Adaptive anti-shake control method and system for the robotic arm of a neurosurgery robot

Abstract

The present application relates to the field of data processing technology, and discloses a method and system for adaptive anti-shake control of a neurosurgery robot arm. The method comprises: collecting tissue contact force, intracranial pressure change, and vascular pulsation frequency, fusing the jitter feature vector through a neurosurgery-specific extended Kalman filter algorithm, generating compensation strategy parameters through a neurosurgery scene adaptive prediction model, and outputting a control signal from a tissue-protective super-twisted sliding mode controller to drive an intracranial constraint flexible actuator to achieve adaptive anti-shake control. The present application significantly improves the jitter suppression accuracy and adaptive adjustment capability of the neurosurgery robot in a complex intracranial environment by constructing a neurosurgery-specific extended Kalman filter algorithm and an LSTM prediction model based on multimodal sensor fusion, combined with a tissue-protective super-twisted sliding mode controller.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method and system for adaptive anti-shake control of a neurosurgery robot arm. Background Art

[0002] With the rapid development of minimally invasive neurosurgery, surgical robots have been widely used in precision surgeries such as deep brain electrode implantation, brain tumor resection, and cerebrovascular intervention. Traditional anti-shake control methods for neurosurgical robots primarily employ fixed-parameter PID controllers or passive compensation techniques based on inertial sensors. These reduce surgical instrument vibration by detecting vibration signals at the end of the robotic arm and applying a counter-compensating force. Existing anti-shake control systems typically use a single sensor to acquire vibration information, employ linear control algorithms for compensation calculations, and implement vibration suppression through rigid actuators. Some advanced systems integrate force feedback sensors and optical tracking devices to detect instrument-tissue contact and spatial position changes, and employ Kalman filtering to fuse multi-sensor data. Some studies have also explored anti-shake methods based on predictive control and adaptive control, improving control performance by establishing system dynamic models and using parameter identification techniques.

[0003] However, existing anti-shake control technologies for neurosurgical robots have significant shortcomings. First, traditional methods fail to fully consider the unique characteristics of the neurosurgical surgical environment, ignoring the impact of physiological factors such as cerebrospinal fluid pulsation, intracranial pressure fluctuations, and vascular pulsation on robotic arm vibration. This results in a lack of targeted and adaptable anti-shake control. Second, existing control algorithms often use fixed parameter designs, failing to dynamically adjust control strategies based on different surgical scenarios (such as surgical depth, instrument type, and risk level of brain functional areas). This makes it difficult to ensure stable anti-shake performance in the complex intracranial environment. Third, traditional rigid compensation actuators are prone to secondary damage to sensitive brain tissue when operating in the confined intracranial space and lack tissue protection mechanisms and safety constraints. Finally, existing methods lack comprehensive consideration of the geometric constraints of the skull drilling point and the safety boundaries of brain functional areas. This makes it impossible to achieve accurate vibration compensation while meeting surgical safety requirements, limiting their effectiveness in high-risk neurosurgery. Summary of the Invention

[0004] This application provides a method and system for adaptive anti-shake control of a neurosurgical robot arm. This method significantly improves the vibration suppression accuracy and adaptive adjustment capability of the neurosurgical robot in complex intracranial environments by constructing a neurosurgery-specific extended Kalman filter algorithm and LSTM prediction model based on multimodal sensor fusion, combined with a tissue-protective super-warp sliding mode controller. Furthermore, through the use of an intracranial constrained flexible actuator and a brain functional area safety monitoring mechanism, the anti-shake control process effectively improves the protection level of sensitive brain tissue and surgical safety.

[0005] In the first aspect, the present application provides an adaptive anti-shake control method for the manipulator arm of a neurosurgical robot, which includes: collecting and processing signals of the contact status of the surgical environment through a multimodal sensor array to obtain tissue contact force, intracranial pressure change and vascular pulsation frequency; performing multi-source signal fusion processing through a neurosurgery-specific extended Kalman filter algorithm based on the tissue contact force, intracranial pressure change and vascular pulsation frequency to obtain a jitter feature vector; inputting the jitter feature vector into a neurosurgery scene adaptive prediction model for jitter trend prediction processing to obtain compensation strategy parameters; performing compensation control signal generation processing through a tissue protection super-twist sliding mode controller based on the compensation strategy parameters to obtain a control output signal; inputting the control output signal into an intracranial constraint flexible actuator for jitter compensation execution processing to realize adaptive anti-shake control of the manipulator end in the intracranial space.

[0006] In a second aspect, the present application provides an adaptive anti-shake control system for a neurosurgery robot arm, the adaptive anti-shake control system for a neurosurgery robot arm comprising:

[0007] An acquisition module is used to collect and process signals of the surgical environment contact status through a multimodal sensor array to obtain tissue contact force, intracranial pressure change, and vascular pulsation frequency;

[0008] a fusion module, configured to perform multi-source signal fusion processing based on the tissue contact force, intracranial pressure change, and vascular pulsation frequency using a neurosurgery-specific extended Kalman filter algorithm to obtain a jitter feature vector;

[0009] A prediction module, configured to input the jitter feature vector into a neurosurgery scenario adaptive prediction model to perform jitter trend prediction processing and obtain compensation strategy parameters;

[0010] a generating module, configured to generate a compensation control signal through a tissue-protective hypertwisted sliding mode controller according to the compensation strategy parameters to obtain a control output signal;

[0011] The execution module is used to input the control output signal into the intracranial constraint flexible actuator for jitter compensation execution processing, thereby realizing adaptive anti-shake control of the end of the robotic arm in the intracranial space.

[0012] In a third aspect, a neurosurgical robot arm adaptive anti-shake control device is provided, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory to enable the neurosurgical robot arm adaptive anti-shake control device to execute the above-mentioned neurosurgical robot arm adaptive anti-shake control method.

[0013] In a fourth aspect, a computer-readable storage medium is provided, wherein instructions are stored in the computer-readable storage medium, which, when executed on a computer, enables the computer to execute the above-mentioned adaptive anti-shake control method for the manipulator arm of a neurosurgical robot.

