Adaptive compensation and dynamic balance control method for joint torque of surgical robot arm

Abstract

A method for adaptive compensation and dynamic balancing control of arm joint torque in a surgical robot is disclosed. The method comprises the following steps: obtaining arm joint motion state parameters; constructing a feature vector based on the motion state, dynamic characteristics, friction characteristics, and load characteristics; and predicting joint torque values ​​using a Gaussian process regression model with a composite kernel function containing basis terms, load terms, and friction terms. A sliding surface is constructed based on the deviation between the actual motion state and the desired motion state, and a control torque is generated by combining the predicted joint torque values ​​with an adaptive gain control structure and a variable saturation function of the boundary layer based on external torque. Based on the deviation between the control torque and the actual measured torque, the sliding control parameters are dynamically adjusted to achieve joint balance. This application can effectively improve the operational performance and safety of surgical robots.

Description

Technical Field

[0001] The present application relates to the technical field of surgical robot control, and in particular to a method for adaptive compensation and dynamic balance control of joint torque of a surgical robot arm. Background Art

[0002] Surgical robots are playing an increasingly important role in precise surgical operations. The torque control of arm joints directly impacts the accuracy and safety of surgical procedures. In practical applications, the dynamic characteristics of robotic arm joints are affected by a variety of factors, including joint posture, load variations, and friction, parameters that are difficult to accurately model using analytical methods. Furthermore, human-machine interaction during surgery introduces external interference and uncertainty into the system, making traditional fixed-parameter control methods difficult to adapt to these dynamic changes. Furthermore, surgical operations place seemingly contradictory demands on robots: they must maintain sufficient positional accuracy while also possessing appropriate compliance to ensure safety.

[0003] To address these issues, researchers have proposed a variety of control methods. Compensatory control methods based on dynamic models, while theoretically sound, rely heavily on precise model parameters. Adaptive control methods can respond to parameter changes, but their parameter adjustment rules are complex. Learning-based control methods, while offering certain advantages, still need improvement in terms of real-time performance and robustness.

[0004] Therefore, it is necessary to study a control method that can adaptively estimate and compensate joint torque while ensuring dynamic balance to improve the operational performance and safety of surgical robots. Summary of the Invention

[0005] This application proposes a method for adaptive compensation and dynamic balance control of the joint torque of a surgical robot arm, which can effectively improve the operational performance and safety of the surgical robot.

[0006] According to one embodiment of the present application, a method for adaptive compensation and dynamic balance control of joint torque of a surgical robot arm is proposed, comprising:

[0007] Obtain the motion state parameters of the arm joint, including position, velocity and acceleration;

[0008] Based on the arm joint motion state, dynamic characteristics, friction characteristics and load characteristics, a feature vector is constructed, and a Gaussian process regression model with a composite kernel function including basis terms, load terms and friction terms is used to predict the joint torque value.

[0009] Based on the deviation between the actual motion state of the arm joint and the expected motion state, a sliding surface is constructed. An adaptive gain control structure and a variable saturation function of the boundary layer based on the external torque are used to generate a control torque combined with the predicted joint torque value to drive the arm joint.

[0010] According to the deviation between the control torque and the actual measured torque, the sliding mode control parameters are dynamically adjusted to adaptively adjust the control torque.

[0011] In some embodiments, the feature vector includes position, velocity, acceleration, gravity term, inertia term, friction term, and load term.

[0012] In some embodiments, predicting joint torque values ​​using a Gaussian process regression model includes:

[0013] The joint torque value τ_pred(x*) is predicted based on the following formula:

[0014] ,

[0015] Among them, k(x*,X)=[k_total(x*,x1),k_total(x*,x2),...,k_total(x*,xN)] represents the vector composed of the composite kernel function values ​​of the test point x* and each training sample point, k(x*,X)T represents the transpose of the vector k(x*,X), and K is the kernel function matrix obtained based on the training sample points. is the observation noise variance, I is the identity matrix, Indicates the matrix The inverse matrix of , y is the training data moment value vector, and the Gaussian process regression model is based on the following composite kernel function:

[0016] k_total(x,x')=k_basic(x,x')+k_load(x,x')+k_friction(x,x'),

[0017] Where x and x' represent the two input state eigenvectors, k_basic(x,x') is the radial basis kernel function, k_load(x,x') is the load kernel function, and k_friction(x,x') is the friction kernel function.

