized decision variables and a total energy function in the form of soft constraints; the framework's automatic differentiation interface is called, and the graph is propagated backward along the differentiable computational graph and the chain rule is used to calculate the precise gradient vector of the total energy with respect to the acupuncture hand posture parameters; the precise gradient vector is processed by non-differentiable operations and engineering optimization to obtain the optimal acupuncture hand robot posture.

[0068] In this embodiment, the gradient calculation of the energy function with respect to the hand posture parameters in this invention can be efficiently completed using modern automatic differentiation frameworks such as PyTorch. The core principle is to construct the entire energy calculation process as a differentiable computational graph, and utilize the framework's automatic differentiation mechanism to obtain the precise gradients of all parameters (hand pose and 15 joint angles, totaling 22 dimensions) at once, without the need for manually deriving complex derivative formulas.

[0069] First, the energy calculation is constructed as a differentiable computation graph. To achieve automatic differentiation, the entire process of generating the hand gesture needs to be represented using differentiable tensor operations within the framework. The specific steps are as follows: 1) Define the attitude parameters as differentiable tensors. In PyTorch, the hand position... Palm facing quaternion (7 dimensions in total) and joint angle vectors All of these need to be set. The framework is created in tensor form, and from there it begins to record all operations performed on these parameters.

[0070] 2) Constructing a forward kinematics computation chain. The approach involves calculating the hand mesh using URDF files, and then using forward kinematics to map the attitude parameters to the spatial poses of each finger link and the 3D coordinates of the contact points. This process involves quaternion rotations, translation transformations, and trigonometric function operations, all of which are natively supported differentiable operations within the framework.

[0071] 3) Achieve differentiable calculations of various energy levels. Contact energy. The squared distance calculation, the piecewise penalty of the max function in the joint limit energy, and the "back-penetration energy" used to replace non-differentiable penetration detection are all implemented by calculating the distance from the object surface point cloud to the hand mesh; these operations are inherently differentiable. For stability energy... The force closure assessment involved can be performed using a differentiable force closure estimator, by constructing a capture matrix G using matrix multiplication and then calculating... This process consists entirely of linear algebra operations, and gradients can be seamlessly calculated within the automatic differentiation framework.

[0072] 4) Sum all the energy terms according to their weights to obtain the total energy. At this point, the framework has built a complete computation graph in memory, recording all operation nodes and dependencies from the initial attitude parameters to the final energy scalar.

[0073] Furthermore, the gradient is obtained by calling the automatic differentiation function. Once the computation graph is constructed, the gradient can be obtained simply by calling the automatic differentiation interface provided by the framework. Taking PyTorch as an example, after calling E_total.backward(.), the framework will propagate backward along the computation graph, automatically calculating the partial derivative of the total energy with respect to each input parameter using the chain rule, and finally obtaining the gradient vector. Its dimensions correspond perfectly to the attitude parameters (22 dimensions). This approach automates the complex calculation of partial derivatives, eliminating the need for developers to manually derive the Jacobian matrix for each energy term.

[0074] Finally, for handling non-differentiable operations and engineering optimization, a practical strategy is provided for a few non-differentiable energy terms. Differentiable approximation uses the Kaolin library to calculate the signed distance from a point to the mesh, replacing discrete collision detection and transforming the originally non-differentiable "yes / no" judgment into a continuously differentiable distance value. In terms of computational efficiency, these frameworks support batch parallel computation: they can simultaneously construct computational graphs of multiple candidate poses, utilizing GPUs to compute the energy and gradients of all candidate poses in parallel, significantly accelerating the iterative optimization process. By increasing the batch size, 1.32 million valid grasps were generated in just 950 GPU hours on an NVIDIA A100, achieving an order-of-magnitude efficiency improvement compared to the original algorithm. This strategy can also be applied to the optimization of the betting pose in this embodiment of the invention, achieving efficient large-scale pose generation and verification.