[0014] The technical solution provided in this application uses a multimodal sensor array to collect and process signals of the surgical environment contact state, enabling simultaneous acquisition of physiological parameters unique to neurosurgery, such as tissue contact force, intracranial pressure change, and vascular pulsation frequency. This significantly improves the accuracy and comprehensiveness of jitter source identification compared to traditional single-sensor solutions. A neurosurgery-specific extended Kalman filter algorithm performs multi-source signal fusion processing tailored to the characteristics of the intracranial environment. By introducing a dynamic equation for cerebral vascular pulsation compensation and brain tissue-specific covariance analysis, it effectively filters out physiological interference such as cerebrospinal fluid pulsation and vascular pulsation, addressing the problem of inaccurate feature extraction in complex intracranial environments with traditional Kalman filter methods. The neurosurgery scenario adaptive prediction model utilizes a specially designed LSTM network structure, comprising a brain tissue memory unit, an intracranial pressure state gate, and a vascular pulsation forget gate. It can dynamically adjust the prediction strategy based on the type of surgical instrument, surgical depth, and risk level of brain functional areas, enabling jitter trend prediction to achieve millisecond-level time accuracy, effectively overcoming the lack of adaptability of traditional linear prediction methods in neurosurgery scenarios. The tissue-protective super-twisted sliding mode controller uses the hyperbolic tangent function to perform nonlinear constraint transformation and vascular pulsation frequency compensation, which not only ensures the rapid convergence characteristics of the control system but also avoids overcompensation impact on sensitive brain tissue. It shortens the control response time to the millisecond level while ensuring that the tissue contact force is always within a safe range.

[0015] The intracranial constraint flexible actuator utilizes a series structure of a shape memory alloy driver and a magnetorheological damper, combined with 7-DOF redundant zero-space motion planning for a robotic arm. This achieves micron-level jitter compensation while meeting geometric constraints at the cranial drilling point. This effectively addresses the operational difficulties and high risk of tissue damage associated with traditional rigid actuators within the confined intracranial space. Adaptive damping significantly reduces oscillation and overshoot during the compensation process. The core advantage of this overall solution in neurosurgical robotic applications lies in the algorithm's neurosurgical scenario-specific design. The neurosurgery-specific extended Kalman filter algorithm is optimized for the periodic characteristics of intracranial pressure fluctuations and vascular pulsation. The brain tissue memory unit of the LSTM prediction model learns the differences in jitter patterns across different brain regions. The tissue-protective design of the super-warp sliding mode controller ensures control safety in the complex intracranial environment. The synergistic effect of these algorithmic features ensures excellent adaptability and stability in high-precision neurosurgery procedures such as deep brain electrode implantation and brain tumor resection. Compared with the existing general anti-shake control method, the present invention achieves more precise jitter suppression effect and higher surgical safety in neurosurgery application scenarios, provides reliable technical support for precise surgical operations in complex intracranial environments, and significantly improves the clinical application effect of neurosurgery robots. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0017] Figure 1 This is a schematic diagram of an embodiment of a method for adaptive anti-shake control of a neurosurgery robot arm in an embodiment of the present application;

[0018] Figure 2 This is a schematic diagram of an embodiment of an adaptive anti-shake control system for a neurosurgery robot arm in an embodiment of the present application;

[0019] Figure 3 It is a schematic block diagram of the structure of the adaptive anti-shake control device of the neurosurgery robot manipulator in an embodiment of the present invention. DETAILED DESCRIPTION

[0020] An embodiment of the present application provides a method and system for adaptive anti-shake control of a neurosurgical robot arm. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or that are inherent to these processes, methods, products or devices.

[0021] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In one embodiment of the present application, an adaptive anti-shake control method for a neurosurgery robot arm includes:

[0022] Step S101: collecting and processing signals of the surgical environment contact state through a multimodal sensor array to obtain tissue contact force, intracranial pressure change, and vascular pulsation frequency;

[0023] Step S102: performing multi-source signal fusion processing using a neurosurgery-specific extended Kalman filter algorithm based on tissue contact force, intracranial pressure change, and vascular pulsation frequency to obtain a jitter feature vector;

[0024] Step S103: Input the jitter feature vector into the neurosurgery scene adaptive prediction model to perform jitter trend prediction processing to obtain compensation strategy parameters;

[0025] Step S104: performing compensation control signal generation processing through a tissue-protective super-twisted sliding mode controller according to compensation strategy parameters to obtain a control output signal;

[0026] Step S105: input the control output signal into the intracranial constraint flexible actuator for vibration compensation execution processing, thereby realizing adaptive anti-shake control of the end of the robotic arm in the intracranial space.

[0027] It is understandable that the execution subject of this application can be a neurosurgery robot arm adaptive anti-shake control system, or a terminal or a server, which is not limited here. The embodiment of this application is described by taking the server as the execution subject as an example.

[0028] Specifically, a multimodal sensor array acquires and processes signals from the surgical environment's contact status. A MEMS accelerometer collects three-axis vibration signals from the robotic arm's end effector, sampling at a frequency of 1000 Hz, generating raw end vibration data containing position, velocity, and acceleration information. A six-axis force feedback sensor detects mechanical changes at the interface between the instrument and surgical tissue. Using the piezoresistive principle of the force sensor, the mechanical force is converted into a voltage signal, which is then converted to a digitized value of the tissue contact force via an analog-to-digital converter. An intracranial pressure monitoring device uses a pressure sensor to acquire intracranial pressure fluctuation signals, reflecting changes in intracranial hemodynamics. A bandpass filter removes low-frequency drift below 0.1 Hz and high-frequency noise above 50 Hz to extract the effective intracranial pressure change. The ECG monitoring device outputs ECG signals, identifies the heartbeat cycle using an R-wave peak detection algorithm, and calculates the heart rate reference frequency, which serves as a baseline reference for vascular pulsation. An optical tracking system monitors changes in the robotic arm's position coordinates in real time. The time-domain position data is converted into frequency-domain information via Fourier transform. Combined with the heart rate reference frequency, the periodic position fluctuations caused by vascular pulsation are identified, thereby determining the vascular pulsation frequency.

[0029] The neurosurgery-specific extended Kalman filter algorithm combines tissue contact force, intracranial pressure change, and vascular pulsation frequency according to a weighted coefficient based on intracranial depth to form a multidimensional neurosurgery state vector. This state vector, consisting of four dimensions: position, velocity, force, and pressure, is predicted in the time domain using dynamic equations for cerebral vascular pulsation compensation, which account for the periodic influence of cerebral vascular pulsation on the system state. During the prediction process, the algorithm calculates the predicted value and prediction error covariance matrix for the current state. These are then compared with the actual sensor observations. Brain tissue-specific covariance analysis is then performed to derive the neurosurgery-specific Kalman gain coefficient. This gain coefficient reflects the weighting between the predicted and observed values. When sensor measurement accuracy is high, the gain coefficient is larger, indicating greater trust in the observed value; when system model accuracy is high, the gain coefficient is smaller, indicating greater trust in the predicted value. The predicted state is corrected and updated using the gain coefficient to obtain a fused state estimate vector that integrates multi-sensor information and system dynamics. Finally, a nonlinear transformation is performed using a neurosurgery scene feature mapping algorithm to reduce the high-dimensional state vector to a jitter feature vector containing key jitter characteristics. The neurosurgery scenario-adaptive prediction model receives the jitter feature vector and first concatenates it with the surgical instrument type code and the surgical depth parameter to form the neurosurgery scenario input feature matrix. The surgical instrument type code uses one-hot encoding to convert different instrument types into digital vectors. The surgical depth parameter reflects the depth of the instrument's operation within the skull. The brain tissue memory unit of the LSTM neural network is specifically designed to learn the temporal characteristics of the brain tissue environment. This memory unit can memorize both long-term jitter patterns and short-term transient changes. The network calculates a hidden state vector through forward propagation, which contains an abstract representation of temporal information. The intracranial pressure state gate determines which historical information to retain or forget based on the current intracranial pressure state, while the vascular pulsation forget gate adjusts the information update rate based on the vascular pulsation frequency. After selective information filtering, a neurosurgery-specific cell state vector is obtained, which contains the most relevant feature information for the neurosurgery scenario. A fully connected layer network performs nonlinear mapping on the cell state vector. A composite function of the multi-layer neural network approximates the complex jitter prediction relationship and outputs a jitter trend prediction result. The neurosurgery adaptive law performs parameter matching calculations on the predicted results based on the biomechanical characteristics of different brain regions and the surgical risk level, and generates compensation strategy parameters including compensation intensity, response time and safety margin.