[0018] In some embodiments, the radial basis kernel function k_basic(x, x') is determined according to the following formula:

[0019] ,

[0020] in, is the basic kernel function signal variance, x i and x' i are the i-th component in the state eigenvectors x and x', respectively, l i is the length scale parameter of the feature corresponding to the i-th component in the state feature vector;

[0021] The load kernel function k_load(x, x') is determined according to the following formula:

[0022] ,

[0023] in, and are the joint positions in the state feature vectors x and x', respectively, and are the load characteristic related components in the state characteristic vectors x and x', is the load signal variance, l m is the length scale parameter of the load characteristic;

[0024] The friction kernel function k_friction(x,x') is determined according to the following formula:

[0025] ,in, and are the joint velocities in the state eigenvectors x and x', and are the friction force characteristic related components in the state eigenvectors x and x', is the friction signal variance, is the length scale parameter of the friction characteristic.

[0026] In some embodiments, a sliding surface is constructed based on the deviation between the actual motion state of the arm joint and the desired motion state, and an adaptive gain control structure and a boundary layer variable saturation function based on an external torque are used to generate a control torque in combination with the predicted joint torque value to drive the arm joint motion, including:

[0027] The sliding surface is constructed based on the following formula:

[0028] ,

[0029] in, is the position error, is the speed error, and are the actual joint positions and velocities, and are the desired joint positions and velocities, respectively, and λ is a preset positive constant;

[0030] The torque compensation is generated based on the following formula:

[0031] ,

[0032] Where K(s) is the adaptive gain function, is the boundary layer width, is the variable saturation function of the boundary layer;

[0033] Combining the predicted joint torque value τ_pred(x*) and the torque compensation amount, the control torque τ_control is obtained:

[0034] τ_control = u +τ_pred(x*).

[0035] In some embodiments, the gain function K(s) is determined according to the following formula:

[0036] K(s) = Kmin + β|s| + γ·m load (q*), where Kmin is the minimum gain value, β is the adaptive gain parameter, and γ is the preset load factor.

[0037] In some embodiments, the saturation function sat(s / Φ) is determined according to the following formula: sat(s / Φ) = s / Φ, when |s| ≤ Φ, sat(s / Φ) = sign(s), when |s| > Φ, where, , is the initial boundary layer width, α is the preset attenuation coefficient, η is the preset safety factor, is the external torque.

[0038] In some embodiments, dynamically adjusting the sliding mode control parameters to adaptively adjust the control torque based on the deviation between the control torque and the actual measured torque includes:

[0039] The torque deviation Δτ is obtained according to the following formula:

[0040] Δτ = τ_control - τ_total,

[0041] Among them, τ_control is the control torque, and τ_total is the joint torque measured in real time;

[0042] When |Δτ| > Δτ0, reduce the adaptive gain parameter β to reduce the control torque,

[0043] When |Δτ| ≤ Δτ0, the adaptive gain parameter β is increased to increase the control torque until the adaptive gain parameter β reaches a preset nominal value, where Δτ0 is a preset torque deviation threshold.

[0044] The surgical robot arm joint torque adaptive compensation and dynamic balance control method proposed in this application has the following advantages:

[0045] By constructing a state feature vector that includes joint motion state, dynamic parameters, friction characteristics, and load characteristics, a Gaussian process regression model with a composite kernel function is used for torque prediction, effectively avoiding the traditional method's reliance on precise dynamic models. Specifically, the designed composite kernel function separately models the basic characteristics, load characteristics, and friction characteristics, improving the accuracy of torque prediction. The load kernel function captures the impact of load changes on torque through load characteristic parameters, while the friction kernel function comprehensively describes friction characteristics by combining Coulomb friction and viscous friction.

[0046] In terms of torque compensation control, this application designs an adaptive gain control structure based on the sliding surface and introduces a variable saturation function in the boundary layer based on external torque to effectively compensate for system uncertainty. The synergistic effect of the adaptive gain function and saturation function ensures control robustness while avoiding the system chatter problem in traditional sliding mode control. At the same time, the control torque generated based on the predicted torque and compensation torque can effectively drive joint motion and ensure tracking accuracy.

[0047] In terms of dynamic balance control, this application monitors the deviation between the control torque and the actual measured torque in real time and dynamically adjusts the sliding mode control parameters, enabling the system to adaptively adjust the control torque. When the torque deviation is large, system compliance is increased by increasing the boundary layer width and reducing the adaptive gain parameter. When the torque deviation is small, control accuracy is maintained by gradually restoring the parameter to its nominal value. This dynamic adjustment strategy based on torque feedback achieves adaptive changes in the control torque, effectively balancing the requirements of operational accuracy and safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the specification and, together with the description, serve to explain the principles of the specification.