[0075] The optimization algorithm chosen is stochastic gradient descent (SGD), and its optimization process is as follows: 1) Initialization: random initialization or initialization of pose based on prior knowledge. Set the learning rate Maximum number of iterations Convergence threshold .

[0076] 2) Iterative optimization: calculate energy function value, calculate gradient, update attitude parameters, and determine convergence.

[0077] 3) Post-processing: physical feasibility check and comfort verification.

[0078] 4) Return to the optimal pose .

[0079] In summary, as Figure 3 As shown, this invention employs a gradient optimization method based on energy functions. By constructing a pressing hand posture energy function, the pressing hand posture generation problem is transformed into an energy minimization problem. First, geometric modeling of the pressing hand's palm and fingers is performed, establishing a three-dimensional geometric model of the pressing hand (including the palm and five fingers) and defining parameterized representations of joint angles. Next, target acupoints and patient body shape are modeled, obtaining the three-dimensional coordinates of the target acupoints and the surface geometric information of the relevant patient areas. Further, the pressing hand posture energy function is constructed, designing an energy function that comprehensively considers multiple objectives, including contact energy, comfort energy, stability energy, and accessibility energy. Then, gradient optimization of the energy function is performed, using a gradient descent-type optimization algorithm to iteratively solve for the pressing hand corresponding to the minimum energy. Finally, post-processing and verification of the posture are conducted, and the optimization results are physically feasible and comfort is evaluated.

[0080] Therefore, the embodiments of the present invention have the following distinguishing technical features compared to the prior art: 1) The energy optimization method in robotic dexterity grasping was innovatively applied to the generation of acupuncture hand posture, solving the technical problem of automated generation of hand posture; 2) An energy function specifically designed for the presser task was developed, taking into account multiple objectives such as contact, stability, comfort, accessibility, and effectiveness; 3) The application of the force closure assessment method in the distribution of pressing force by the presser is proposed to ensure that the presser provides a stable and appropriate pressing force. 4) A hand comfort assessment system was established, incorporating ergonomic factors such as joint limits and joint stress into the energy function.

[0081] In summary, the embodiments of the present invention have the following advantages compared with the prior art: 1) It can automatically generate a suitable pressing posture based on the patient's acupoint location and body shape characteristics, solving the problem of the "one-size-fits-all" approach of traditional methods; 2) By using the force closure and contact force balance terms in the energy function, the pressure applied by the hand is ensured to provide appropriate and evenly distributed pressure, thereby improving patient comfort; 3) Introduce ergonomic energy parameters such as joint limits and joint stress to avoid discomfort to the human body caused by the pressing posture; 4) Energy-optimized methods have good generalization performance and can handle unseen acupoint locations and patient body types; 5) Compared to end-to-end learning methods, gradient-based optimization methods typically require less training data and computational resources.

[0082] Reference Figure 2 An energy-optimized acupuncture hand gesture generation system includes: The first module 201 is used to construct a three-dimensional geometric model of the acupuncture hand robot and define the hand posture representation; The second module 202 is used to combine contact energy, comfort energy, stability energy and accessibility energy to construct a multi-dimensional target hand gesture energy optimization function; The third module 203 is used to transform the multi-dimensional target hand posture energy optimization function into an energy minimization extreme value constraint problem. Combining the three-dimensional geometric model of the acupuncture hand robot with the hand posture representation, the optimal acupuncture hand robot posture is obtained by solving the problem through a gradient descent optimization algorithm.

[0083] The content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0084] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this is not intended to limit the scope of the embodiments of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the embodiments of the present application.

Claims

1. A method for generating acupuncture hand gestures based on energy optimization, characterized in that, Includes the following steps: Construct a three-dimensional geometric model of the acupuncture hand robot and define the hand posture representation; By combining contact energy, comfort energy, stability energy, and accessibility energy, a multi-dimensional target hand gesture energy optimization function is constructed. The multi-dimensional objective hand pose energy optimization function is transformed into an energy minimization extreme value constraint problem. Combining the three-dimensional geometric model of the acupuncture hand robot with the hand pose representation, the optimal acupuncture hand robot pose is obtained by solving the problem using a gradient descent optimization algorithm.