[0030] The tissue-protective super-warp sliding mode controller generates a compensation control signal based on the compensation strategy parameters. First, the compensation strategy parameters are ratioed with the real-time tissue contact force to obtain the tissue-protective sliding mode surface function. This function describes the degree to which the system state deviates from the desired trajectory. When the sliding mode surface function is zero, the system is in an ideal state. A hyperbolic tangent function is used to perform a nonlinear constraint transformation on the sliding mode surface function, mapping the unbounded sliding mode surface function to a bounded interval to prevent tissue damage caused by excessive control signals. This results in a tissue safety margin constraint coefficient. The super-warp algorithm for vascular pulsation frequency compensation uses cosine modulation to combine the tissue safety margin constraint coefficient with the vascular pulsation frequency to obtain a control gain coefficient for vascular pulsation compensation. This gain coefficient varies with the vascular pulsation cycle, reducing the control intensity during peak pulsation and increasing it during valley pulsation. The super-warp algorithm calculates the first control term by taking the square root of the sliding mode surface function and then using the sign function to determine the control direction to obtain a directional control flag. The second control term uses an integral operation to process the directional control flag, forming a continuous control action. The two control terms are vector-synthesized to produce a discontinuous control component with finite-time convergence. An exponential decay integrator smooths the discontinuous control component, eliminating high-frequency jitter in the control signal through the exponential decay characteristics, resulting in a continuous compensation control signal. Finally, a safety check is performed against the intracranial pressure constraint parameters to ensure that the control signal does not cause intracranial pressure to exceed the safe range, and the final control output signal is output.

[0031] After receiving the control output signal, the intracranial constrained flexible actuator first decomposes it into X, Y, and Z components, and verifies its feasibility using a constrained kinematic model of the cranial burr point. This model ensures that all compensatory motions strictly adhere to the linear constraints from the cranial burr point to the target position, preventing the instrument from deviating from the predetermined trajectory and damaging surrounding tissue. Upon successful verification, a constraint satisfaction flag is generated, which controls the activation of subsequent actuators. The shape memory alloy actuator is activated by current modulation based on the constraint satisfaction flag. By controlling the heating current, the alloy's crystal structure is altered, producing a preset shape change and achieving micron-level precision compensation displacement. The magnetorheological damper adjusts the rheological properties of the magnetorheological fluid by controlling the excitation current. Increasing the excitation current increases the viscosity of the magnetorheological fluid and the damping coefficient, thereby achieving adaptive damping parameters. Zero-space motion planning for the 7-DOF redundant manipulator utilizes the manipulator's redundant degrees of freedom. While satisfying the end position constraints, an optimization algorithm is used to plan the compensatory motion trajectory in the joint space and generate intracranial compensation motion commands. After receiving the motion instructions, the robotic arm joint controller accurately controls the angles of each joint through the servo motor to achieve real-time anti-shake compensation action at the end of the robotic arm.

[0032] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0033] The three-axis vibration signal of the end effector of the robot arm is sampled and processed at high frequency by the MEMS accelerometer to obtain the original vibration data of the end effector;

[0034] The contact interface between the instrument and the surgical tissue is detected and processed using a 6-axis force feedback sensor to obtain the tissue contact force.

[0035] The pressure fluctuation signal collected by the intracranial pressure monitoring device is processed by a digital filter to extract the signal and obtain the intracranial pressure change;

[0036] The physiological signal output by the ECG monitoring device is analyzed and processed periodically using a peak detection algorithm to obtain the heart rate reference frequency;

[0037] The position data of the optical tracking system is subjected to frequency domain transformation according to the heart rate reference frequency to obtain the blood vessel pulsation frequency.

[0038] Specifically, a MEMS accelerometer is an inertial sensor manufactured using microelectromechanical systems technology. It contains a microstructure consisting of a mass, spring, and capacitor plates. When the robotic arm tip vibrates, the mass moves due to inertia, changing the distance between the capacitor plates and causing a change in capacitance. The sensor converts this capacitance change into a voltage signal. The triaxial vibration signal is then digitized using an analog-to-digital converter at a sampling frequency of 1000 Hz, generating raw end vibration data containing acceleration values ​​in the X, Y, and Z directions. Each sampling point includes a timestamp and the corresponding acceleration amplitude, forming a continuous time series. A six-axis force feedback sensor operates based on the strain gauge principle. Multiple strain gauges are installed within the sensor. When the instrument contacts the surgical tissue, the sensor undergoes a slight deformation due to the force, and the resistance of the strain gauges changes accordingly. This resistance change is converted into a voltage difference signal using a Wheatstone bridge circuit. This signal is then amplified by a signal amplifier and fed into the data acquisition system. The sensor simultaneously detects the force components and torque components in the three directions. Matrix operations are used to convert the six-dimensional torque data into the actual contact force of the instrument tip, resulting in a numerical representation of the tissue contact force. During the data collection process, the sensor continuously monitors the changing trend of the force and records the magnitude, direction and point of application of the force.

[0039] The intracranial pressure monitoring device uses a piezoresistive pressure sensor, which is implanted in the skull or obtains intracranial pressure signals through lumbar puncture. The sensor converts pressure changes into electrical signals, which contain multiple components such as heartbeat, respiration, and intracranial pressure fluctuations. The digital filter uses bandpass filtering technology, setting the passband range to 0.1Hz to 10Hz to filter out DC offset and high-frequency noise interference. During the filtering process, the signal first passes through a high-pass filter to remove slow drifts below 0.1Hz, and then passes through a low-pass filter to remove fast fluctuations above 10Hz. The filtered signal mainly reflects the periodic changes in intracranial pressure caused by heartbeat and respiration. The change in intracranial pressure is obtained by calculating the difference between the signals before and after filtering.