[0049] Figure 1 A flow chart of a method for adaptive compensation and dynamic balance control of joint torque of a surgical robot arm according to one embodiment of the present application is shown.

[0050] Figure 2 It is a structural diagram of an electronic device shown in at least one embodiment of the present application. DETAILED DESCRIPTION

[0051] Exemplary embodiments will be described in detail herein, with examples illustrated in the accompanying drawings. In the following description, when referring to the drawings, identical numerals in different figures represent identical or similar elements, unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all embodiments consistent with the present application. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present application, as detailed in the appended claims.

[0052] The embodiments of the present application can be applied to a computer system / server that can operate in conjunction with numerous other general-purpose or special-purpose computing system environments or configurations. Examples of well-known computing systems, environments, and / or configurations suitable for use with the computer system / server include, but are not limited to, personal computer systems, server computer systems, thin clients, thick clients, handheld or laptop devices, microprocessor-based systems, set-top boxes, programmable consumer electronics, network personal computers, minicomputer systems, mainframe computer systems, and distributed cloud computing technology environments including any of the above, among others.

[0053] Computer systems / servers may be described in the general context of computer system-executable instructions, such as program modules, executed by a computer system. Generally, program modules may include routines, programs, objects, components, logic, data structures, and the like, that perform specific tasks or implement specific abstract data types. Computer systems / servers may be implemented in a distributed cloud computing environment where tasks are performed by remote processing devices linked through a communications network. In a distributed cloud computing environment, program modules may be located on local or remote computer system storage media, including storage devices.

[0054] Figure 1 A flowchart of a method for adaptive compensation and dynamic balance control of joint torque of a surgical robot arm according to an embodiment of the present application is shown. As shown in the figure, the method includes steps 1 to 4.

[0055] Step 1: Obtain the motion state parameters of the arm joint, including position, velocity and acceleration.

[0056] In the field of robot arm joint control, the joint position can usually be directly measured by a high-precision encoder; the joint speed can be obtained by differential calculation of the encoder signal, or it can be directly measured using a dedicated speed sensor (such as a Hall sensor, resolver, etc.); the joint acceleration can usually be measured using an acceleration sensor, or obtained by differential calculation of the speed signal.

[0057] In some implementations, a combination of joint encoders and accelerometers can be employed. The encoders measure joint position and obtain velocity signals through numerical differentiation, while the accelerometers directly measure joint acceleration. To obtain high-quality motion state parameters, the encoder-measured position signals and the calculated velocity signals can be processed using low-pass filtering to reduce noise. Simultaneously, the accelerometer signals are fused using a Kalman filter algorithm to improve the accuracy and reliability of acceleration measurements. These state parameters are used for subsequent torque estimation and control calculations.

[0058] Step 2: construct a feature vector based on the arm joint motion state, dynamic characteristics, friction characteristics and load characteristics, and use a Gaussian process regression model with a composite kernel function containing basic terms, load terms and friction terms to predict the joint torque value.

[0059] In this step, the feature vector is first constructed, and then the torque prediction is realized based on the Gaussian process regression model of the composite kernel function.

[0060] The eigenvector is a set of key parameters that reflect the characteristics of the system. Reasonable eigenvector construction has an important impact on the performance of the prediction model.

[0061] In some embodiments, the feature vector includes position, velocity, acceleration, gravity term, inertia term, friction term, and load term. In one example, the feature vector , ,in is the motion state parameter; gravity term Can be defined as ; Inertia term middle Can be defined as , I joint is the moment of inertia of the joint itself, m is the load mass, is the distance from the load to the joint center; the friction term Can be defined as , f c is the Coulomb friction coefficient, f v is the viscous friction coefficient, is the velocity sign function; the load term Can be defined as , τ is the measured joint torque, c is the bias term, g is the gravitational acceleration constant, and l is the length of the lever arm. Those skilled in the art can also obtain the components of the eigenvector using other methods.

[0062] The feature vector design according to this embodiment fully considers the dynamic characteristics of the arm joint system: the motion state characteristics (position, velocity, acceleration) reflect the real-time motion characteristics of the system, the dynamic characteristics (gravity term, inertia term) reflect the physical characteristics of the system, and the friction characteristics and load characteristics (friction term, load term) specifically describe these two important influencing factors.

[0063] The Gaussian Process Regression (GPR) model is a nonparametric probabilistic model used for function regression. It defines a Gaussian process to describe the prior distribution of the target function and uses Bayesian inference to predict new data points, while also providing an estimate of the uncertainty of the prediction results. The kernel function is a core component of the GPR model. It defines the similarity between data points and determines how the GPR model makes predictions. The length scale parameter is a key parameter in the kernel function, controlling the "width" or "range of influence" of the kernel function. The length scale parameter determines the degree to which the distance between data points affects the kernel function value. If the data on some components may vary significantly, a smaller length scale can be used; if the data on other components may vary more gradually, a larger length scale can be used.