2. The method for generating acupuncture hand posture based on energy optimization according to claim 1, characterized in that, The three-dimensional geometric model of the acupuncture hand-pressing robot has a multi-joint linkage mechanism. The hand-pressing posture representation includes the palm pose, joint angles, and the total parameter dimensions of the acupuncture hand-pressing robot. The multi-joint linkage mechanism specifically includes palm joints, finger joints, and joint angle vectors, wherein: The palm joint serves as a reference rigid body, defining the center position and orientation of the palm. Each of the five finger joints includes a metacarpophalangeal joint, a proximal interphalangeal joint, and a distal interphalangeal joint.

3. The method for generating acupuncture hand posture based on energy optimization according to claim 2, characterized in that, The multi-dimensional target hand gesture energy optimization function specifically includes contact energy, stability energy, comfort energy, accessibility energy, and effectiveness energy, among which: The contact energy is used to measure the degree of contact between the pressing hand and the target acupoint area. The contact energy includes positional contact energy and force contact energy. The stability energy is used to evaluate the stability and anti-interference ability of the presser's posture. The stability energy includes force containment energy and contact force balance energy. The force containment energy is used to evaluate whether the presser can resist external force interference in any direction. The contact force balance energy is used to ensure that the forces at each contact point are in a balanced state. The comfort energy is used to assess the impact of hand posture on the human body. The comfort energy includes joint limit energy, joint stress energy, and finger spread energy. The joint limit energy is used to penalize postures that exceed the joint range of motion. The joint stress energy is used to penalize postures that exceed the preset requirement for joint flexion. The finger spread energy is used to penalize postures that exceed the preset requirement for finger spread. The reachability energy is used to ensure that the generated posture is within the range of motion of the hand. The reachability energy includes joint range energy and palm reachability energy. The joint range energy is used to ensure that all joint angles are within the reachable range, and the palm reachability energy is used to ensure that the palm can reach the target position. The effective energy is used to eliminate physically infeasible or uncomfortable postures. The effective energy includes self-collision energy and finger gap energy. The self-collision energy is used to punish self-collision of the pressing hand, and the finger gap energy is used to ensure that an appropriate gap is maintained between adjacent fingers.

4. The method for generating acupuncture hand posture based on energy optimization according to claim 3, characterized in that, The step of transforming the multi-dimensional target hand posture energy optimization function into an energy minimization extremum constraint problem, combining the three-dimensional geometric model of the acupuncture hand robot with the hand posture representation, and solving it using a gradient descent optimization algorithm to obtain the optimal acupuncture hand robot posture, specifically includes: Combining the three-dimensional geometric model and hand posture representation of the acupuncture hand robot, the multi-dimensional objective hand posture energy optimization function is transformed into an energy minimization extreme value constraint problem; Based on the gradient calculation method of the automatic differentiation framework, the energy minimization extreme value constraint problem is optimized and solved to obtain the optimal posture of the acupuncture hand robot.

5. The method for generating acupuncture hand posture based on energy optimization according to claim 4, characterized in that, The step of combining the three-dimensional geometric model of the acupuncture hand-pressing robot with the hand-pressing posture representation to transform the multi-dimensional objective hand-pressing posture energy optimization function into an energy minimization extremum constraint problem specifically includes: The three-dimensional geometric model and hand posture representation of the acupuncture hand-pressing robot are converted into palm pose parameters and finger joint angle vectors, and parameterized decision variables are defined. The physical and physiological hard constraints in the multi-dimensional target hand posture energy optimization function are transformed into differentiable penalty terms and weighted summation is performed to construct the total energy function in the form of soft constraints. Substituting the parameterized decision variables into the total energy function in the form of soft constraints yields the energy minimization extreme value constraint problem.