[0040] ECG monitoring equipment outputs a standard ECG signal, which reflects the temporal evolution of the heart's electrical activity. Peak detection algorithms specifically identify the R wave in the ECG signal. The R wave is the most prominent characteristic waveform in the ECG and represents the onset of ventricular depolarization. The algorithm first preprocesses the ECG signal to remove baseline drift and myoelectric interference. It then sets a threshold to detect peaks exceeding a specific amplitude. The RR interval is calculated by calculating the time interval between two adjacent R waves. The average RR interval is then divided by 60 seconds to calculate the heart rate (beats per minute), or the baseline heart rate. This rate reflects the rhythmic activity of the cardiovascular system and serves as a reference standard for vascular pulsation analysis.

[0041] The optical tracking system uses an infrared camera array to capture the spatial position coordinates of the end of the robotic arm in real time. The system records position data at a frequency of 200 Hz, forming a continuous position time series. The frequency domain transformation process uses the fast Fourier transform algorithm to convert the time domain position data into frequency domain spectrum lines. During the transformation process, the algorithm calculates the amplitude and phase information of each frequency component and identifies the periodic components in the position data. By comparing the frequency domain analysis results with the heart rate reference frequency, the peak of the spectrum line close to the heart rate frequency is found. This peak corresponds to the tiny position fluctuation of the end of the robotic arm caused by vascular pulsation. The algorithm extracts the center frequency of this frequency component as the vascular pulsation frequency and records its amplitude information to reflect the pulsation intensity.

[0042] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0043] The tissue contact force, intracranial pressure change and vascular pulsation frequency are weighted and combined according to the intracranial depth weight coefficient to obtain the neurosurgery multidimensional state vector;

[0044] Based on the multi-dimensional state vector of neurosurgery, the dynamic equation of cerebral vascular pulsation compensation is used to perform time domain prediction calculation and processing to obtain the intracranial environment prediction state matrix;

[0045] Perform brain tissue-specific covariance analysis on the intracranial environment prediction state matrix and measured sensor data to obtain neurosurgery-specific Kalman gain coefficients.

[0046] The predicted state matrix is ​​corrected under the intracranial pressure constraint according to the neurosurgery-specific Kalman gain coefficient to obtain the fusion state estimation vector;

[0047] The fusion state estimation vector is processed by nonlinear transformation through the neurosurgery scene feature mapping algorithm to obtain the jitter feature vector.

[0048] Specifically, the intracranial depth weight coefficient is dynamically calculated based on the operating depth of the surgical instrument within the skull. The deeper the depth, the smaller the weight coefficient, reflecting the reduced sensitivity of deep tissue to external interference. The weighted combination process multiplies the tissue contact force by the corresponding force weight coefficient, the intracranial pressure change by the pressure weight coefficient, and the vascular pulsation frequency by the frequency weight coefficient. The three weighted results are then linearly combined according to a preset ratio. During the combination process, the force component accounts for 40% of the total weight, the pressure component accounts for 30%, and the frequency component accounts for 30%. Ultimately, a multidimensional neurosurgical state vector containing four dimensions: position, velocity, force, and pressure is obtained.

[0049] The dynamic equations for cerebral vascular pulsation compensation specifically account for the periodic effects of vascular pulsation on the system state. The equation structure consists of the state transition matrix multiplied by the current state vector plus the control input matrix multiplied by the control vector, plus the vascular pulsation interference term. The vascular pulsation interference term takes the form of a sinusoidal function, with a frequency corresponding to the vascular pulsation frequency and an amplitude determined by the vascular pulsation intensity. Time-domain prediction calculations advance the dynamic equations by one sampling period on the time axis to calculate the predicted state value for the next moment. During the prediction process, the algorithm uses the current state vector and control input, combined with the system's dynamic characteristics, to infer the future state value. Simultaneously, a prediction error covariance matrix is ​​calculated, reflecting the degree of uncertainty in the prediction results. The diagonal elements represent the predicted variance of each state component, while the off-diagonal elements represent the correlation between state components. Ultimately, the predicted state matrix for the intracranial environment is formed.

[0050] The brain tissue-specific covariance analysis process is designed specifically for the biomechanical properties of brain tissue. The analysis process calculates the residuals between the predicted state matrix of the intracranial environment and the measured sensor data to obtain a prediction error vector. Covariance analysis calculates the statistical characteristics of the prediction error, including the error mean, variance, and covariance information. During the analysis, the algorithm adjusts the parameters of the covariance matrix based on the viscoelastic properties of brain tissue, with larger covariance values ​​corresponding to soft tissue regions and smaller covariance values ​​corresponding to hard tissue regions. The numerator and denominator of the Kalman gain are calculated by performing a matrix operation on the prediction error covariance matrix and the observation noise covariance matrix. The neurosurgery-specific Kalman gain coefficient is equal to the prediction error covariance matrix multiplied by the transpose of the observation matrix, divided by the sum of the observation matrix multiplied by the prediction error covariance matrix multiplied by the transpose of the observation matrix, plus the observation noise covariance matrix.

[0051] The state correction process under the intracranial pressure constraint multiplies the neurosurgery-specific Kalman gain coefficient with the observation residual to obtain the state correction amount. The observation residual is equal to the actual sensor measurement value minus the predicted observation value calculated by the predicted state matrix through the observation equation. During the correction process, the algorithm checks whether the corrected state meets the intracranial pressure safety constraint. When the intracranial pressure exceeds the safety range, the algorithm automatically reduces the correction amplitude of the relevant state components. The state correction adopts a component-by-component processing method. The position component correction has the highest priority, followed by the velocity component. The force and pressure component corrections need to meet physiological safety limits. The correction calculation adds the predicted state vector to the Kalman gain matrix and multiplies it by the observation residual vector. At the same time, the covariance matrix of the state estimate is updated to obtain the fused state estimate vector.

[0052] The neurosurgery scene feature mapping algorithm uses nonlinear transformation technology to convert the high-dimensional fused state estimate vector into a low-dimensional jitter feature representation. The mapping algorithm first normalizes the fused state estimate vector, unifying the numerical range of each component to between zero and one. Principal component analysis is then used to identify the main direction of change in the state vector, and the principal component containing the maximum variance information is extracted as the feature basis. The nonlinear transformation uses a radial basis function network structure, with the input being the standardized state vector and the output being a fixed-dimensional feature vector. During the transformation process, the hidden layer nodes of the network correspond to different jitter patterns, and the output layer performs a weighted combination of the hidden layer activation values ​​to form the final jitter feature vector.