[0064] In some embodiments, predicting joint torque values ​​using a Gaussian process regression model includes:

[0065] Predict joint torque values ​​based on the following formula :

[0066] ,

[0067] in, , indicating the test point and the vector composed of the composite kernel function values ​​of each training sample point, T represents vector The transpose of , K is the kernel function matrix obtained according to the training sample points, is the observation noise variance, I is the identity matrix, Indicates the matrix The inverse matrix of , y is the training data moment value vector, and the Gaussian process regression model is based on the following composite kernel function:

[0068] ,

[0069] Where x and x' represent the two input state eigenvectors, k_basic(x,x') is the radial basis kernel function, k_load(x,x') is the load kernel function, and k_friction(x,x') is the friction kernel function.

[0070] The composite kernel function k_total(x,x') used in the composite kernel function model of this embodiment includes a base term, a load term, and a friction term, which can capture the base characteristics, load characteristics, and friction characteristics of the system respectively. The term acts as observation noise and also plays a role of regularization to avoid overfitting.

[0071] In some embodiments, the radial basis kernel function is determined according to the following formula: : ,

[0072] in, is the basic kernel function signal variance, and are the i-th components in the state eigenvectors x and x', respectively, is the length scale parameter of the feature corresponding to the i-th component in the state feature vector;

[0073] The load kernel function k_load(x, x') is determined according to the following formula:

[0074] ,

[0075] Where q and q' are the joint positions in the state feature vectors x and x' respectively, and are the load characteristic related components in the state characteristic vectors x and x', is the load signal variance, l m is the length scale parameter of the load characteristic;

[0076] The friction kernel function k_friction(x,x') is determined according to the following formula:

[0077] ,

[0078] in, and are the joint velocities in the state eigenvectors x and x', and are the friction force characteristic related components in the state eigenvectors x and x', is the friction signal variance, is the length scale parameter of the friction characteristic.

[0079] In this implementation, the basic kernel function k_basic in the composite kernel function uses radial basis functions (RBFs) to measure the relationship between samples based on the similarity of state features. The load kernel function k_load specifically models load characteristic parameters, where the load characteristic term can be derived by inversely solving the torque measurements. The friction kernel function k_friction combines Coulomb friction and viscous friction to model the system. This multi-kernel design allows the model to more accurately describe the nonlinear characteristics of the system and improve the accuracy of arm joint torque prediction.

[0080] Before using the GPR model for predictions based on test points, the model can be fully trained offline. Training data can first be collected in the arm joint's actual operating environment. By designing a well-designed experimental plan, the arm joint's state parameters and corresponding torque values ​​are collected under different operating conditions, including varying joint positions, movement speeds, and load masses, to ensure data representativeness and completeness. Typically, hundreds to thousands of sets of sample data are collected to construct a training dataset {X, y}, where X is the feature vector sample set and y is the corresponding torque value vector.

[0081] Before model training, the collected raw data can be preprocessed. Normalization is used to map each characteristic component to the [0, 1] interval, eliminating dimensional differences between physical quantities and improving model training effectiveness. Abnormal data can also be screened and processed to ensure training data quality.

[0082] The core of model training is to determine the hyperparameters of the kernel function. The maximum likelihood estimation method can be used to optimize the log-likelihood function of the observed data to determine the various parameters in the composite kernel function, including the signal variance. and the length scale parameter l corresponding to each characteristic component i The optimization process can be implemented using common algorithms such as gradient descent. Considering the non-convexity of the hyperparameter optimization problem, multiple random initializations are usually used to obtain better optimization results.

[0083] To evaluate the model's predictive performance, cross-validation can be used. The dataset is randomly divided into a training set and a test set in an 8:2 ratio. The model is trained using the training set, and the predictive performance is evaluated on the test set. Evaluation metrics include root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²). If predictive performance does not meet requirements, the quality and coverage of the training data can be improved by optimizing the data collection scheme, including increasing the number of sampling points, refining the sampling conditions, and improving the signal-to-noise ratio of the data collection.

[0084] In practical applications, to enable the GPR model to adapt to dynamic changes in system characteristics, incremental learning can be used to regularly update the model. New sample data is continuously collected during online operation and regularly used to retrain the model and update the kernel function parameters to maintain the model's prediction accuracy. A fixed update cycle can be set, or model updates can be triggered adaptively based on the changing trend of the prediction error.