6. The method for generating acupuncture hand posture based on energy optimization according to claim 5, characterized in that, The step of transforming the physical and physiological hard constraints in the multi-dimensional target hand posture energy optimization function into differentiable penalty terms and then performing a weighted summation to construct a soft-constraint form of the total energy function specifically includes: Based on the contact energy in the multi-dimensional target pressure hand posture energy optimization function, the sum of squared distances between the fingertip contact point and the target acupoint is minimized to drive the fingertip to align with the acupoint, thus constructing a contact energy constraint term; Based on the stability energy in the multi-dimensional target presser posture energy optimization function, and based on a differentiable force closed approximation estimator, the pressing stability of the posture is quantified, and a stability energy constraint term is constructed. Based on the comfort energy in the multi-dimensional target hand posture energy optimization function, the hard boundary of the joint angle is transformed into a funnel-shaped penalty function to construct the joint limit energy constraint term; Based on the reachability energy and effectiveness energy in the multi-dimensional target bettor posture energy optimization function, differentiable self-collision and skin penetration detection are realized by point cloud distance calculation. When penetration occurs, a penalty is applied, and a penetration penalty energy constraint term is constructed. By combining the contact energy constraint, stability energy constraint, joint limit energy constraint, and penetration penalty energy constraint, a weighted sum is performed to construct the total energy function in the form of soft constraints.

7. The method for generating acupuncture hand posture based on energy optimization according to claim 6, characterized in that, The gradient calculation method based on the automatic differentiation framework optimizes the energy minimization extremum constraint problem to obtain the optimal posture of the acupuncture hand robot. This step specifically includes: By combining parameterized decision variables with the total energy function in the form of soft constraints, a differentiable computational graph is constructed. The framework's automatic differentiation interface is invoked, and the algorithm propagates backward along the differentiable computation graph and uses the chain rule to calculate the precise gradient vector of the total energy with respect to the bettor's posture parameters. By processing the precise gradient vector through non-differentiable manipulation and engineering optimization, the optimal posture of the acupuncture hand robot is obtained.

8. The method for generating acupuncture hand posture based on energy optimization according to claim 7, characterized in that, The step of constructing a differentiable computational graph by combining parameterized decision variables and the total energy function in the form of soft constraints specifically includes: Convert parameterized decision variables into differentiable tensors; By mapping differentiable tensors to the spatial poses of each link in the finger and the three-dimensional coordinates of the contact points through forward kinematics, a forward kinematics computation chain is constructed. The total energy function in the form of soft constraints is transformed into a differentiable computation function, which is then substituted into the forward kinematics computation chain for weighted summation to construct a differentiable computation graph.

9. The method for generating acupuncture hand posture based on energy optimization according to claim 8, characterized in that, The step of processing the precise gradient vector through non-differentiable manipulation and engineering optimization to obtain the optimal posture of the acupuncture hand robot specifically includes: Discrete collision detection was performed on the acupuncture hand-pressing robot. Differentiable approximation was performed using the signed distance calculation of point clouds in the Kaolin library to obtain the discrete collision detection results that are differentiable approximations. By leveraging the batch parallel capabilities of GPUs, a computational graph of multiple candidate poses is constructed. Combined with the discrete collision detection results that are differentiable approximations, energy and gradients are calculated in parallel. Starting from the initial posture, the posture parameters are iteratively updated along the negative gradient direction until the total energy converges, and the optimal acupuncture hand robot posture is output.

10. An energy-optimized acupuncture hand gesture generation system, characterized in that, Includes the following modules: The first module is used to construct the three-dimensional geometric model of the acupuncture hand robot and define the hand posture representation; The second module is used to combine contact energy, comfort energy, stability energy and accessibility energy to construct a multi-dimensional target hand gesture energy optimization function; The third module is used to transform the multi-dimensional target hand posture energy optimization function into an energy minimization extreme value constraint problem. Combining the three-dimensional geometric model of the acupuncture hand robot with the hand posture representation, the optimal acupuncture hand robot posture is obtained by solving the problem through a gradient descent optimization algorithm.