[0053] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0054] The jitter feature vector is combined with the surgical instrument type code and surgical depth parameters to obtain the neurosurgery scene input feature matrix;

[0055] Based on the neurosurgery scene input feature matrix, the brain tissue memory unit of the LSTM neural network performs temporal feature learning processing to obtain the hidden state vector;

[0056] The hidden state vector is selectively filtered through the intracranial pressure state gate and the vascular pulsation forget gate to obtain the neurosurgery-specific cell state vector;

[0057] According to the neurosurgery-specific cell state vector, a nonlinear mapping transformation is performed through a fully connected layer network to obtain the jitter trend prediction result;

[0058] The jitter trend prediction results are matched with the neurosurgery adaptive law to obtain the compensation strategy parameters.

[0059] Specifically, the splicing and combination process uses the jitter feature vector as the base data. The surgical instrument type encoding uses one-hot encoding to convert different instruments into digital vectors. The electrode introducer is encoded as a vector of 100, the biopsy needle is encoded as 010, and the microscissors are encoded as 001. The surgical depth parameter is directly represented using a numerical value in millimeters. A matrix splicing operation combines the three data sources column-wise to form a neurosurgery scenario input feature matrix containing jitter features, instrument information, and depth information. Each row of the matrix represents the complete feature information for a time point, and the number of columns is equal to the jitter feature dimension plus the instrument encoding dimension plus the depth parameter dimension.

[0060] The brain tissue memory unit of the LSTM neural network is specifically designed to handle the temporal dependencies found in neurosurgery. The unit contains an input gate, an output gate, and a candidate value calculation module. Temporal feature learning processes the neurosurgical scene input feature matrix, which is fed into the network in chronological order. The feature vector at each time step passes through the input gate, which determines which new information should be stored in the cell state. The input gate's activation function uses a sigmoid function, mapping the input signal to a range between zero and one. Values ​​closer to one indicate greater importance of the corresponding feature. Candidate value calculation uses the tanh activation function to generate new candidate information, which describes the potential state update at the current moment. The network multiplies the input gate output by the candidate value to determine the actual amount of information added to the cell state. The output gate calculates output weights based on the current input and the previous hidden state, determining which portion of the cell state is used as the output at the current moment. The hidden state vector is equal to the output gate activation value multiplied by the tanh function applied to the cell state. This vector contains all relevant temporal information from the beginning of the input sequence to the current moment.

[0061] The intracranial pressure (ICP) state gate calculates a gating signal based on the deviation of the current ICP measurement from the normal ICP range. When ICP exceeds the normal range, the state gate reduces its retention of historical information, highlighting the importance of the current abnormal state. The vascular pulsation forget gate adjusts the rate of information forgetting based on the amplitude of changes in vascular pulsation frequency. When the pulsation frequency is stable, the forget gate maintains a low forgetting rate, retaining more historical pattern information; when the pulsation frequency changes rapidly, the forget gate increases the forgetting rate, quickly adapting to the new pulsation pattern. Selective information filtering (SIF) performs element-wise multiplication of the hidden state vector with the outputs of the ICP state gate and the vascular pulsation forget gate, respectively. The state gate controls the state component related to ICP, while the forget gate controls the state component related to vascular pulsation. The filtered information is combined through weighted summation into a neurosurgery-specific cell state vector, which highlights the most relevant features of the current neurosurgery scenario.

[0062] The fully connected layer network consists of three layers. The input layer receives the neurosurgery-specific cell state vector, the hidden layer performs nonlinear transformation through the ReLU activation function, and the output layer generates the jitter trend prediction result. The nonlinear mapping transformation process first multiplies the cell state vector with the first-layer weight matrix and adds the bias vector, and then uses the ReLU function to set negative values ​​to zero, retaining positive features. The second layer continues to perform linear transformation and ReLU activation to further extract high-order feature representations. The output layer uses a linear activation function to directly output the predicted value, which includes three components: position jitter amplitude, velocity jitter frequency, and force jitter intensity at the future moment. The network trains weight parameters through the backpropagation algorithm, and the loss function uses the mean square error to measure the difference between the predicted value and the actual observed value.

[0063] The neurosurgery adaptive law formulates parameter matching rules based on the biomechanical properties of different brain regions and the surgical risk level. The motor cortex corresponds to a high risk level, the language area corresponds to an extremely high risk level, and non-functional areas correspond to a moderate risk level. The parameter matching calculation multiplies the position jitter component from the jitter trend prediction results by the position compensation gain, the velocity jitter component by the velocity compensation gain, and the force jitter component by the force compensation gain. The compensation gain is dynamically adjusted based on the surgical risk level. High-risk areas use smaller compensation gains to avoid overcompensation, while low-risk areas use larger compensation gains to enhance jitter suppression. The calculation process also considers the inertial characteristics of surgical instruments, with heavier instruments receiving smaller response time constants and lighter instruments receiving larger response time constants. The final compensation strategy parameters output include the triaxial compensation strength, response time constant, and safety margin threshold.

[0064] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0065] The compensation strategy parameters are processed with the real-time tissue contact force in a ratio operation to obtain the tissue protective sliding surface function;

[0066] Based on the tissue protective sliding surface function, the nonlinear constraint transformation is processed by hyperbolic tangent function to obtain the tissue safety boundary limit coefficient;

[0067] The tissue safety boundary limit coefficient is processed by the control law calculation of the super-distortion algorithm of vascular pulsation frequency compensation to obtain the discontinuous control component;

[0068] The intermittent control component is smoothed by an exponential decay integrator to obtain a continuous compensation control signal;

[0069] The continuous compensation control signal is subjected to safety verification processing with the intracranial pressure constraint parameter to obtain a control output signal.

[0070] Specifically, the ratio operation process uses the compensation intensity component in the compensation strategy parameter as the numerator and the real-time tissue contact force as the denominator for division calculation. When the tissue contact force is large, the ratio is small, and when the tissue contact force is small, the ratio is large. This ratio reflects the relative relationship between the required compensation intensity and the current tissue load. The tissue protective sliding surface function is defined as a linear combination of the tracking error and the result of the ratio operation. The tracking error is equal to the expected position minus the actual position. A sliding surface function value of zero indicates that the robotic arm is in an ideal tracking state. The degree to which the function value deviates from zero reflects the size of the deviation and the intensity of the compensation requirement. The calculation process of the sliding surface function takes into account the protective effect of the tissue contact force. When the contact force increases, the slope of the sliding surface automatically decreases, reducing the control response intensity and avoiding excessive impact on sensitive tissues.