[0085] In step 3, a sliding surface is constructed based on the deviation between the actual motion state of the arm joint and the expected motion state. The control torque is generated by combining the predicted joint torque value with the adaptive gain control structure and the variable saturation function of the boundary layer based on the external torque.

[0086] Sliding mode control is a robust control method for nonlinear systems. The sliding surface is a special manifold in the system state space where the system state slides along it while maintaining the desired dynamic characteristics. The boundary layer is a region near the sliding surface that is used to reduce system chatter. The control torque is converted into motor current by the motor driver, which then generates a corresponding torque that acts on the arm joint.

[0087] In some embodiments, generating the control torque based on the sliding surface includes constructing the sliding surface based on the following formula: ,in, is the position error, is the speed error, and are the actual joint positions and velocities, and are the desired joint positions and velocities, respectively, and λ is a preset positive constant. The torque compensation is generated based on the following formula: , where K(s) is the adaptive gain function, Φ is the boundary layer width, and sat(s / Φ) is the boundary layer variable saturation function; Combined with the predicted joint torque value And the torque compensation, the control torque τ_control is obtained: .

[0088] According to this embodiment, position tracking e and velocity tracking ė are unified into one control target. The control torque is obtained by sliding surface and predicted torque. , where u = -K(s)·sat(s / Φ) is the torque compensation.

[0089] In some embodiments, the gain function K(s) is determined according to the following formula:

[0090] K(s) = Kmin + β|s| + γ·m load (q*),

[0091] Where Kmin is the minimum gain value, β is the adaptive gain parameter, and γ is the preset load factor.

[0092] The adaptive gain function K(s) according to this embodiment includes three parts: the basic gain Kmin ensures the basic control performance, the adaptive term β|s| is dynamically adjusted according to the tracking error, and the load compensation term Consider the effects of load changes.

[0093] In some embodiments, the saturation function sat(s / Φ) is determined according to the following formula:

[0094] sat(s / Φ) = s / Φ, when |s| ≤ Φ,

[0095] sat(s / Φ) = sign(s), when |s| > Φ,

[0096] in, , is the initial boundary layer width, α is the preset attenuation coefficient, η is the preset safety factor, and τext is the external torque.

[0097] τext refers to the interactive torque from the external environment. A torque sensor can be installed at the joint to directly measure the external torque τext, or the residual method can be used to estimate the external torque τext. The torque sensors commonly used in direct measurement methods include strain gauge and piezoelectric types, which have the advantages of fast response and high precision, but the system cost is relatively high. The residual method estimates the external torque by comparing the measured total torque and the internal torque of the system calculated by the model. It is simple to implement but is more dependent on the accuracy of the model. An example of this embodiment uses a six-dimensional force / torque sensor to directly measure the external torque, ensuring higher control performance.

[0098] According to this embodiment, the boundary layer variable saturation function sat(s / Φ) is dynamically adjusted by adjusting the boundary layer width. To achieve adaptive changes in the control torque, s and τext will be adjusted online according to the motion state and torque feedback.

[0099] The above embodiment introduces adaptability into the gain function and saturation function, enabling the controller to dynamically adjust control parameters according to the system state and external environment, while combining torque prediction to achieve precise adaptive compensation and control.

[0100] Step 4: According to the deviation between the control torque and the actual measured torque, the sliding mode control parameters are dynamically adjusted to adaptively adjust the control torque.

[0101] In some embodiments, the torque deviation Δτ is obtained according to the following formula:

[0102] Δτ = τ_control - τ_total,

[0103] Among them, τ_control is the control torque, and τ_total is the joint torque measured in real time;

[0104] When |Δτ| > Δτ0, reduce the adaptive gain parameter β to reduce the control torque,

[0105] When |Δτ| ≤ Δτ0, the adaptive gain parameter β is increased to increase the control torque until the adaptive gain parameter β reaches a preset nominal value, where Δτ0 is a preset torque deviation threshold.

[0106] τ_total refers to the total joint torque, measured by the joint motor's current feedback or torque sensor. It includes the combined effects of internal system torques (such as gravity, inertia, friction, etc.) and external torques. When |Δτ| > Δτ0, the control action is too strong, and the adaptive gain parameter β can be adjusted and reduced to increase system compliance. When |Δτ| ≤ Δτ0, the torque control error is within an acceptable range, and β can be gradually restored to the preset nominal value to ensure control accuracy. This parameter adjustment strategy based on torque feedback avoids complex multi-parameter coupled adjustments while effectively adjusting the control torque through changes in a single parameter, β.

[0107] In some examples, β can be adjusted according to the following principles: a rapid response when decreasing to ensure safety, and a gradual approach when increasing to avoid system oscillation. Specifically, β can be decreased using an exponential decay approach and increased using a linear increase approach. Furthermore, a reasonable setting of the torque deviation threshold Δτ0 can be used to achieve a balance between system performance and safety.

[0108] The dynamic balance control according to this embodiment has a simple structure and is easy to implement; the parameter adjustment mechanism is clear, which facilitates system debugging; and it can effectively adapt to different working conditions to ensure the dynamic performance of the system.

[0109] The method for adaptive compensation and dynamic balance control of surgical robot arm joint torques proposed in the above embodiment uses Gaussian process regression based on physical characteristics for torque prediction, avoiding reliance on precise dynamic models. Composite kernel functions are used to model basic characteristics, load characteristics, and friction characteristics, respectively, improving the accuracy of torque prediction. In the control strategy, a single-parameter adaptive adjustment mechanism based on torque deviation is adopted. Dynamic balance of the control torque is achieved through the dynamic variation of the adaptive gain parameter β, while the boundary layer width is determined by the natural response of the sliding surface s and the external torque τext. The solution proposed in this embodiment ensures control accuracy while avoiding the complexity brought about by multi-parameter coupled adjustment, achieving an effective balance between system performance and safety.

[0110] An application example of this application is given below.

[0111] The experimental platform consists of a single-joint test bench, including: a permanent magnet synchronous motor (power 1kW), a harmonic reducer (reduction ratio 100:1), a torque sensor (range 20N·m), an encoder (resolution 17 bits) and a six-dimensional force / torque sensor (range 100N, 10N·m).

[0112] First, collect training data.

[0113] The joint positions were set in 10° intervals from 0° to 90°. Five different speeds were set at each position (-0.5 rad / s, -0.2 rad / s, 0 rad / s, 0.2 rad / s, and 0.5 rad / s). Three different masses (0.2 kg, 0.5 kg, and 0.8 kg) were loaded. The sampling frequency was 1 kHz, and 10 seconds of data were collected for each condition. A total of 500 valid data sets were obtained for all conditions, each of which included both raw data collection and feature engineering processing.

[0114] Raw data collection collects joint motion status (position ,speed , acceleration ) and the corresponding joint torque values. Feature engineering is used to convert raw data into feature vectors , the components are position, velocity, acceleration, gravity term, inertia term, friction term, and load term.

[0115] Next, train the GRP model.

[0116] The collected data is divided into training set and test set at an 8:2 ratio. The model parameters are set as follows:

[0117] Basic kernel function signal variance: ,

[0118] Load kernel function signal variance: ,

[0119] Friction kernel function signal variance: ,

[0120] Observation noise variance: ;

[0121] Length scale parameter of state characteristics:

[0122] Position component length scale: 0.5;

[0123] Velocity component length scale: 0.3

[0124] Acceleration component length scale: 0.2

[0125] Gravity characteristic component length scale: 0.4

[0126] Inertial characteristic component length scale: 0.3

[0127] Friction characteristic component length scale: 0.2

[0128] Load characteristic component length scale: 0.4.

[0129] The evaluation indicators of the training results on the test set are as follows:

[0130] Root mean square error ;

[0131] Mean absolute error ;

[0132] Coefficient of determination .

[0133] Then, set the basic parameters of the sliding surface control.

[0134] Perform position control experiments, and the expected trajectory is .

[0135] The basic parameters of sliding mode control are as follows:

[0136] Sliding surface coefficient: λ = 2.0,

[0137] Minimum gain: Kmin = 1.0,

[0138] Initial adaptive gain: β = 2.0,

[0139] Load factor: γ = 0.8.

[0140] The boundary layer parameters are as follows:

[0141] Initial boundary layer width: = 0.1,

[0142] Attenuation coefficient: α = 0.5,

[0143] Safety factor: η = 0.2.

[0144] The adaptive adjustment parameters are as follows:

[0145] Torque deviation threshold: Δτ0 = 0.3N·m,

[0146] Nominal adaptive gain value: βn = 2.0.

[0147] Finally, the experimental results are evaluated.

[0148] The position tracking performance is evaluated as follows:

[0149] Steady-state error is less than 0.005rad,

[0150] The maximum dynamic error is less than 0.05rad,

[0151] The overshoot is less than 5%.

[0152] The torque control performance is evaluated as follows:

[0153] The relative error of torque prediction is less than 3%.