[0071] The hyperbolic tangent function maps the unbounded tissue-protective sliding surface function value to a bounded interval between negative one and positive one. The function takes the form of a hyperbolic tangent function with the sliding surface function value as the independent variable. Nonlinear constrained transformation processing utilizes the saturation characteristics of the hyperbolic tangent function. When the sliding surface function value is small, the transformation result is approximately equal to the input value, maintaining a linear relationship; when the sliding surface function value is large, the transformation result approaches the saturation value, limiting the output amplitude. The tissue safety boundary limit coefficient is equal to the output of the hyperbolic tangent transformation. This coefficient automatically adjusts the intensity range of the control signal to ensure that the compensation action does not exceed the tissue safety tolerance. During the transformation process, the steepness of the hyperbolic tangent function is determined by a preset safety factor. A larger safety factor is used in soft tissue areas to obtain a flatter transformation curve, while a smaller safety factor is used in hard tissue areas to obtain a steeper transformation characteristic.

[0072] The super-distorted algorithm for vascular pulsation frequency compensation specifically designs a control law for vascular pulsation interference. The algorithm modulates the tissue safety boundary limit coefficient and the vascular pulsation frequency with a cosine function. The modulated result serves as the control gain to dynamically adjust the strength of the super-distorted control law. The control law calculation process includes the calculation of two control terms. The first control term takes the square root of the sliding surface function and multiplies it by the sign function. The square root operation reduces the growth rate of the control signal, and the sign function determines the control direction. The second control term directly integrates the sign of the sliding surface function. The integration process accumulates the control effect and compensates for the system's modeling errors and external interference. The two control terms are vector-synthesized according to preset weight coefficients. The first control term has a larger weight for rapid response, while the second control term has a smaller weight for precise tracking. The resulting synthesis results in a discontinuous control component with finite-time convergence characteristics.

[0073] The exponential decay integrator uses an exponential function as the integral kernel function and performs time-domain integration operations on the discontinuous control component. During the integration process, the weight of the recent control signal is relatively large, while the weight of the historical control signal decays exponentially. Smoothing processing eliminates high-frequency jump components in the discontinuous control component through integration operations. The time constant of the integrator determines the degree of smoothing. When the time constant is small, the smoothing effect is weaker but the response speed is faster. When the time constant is large, the smoothing effect is stronger but the response speed is slower. The continuous compensation control signal is equal to the output of the exponential decay integrator. This signal maintains the control performance of the original control law while being continuous and differentiable, avoiding impact on the actuator. The integrator also has an anti-saturation function. When the integral value exceeds the preset range, it automatically limits the amplitude to prevent excessive control signals from damaging the actuator.

[0074] The safety verification process converts the continuous compensation control signal into the corresponding predicted value of the intracranial pressure change. By establishing a mapping relationship between the control signal and the intracranial pressure change, the impact of the compensation action on the intracranial pressure is evaluated. The verification process compares the predicted intracranial pressure change with the safety threshold. When the predicted change exceeds the safety range, the verification module automatically reduces the amplitude of the control signal to ensure that the intracranial pressure always remains within the physiological safety range. The intracranial pressure constraint parameters include the maximum allowable pressure change, the pressure change rate limit, and the cumulative pressure deviation threshold. The verification process must simultaneously meet all three constraints. If any constraint condition is not met, the verification module adjusts the control signal according to the constraint priority. The absolute pressure constraint has the highest priority, the pressure change rate constraint has the second highest priority, and the cumulative deviation constraint has the lowest priority. The control output signal output after safety verification maintains the effectiveness of anti-shake control while ensuring the safety requirements of neurosurgery.

[0075] In a specific embodiment, the step of performing control law calculation processing on the tissue safety boundary limit coefficient using a super-distortion algorithm for vascular pulsation frequency compensation to obtain a discontinuous control component may specifically include the following steps:

[0076] The tissue safety boundary limit coefficient and the blood vessel pulsation frequency are modulated by a cosine function to obtain a control gain coefficient for blood vessel pulsation compensation;

[0077] Performing square root operation on the sliding surface function based on the control gain coefficient of vascular pulsation compensation to obtain the first control term of the super-distortion algorithm;

[0078] The first control item of the super-twist algorithm is processed by a sign function to determine the direction and obtain a direction control flag.

[0079] According to the direction control flag, an integral operation is performed through the calculation formula of the second control term of the super-twisted algorithm to obtain the second control term of the super-twisted algorithm;

[0080] The first control item of the super-distortion algorithm and the second control item of the super-distortion algorithm are subjected to vector synthesis processing to obtain a discontinuous control component.

[0081] Specifically, the cosine function modulation process uses the tissue safety boundary limit coefficient as the modulation amplitude, and the blood vessel pulsation frequency multiplied by the current time and then multiplied by twice pi as the phase angle of the cosine function. The modulation formula is:

[0082] in represents the control gain coefficient of vascular pulsation compensation, represents the organizational safety boundary limitation coefficient, represents the vascular pulsation frequency, and t represents the current time. During the modulation process, the cosine value approaches positive one when the vascular pulsation is in the systolic phase, and approaches negative one when the vascular pulsation is in the diastolic phase. The modulated signal fluctuates periodically with the vascular pulsation. The control gain coefficient for vascular pulsation compensation dynamically adjusts the control intensity based on the vascular pulsation phase, reducing the control gain at the vascular pulsation peak to avoid superimposed interference and increasing the control gain at the vascular pulsation valley to enhance jitter suppression. Modulation processing ensures that the control signal remains synchronized with the vascular pulsation, and phase compensation reduces the impact of vascular pulsation on the positioning accuracy of the robotic arm.

[0083] The square root operation takes the absolute value of the sliding surface function as the square root, and the square root result is obtained by the square root function. The calculation formula is:

[0084] in Represents the first control item of the super-warping algorithm, represents the sliding surface function value, Represents a sign function. The square root operation has the mathematical property of reducing the growth rate of a function. When the sliding surface function value is small, the square root result is close to the original value. When the sliding surface function value is large, the square root result is significantly smaller than the original value. The control gain coefficient of vascular pulsation compensation is multiplied by the square root result to obtain a gain-adjusted square root value, which reflects the control strength requirement after considering the influence of vascular pulsation. The amplitude of the first control item of the super-warp algorithm is determined by the square root operation, and the direction is determined by the sign function. This control item is mainly responsible for quickly driving the system state close to the sliding surface.

[0085] The sign function performs direction judgment processing and inputs the value of the first control item of the super-distortion algorithm into the sign function. The mathematical expression of the sign function is:

[0086] in Represents the direction control flag. The sign function outputs three possible results: positive one, negative one, or zero. When the first control item is a positive number, positive one is output; when the first control item is a negative number, negative one is output; and when the first control item is zero, zero is output. The direction control flag indicates the direction of the current control action. Positive one indicates that a positive control action is required, negative one indicates that a negative control action is required, and zero indicates that no control action is required. The change in the flag reflects the change in the sign of the sliding surface function. When the system state oscillates on both sides of the sliding surface, the flag switches between positive one and negative one accordingly, forming a typical sliding mode control feature. The direction judgment process ensures that the control algorithm always drives the system state in the correct direction, avoiding system divergence caused by incorrect control direction.