[0154] The root mean square error of torque control deviation is less than 0.3N·m.

[0155] The system has good suppression capability against external disturbance torque of 1N·m.

[0156] The adaptive characteristics are evaluated as follows:

[0157] β changes adaptively within the range of 0.5~1.0,

[0158] System response time is less than 50ms,

[0159] There is no obvious oscillation phenomenon.

[0160] Experimental results show that the surgical robot arm joint torque adaptive compensation and dynamic balance control method proposed in this application can achieve precise position tracking control while maintaining good dynamic response characteristics and anti-interference capabilities.

[0161] To verify the effectiveness of this scheme, it is compared with traditional PID control and adaptive control based on dynamic model. The experiment is carried out under the same conditions, setting the expected trajectory as q d =π / 4·sin(t), each experiment was repeated 5 times and the average value was taken. The experimental results are shown in Table 1.

[0162] Table 1 Control performance comparison results

[0163]

[0164] As can be seen in Table 1, compared with traditional PID control and dynamic model-based adaptive control, the proposed method for adaptive compensation and dynamic balance control of surgical robot arm joint torques has significant advantages in terms of position tracking accuracy, torque control accuracy, dynamic response speed, and disturbance resistance. In particular, under external disturbances, the proposed method exhibits improved robustness, with a maximum disturbance resistance error of only 0.015 rad, a 64.3% reduction compared to the 0.042 rad of PID control.

[0165] Figure 2An electronic device provided for at least one embodiment of the present application includes a memory and a processor, the memory is used to store computer instructions that can be executed on the processor, and the processor is used to implement the surgical robot arm joint torque adaptive compensation and dynamic balance control method described in any embodiment or implementation of the present application when executing the computer instructions.

[0166] At least one embodiment of the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for adaptive compensation and dynamic balance control of the joint torque of the surgical robot arm as described in any embodiment or implementation of the present application.

[0167] Those skilled in the art will appreciate that one or more embodiments of this specification may be provided as methods, systems, or computer program products. Thus, one or more embodiments of this specification may take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware. Furthermore, one or more embodiments of this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code.

[0168] The various embodiments in this specification are described in a progressive manner. Similar portions between the various embodiments can be referenced to each other. Each embodiment focuses on the differences between the other embodiments. In particular, the data processing device embodiment is generally similar to the method embodiment, so its description is relatively simple. For relevant portions, refer to the description of the method embodiment.

[0169] The foregoing description of this specification describes specific embodiments. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that described in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order shown or sequential order to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0170] Although this specification includes many specific implementation details, these should not be interpreted as limiting the scope of any invention or the scope of protection claimed, but are mainly used to describe the features of specific embodiments of specific inventions. Certain features described in multiple embodiments within this specification may also be implemented in combination in a single embodiment. On the other hand, the various features described in a single embodiment may also be implemented separately in multiple embodiments or in any suitable sub-combination. In addition, although features may work in certain combinations as described above and even initially claimed as such, one or more features from the claimed combination may be removed from the combination in some cases, and the claimed combination may point to a sub-combination or a variation of the sub-combination.

[0171] Similarly, although operations are depicted in a particular order in the accompanying drawings, this should not be understood as requiring that these operations be performed in the particular order shown or performed sequentially, or that all illustrated operations be performed to achieve the desired results. In some cases, multitasking and parallel processing may be advantageous. In addition, the separation of various system modules and components in the above-described embodiments should not be understood as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated together in a single software product, or packaged into multiple software products.

[0172] Thus, specific embodiments of the subject matter have been described. Other embodiments are within the scope of the following claims. In some cases, the actions recited in the claims can be performed in a different order and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the particular order shown or sequential sequence to achieve the desired results. In some implementations, multitasking and parallel processing may be advantageous.

[0173] The above description is merely a preferred embodiment of one or more embodiments of this specification and is not intended to limit one or more embodiments of this specification. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of one or more embodiments of this specification shall be included in the scope of protection of one or more embodiments of this specification.

Claims

1. A method for adaptive compensation and dynamic balance control of joint torque of a surgical robot arm, characterized in that: include: Obtain the motion state parameters of the arm joint, including position, velocity and acceleration; The state feature vector is constructed based on the arm joint motion state, dynamic characteristics, friction characteristics, and load characteristics. The joint torque value is predicted using a Gaussian process regression model containing the following composite kernel function: k_total(x,x')=k_basic(x,x')+k_load(x,x')+k_friction(x,x'), Where x and x' represent two input state eigenvectors, k_basic(x,x') is the radial basis kernel function, k_load(x,x') is the load kernel function, and k_friction(x,x') is the friction kernel function. The radial basis kernel function k_basic(x,x') is determined according to the following formula: k_basic(x,x')= ²exp(-Σ(x i -x' i )² / (2l i ²)), in, ² is the variance of the basic kernel function signal, x i and x' i are the i-th component in the state eigenvectors x and x', respectively, l i is the length scale parameter of the feature corresponding to the i-th component in the state feature vector, The load kernel function k_load(x, x') is determined according to the following formula: k_load(x,x')= ²exp(- m load (q)–m load (q') ² / l m ²), Among them, q and q' are the joint positions in the state feature vectors x and x', respectively, and m load (q) and m load (q') are the load characteristic related components in the state characteristic vectors x and x', ² is the load signal variance, l m is the length scale parameter of the load characteristic, The friction kernel function k_friction(x, x') is determined according to the following formula: k_friction(x,x')= ²exp(- f( )-f( ') ² / l f ²), in, and ' are the joint velocities in the state feature vectors x and x', respectively, f( ) and f( ') are the friction force characteristic related components in the state eigenvectors x and x', ² is the friction signal variance, l f is the length scale parameter of the friction characteristic; Based on the deviation between the actual motion state of the arm joint and the expected motion state, a sliding surface is constructed. An adaptive gain control structure and a variable saturation function of the boundary layer based on the external torque are used to generate the control torque in combination with the predicted joint torque value. According to the deviation between the control torque and the actual measured torque, the sliding mode control parameters are dynamically adjusted to adaptively adjust the control torque.

2. The method according to claim 1, characterized in that The state characteristic vector includes position, velocity, acceleration, gravity term, inertia term, friction term, and load term.

3. The method according to claim 1, characterized in that The Gaussian process regression model is used to predict joint torque values ​​including: The joint torque value τ_pred(x ): τ_pred(x )=k(x ,X) T (K+σ n ²I) -1 y, Among them, k(x ,X)=[k_total(x , ),k_total(x , ),...,k_total(x ,xN)], indicating the test point x The vector composed of the composite kernel function value of each training sample point, k(x ,X) T Represents vector k(x ,X), K is the kernel function matrix obtained according to the training sample points, σ n ² is the observation noise variance, I is the unit matrix, (K+σ n ²I) -1 Indicates the matrix (K+σ n ²I), y is the inverse matrix of the training data moment value vector.

4. The method according to claim 3, characterized in that The sliding surface is constructed based on the deviation between the actual motion state of the arm joint and the expected motion state. The adaptive gain control structure and the variable saturation function of the boundary layer based on the external torque are used to generate the control torque in combination with the predicted joint torque value, including: The sliding surface is constructed based on the following formula: s = + λe, where e = q – q d is the position error, = - d is the velocity error, q and are the actual joint positions and velocities, q d and d are the desired joint positions and velocities, respectively, and λ is a preset positive constant; The torque compensation is generated based on the following formula: u = -K(s)·sat(s / Φ), Where K(s) is the adaptive gain function, Φ is the boundary layer width, and sat(s / Φ) is the variable saturation function of the boundary layer; Combined with the predicted joint torque value τ_pred(x ) and torque compensation, the control torque τ_control is obtained: τ_control = u +τ_pred(x )。 5. The method according to claim 4, characterized in that The gain function K(s) is determined according to the following formula: K(s) = Kmin + β yes + γ·m load (what ), Among them, Kmin is the minimum gain value, β is the adaptive gain parameter, γ is the preset load coefficient, m load (q ) is the load term, q is the actual joint position.

6. The method according to claim 4, characterized in that The saturation function sat(s / Φ) is determined according to the following formula: sat(s / Φ) = s / Φ, then s ≤ Φ, sat(s / Φ) = sign(s), when s > Φ, Among them, sign is the sign function, Φ= ·exp(-α s )·(1 - η· τext ), is the initial boundary layer width, α is the preset attenuation coefficient, η is the preset safety factor, and τext is the external torque.

7. The method according to claim 5, characterized in that According to the deviation between the control torque and the actual measured torque, the sliding mode control parameters are dynamically adjusted to adaptively adjust the control torque, including: The torque deviation Δτ is obtained according to the following formula: Δτ = τ_control - τ_total, Among them, τ_control is the control torque, and τ_total is the joint torque measured in real time; when Δτ > Δτ0, reduce the adaptive gain parameter β to reduce the control torque, when Δτ ≤ Δτ0, increase the adaptive gain parameter β to increase the control torque until the adaptive gain parameter β reaches a preset nominal value, where Δτ0 is a preset torque deviation threshold.

Images