[0087] The calculation formula of the second control term of the super-twisted algorithm adopts the integral operation form, and the mathematical expression is:

[0088] in Represents the second control item of the super-warping algorithm, represents the integral gain coefficient, Represents the integral variable. The integral operation process performs time integration on the direction control flag starting from the moment the algorithm is started. The integral step is equal to the sampling period of the control system. In each sampling period, the current flag value is multiplied by the sampling period and then added to the integral cumulative value. When the direction control flag is positive one, the integral value increases; when the flag is negative one, the integral value decreases; when the flag is zero, the integral value remains unchanged. The integral gain coefficient is determined according to the dynamic characteristics of the system and the expected convergence speed. The second control term of the super-twist algorithm is mainly responsible for eliminating the steady-state error of the system and resisting continuous interference. It ensures that the system eventually converges to the sliding surface through the integral action.

[0089] The vector synthesis process linearly combines the first control item and the second control item of the super distortion algorithm according to the preset weight coefficient. The synthesis formula is:

[0090] in represents the discontinuous control component, represents the weight coefficient of the first control item, Represents the weight coefficient of the second control item. The weight coefficient of the first control item is usually set to a large value to ensure fast response characteristics, while the weight coefficient of the second control item is set to a relatively small value to ensure steady-state accuracy. The discontinuous control component inherits the fast convergence characteristics of the first control item and the steady-state accuracy characteristics of the second control item, forming a composite control signal with finite time convergence capability. The phase relationship between the two control items is also considered in the vector synthesis process. By adjusting the ratio of the weight coefficients to balance the requirements of rapidity and stability, an intermittent control component suitable for the precision requirements of neurosurgery is obtained.

[0091] In a specific embodiment, the process of executing step S1065 may specifically include the following steps:

[0092] The control output signal is decomposed into three-axis components of X, Y, and Z, and the feasibility verification process is carried out through the kinematic model of the skull drilling point constraint to obtain the constraint satisfaction flag.

[0093] Based on the constraint satisfaction flag, the shape memory alloy actuator is activated by current modulation to obtain a fine-tuning compensation displacement vector;

[0094] The fine-tuning compensation displacement vector is processed by adjusting the damping coefficient through the excitation current control of the magnetorheological damper to obtain the adaptive damping parameters;

[0095] Based on the adaptive damping parameters, the compensation trajectory is generated through the 7-DOF redundant manipulator zero-space motion planning to obtain the intracranial space compensation motion instructions;

[0096] The intracranial space compensation motion command is input into the robotic arm joint controller for real-time execution processing to obtain the anti-shake compensation action of the robotic arm end.

[0097] Specifically, the control output signal decomposition process splits the three-dimensional control vector into three independent components of the X-axis, Y-axis, and Z-axis. Each component represents the compensation demand intensity in the direction of the corresponding coordinate axis. The skull drilling point constraint kinematic model is established based on the geometric constraint relationship from the skull drilling point to the surgical target point. The model requires that the instrument motion trajectory must pass through the drilling point and maintain a straight path. The feasibility verification process determines the feasibility of the movement by calculating whether the compensated instrument position deviates from the predetermined straight line trajectory. The verification algorithm calculates the distance between the instrument tip position and the ideal straight line. When the distance exceeds the safety threshold, the verification fails. When the distance is within the safety range, the verification passes. The constraint satisfaction flag is binary coded. The flag is set to one when the verification passes and is set to zero when the verification fails. This flag controls the activation state of the subsequent actuator.

[0098] A shape memory alloy actuator is a smart material actuator that achieves shape change by changing the alloy's crystal structure through electric current heating. The shape memory alloy wire inside the actuator is in the martensite phase at low temperatures and transforms into the austenite phase at high temperatures, restoring the preset shape. The current modulation activation process determines whether to apply an activation current to the actuator based on the state of the constraint satisfaction flag. When the flag is one, the current modulation module outputs the preset activation current, and when the flag is zero, the current output is zero. The magnitude of the activation current is dynamically adjusted based on the amplitude of the required compensation displacement. There is a nonlinear relationship between current and displacement, and the current value is determined by table lookup or polynomial fitting methods. The fine-tuning compensation displacement vector is equal to the three-dimensional displacement output generated by the shape memory alloy actuator under the action of the activation current. The components of this vector correspond to the micron-level displacement adjustment in the X, Y, and Z directions.

[0099] A magnetorheological damper consists of two main components: magnetorheological fluid (MRF) and an electromagnetic coil. MRF is a special fluid containing micron-sized magnetic particles. In the absence of a magnetic field, the magnetic particles align to form a chain-like structure, causing the fluid's viscosity to increase dramatically. Excitation current control modulates the magnetic field strength by adjusting the excitation current of the electromagnetic coil, thereby controlling the rheological properties of the MRF. The damping coefficient adjustment process uses the amplitude of the fine-tuned compensation displacement vector as an input parameter and determines the corresponding excitation current intensity through a mapping relationship. Larger compensation displacements correspond to stronger excitation currents, resulting in greater damping. The adaptive damping parameter is equal to the equivalent damping coefficient of the MR damper at the current excitation current. This parameter changes dynamically with the compensation demand. When the compensation amplitude is large, the damping parameter increases to suppress overshoot, and when the compensation amplitude is small, the damping parameter decreases to ensure a fast response.

[0100] The 7-DOF redundant manipulator has seven rotational joints, an additional redundant joint compared to a 6-DOF manipulator. The presence of these redundant joints enables internal reconfiguration while maintaining the end effector's position. Null-space motion planning is a core technology for redundant manipulator control. Null space refers to the subspace of joint motions that does not affect the end effector's position and posture. Joint motions within this null space do not alter the end effector's position. Compensatory trajectory generation utilizes the redundant joints to plan compensatory motions within this null space. The Jacobian matrix for the current joint configuration is first calculated, followed by the null-space basis vectors of the Jacobian matrix. These null-space basis vectors span the joint motion directions that do not affect the end effector's position. The trajectory generation algorithm searches for the optimal joint motion within the null space that satisfies the adaptive damping parameter requirements. An optimization algorithm is then used to minimize joint velocities and accelerations while satisfying the damping constraints. The intracranial space compensatory motion instructions contain the angle increments and angular velocity commands for the seven joints, describing the coordinated motion pattern of each joint in the manipulator.

[0101] The robotic arm joint controller utilizes a dual-loop position-velocity control structure. The outer loop, a position controller, tracks joint angles, while the inner loop, a velocity controller, regulates joint angular velocities. Real-time processing decomposes intracranial space compensation motion commands into reference signals for each joint. The position controller calculates the deviation between the current joint angle and the reference angle and generates a velocity reference signal using a proportional-integral-derivative control algorithm. The velocity controller receives the velocity reference signal and actual joint angular velocity feedback, calculates the velocity deviation, and generates the motor drive torque using a control algorithm. The servo motors in each joint synchronously execute compensation motions based on the drive torque commands. The motor encoders provide real-time feedback on joint position and velocity, forming a closed-loop control loop. The end-of-arm anti-shake compensation motion is the result of coordinated motions of seven joints. Precise joint control enables micron-level jitter compensation within the confined intracranial space.

[0102] The above describes the adaptive anti-shake control method of the neurosurgery robot manipulator in the embodiment of the present application. The following describes the adaptive anti-shake control system of the neurosurgery robot manipulator in the embodiment of the present application. Figure 2 In one embodiment of the present application, an adaptive anti-shake control system for a neurosurgery robot arm includes:

[0103] An acquisition module is used to collect and process signals of the surgical environment contact status through a multimodal sensor array to obtain tissue contact force, intracranial pressure change, and vascular pulsation frequency;

[0104] a fusion module, configured to perform multi-source signal fusion processing based on the tissue contact force, intracranial pressure change, and vascular pulsation frequency using a neurosurgery-specific extended Kalman filter algorithm to obtain a jitter feature vector;

[0105] A prediction module, configured to input the jitter feature vector into a neurosurgery scenario adaptive prediction model to perform jitter trend prediction processing and obtain compensation strategy parameters;

[0106] a generating module, configured to generate a compensation control signal through a tissue-protective hypertwisted sliding mode controller according to the compensation strategy parameters to obtain a control output signal;

[0107] The execution module is used to input the control output signal into the intracranial constraint flexible actuator for jitter compensation execution processing, thereby realizing adaptive anti-shake control of the end of the robotic arm in the intracranial space.

[0108] above Figure 2 The adaptive anti-shake control system of the neurosurgical robot arm in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The adaptive anti-shake control device of the neurosurgical robot arm in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0109] Reference Figure 3 In an embodiment of the present invention, a neurosurgery robot arm adaptive anti-shake control device is provided. The neurosurgery robot arm adaptive anti-shake control device can be a server, and its internal structure can be as follows: Figure 3 As shown. The neurosurgical robot arm adaptive anti-shake control device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the computer-designed processor is used to provide computing and control capabilities. The memory of the neurosurgical robot arm adaptive anti-shake control device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the neurosurgical robot arm adaptive anti-shake control device is used to store the corresponding data in this embodiment. The network interface of the neurosurgical robot arm adaptive anti-shake control device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0110] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the adaptive anti-shake control device for the mechanical arm of a neurosurgical robot to which the solution of the present invention is applied.

[0111] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores instructions, and when the instructions are run on a computer, the computer executes the steps of the adaptive anti-shake control method for the mechanical arm of a neurosurgical robot.

[0112] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0113] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a neurosurgery robot arm adaptive anti-shake control device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program code.

[0114] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for adaptive anti-shake control of a neurosurgery robot arm, characterized in that: The method comprises: The multimodal sensor array is used to collect and process signals of the surgical environment contact status to obtain tissue contact force, intracranial pressure change, and vascular pulsation frequency; Performing multi-source signal fusion processing using a neurosurgery-specific extended Kalman filter algorithm based on the tissue contact force, intracranial pressure change, and vascular pulsation frequency to obtain a jitter feature vector; Inputting the jitter feature vector into a neurosurgery scene adaptive prediction model to perform jitter trend prediction processing to obtain compensation strategy parameters; Performing compensation control signal generation processing through a tissue-protective super-twisted sliding mode controller according to the compensation strategy parameters to obtain a control output signal; The control output signal is input into the intracranial constraint flexible actuator for jitter compensation execution processing, thereby realizing adaptive anti-shake control of the end of the robotic arm in the intracranial space.

2. The method for adaptive anti-shake control of a neurosurgery robot arm according to claim 1, characterized in that: The method of collecting and processing signals of the surgical environment contact state by a multimodal sensor array to obtain tissue contact force, intracranial pressure change, and vascular pulsation frequency includes: The three-axis vibration signal of the end effector of the robot arm is sampled and processed at high frequency by the MEMS accelerometer to obtain the original vibration data of the end effector; The contact interface between the instrument and the surgical tissue is detected and processed using a 6-axis force feedback sensor to obtain the tissue contact force. The pressure fluctuation signal collected by the intracranial pressure monitoring device is processed by a digital filter to extract the signal and obtain the intracranial pressure change; The physiological signal output by the ECG monitoring device is analyzed and processed periodically using a peak detection algorithm to obtain the heart rate reference frequency; The position data of the optical tracking system is subjected to frequency domain transformation processing according to the heart rate reference frequency to obtain the blood vessel pulsation frequency.

3. The method for adaptive anti-shake control of a neurosurgery robot arm according to claim 1, characterized in that: The multi-source signal fusion processing is performed based on the tissue contact force, the intracranial pressure change, and the vascular pulsation frequency using a neurosurgery-specific extended Kalman filter algorithm to obtain a jitter feature vector, including: The tissue contact force, intracranial pressure change and vascular pulsation frequency are weighted and combined according to the intracranial depth weight coefficient to obtain a neurosurgery multidimensional state vector; Based on the multi-dimensional state vector of neurosurgery, a time domain prediction calculation process is performed through the dynamic equation of cerebral vascular pulsation compensation to obtain an intracranial environment prediction state matrix; Performing brain tissue-specific covariance analysis on the intracranial environment prediction state matrix and the measured sensor data to obtain a neurosurgery-specific Kalman gain coefficient; Performing state correction processing on the predicted state matrix under intracranial pressure constraints according to the neurosurgery-specific Kalman gain coefficient to obtain a fusion state estimation vector; The fusion state estimation vector is subjected to nonlinear transformation processing by a neurosurgery scene feature mapping algorithm to obtain a jitter feature vector.

4. The method for adaptive anti-shake control of a neurosurgery robot arm according to claim 1, wherein: The step of inputting the jitter feature vector into the neurosurgery scene adaptive prediction model to perform jitter trend prediction processing to obtain compensation strategy parameters includes: The jitter feature vector is combined with the surgical instrument type code and the surgical depth parameter to obtain a neurosurgery scene input feature matrix; Based on the neurosurgery scene input feature matrix, a temporal feature learning process is performed through the brain tissue memory unit of the LSTM neural network to obtain a hidden state vector; Performing selective information filtering on the hidden state vector through an intracranial pressure state gate and a vascular pulsation forget gate to obtain a neurosurgery-specific cell state vector; Performing nonlinear mapping transformation processing on the neurosurgery-specific cell state vector through a fully connected layer network to obtain a jitter trend prediction result; The jitter trend prediction result is subjected to parameter matching calculation processing with the neurosurgery adaptive law to obtain compensation strategy parameters.

5. The method for adaptive anti-shake control of a neurosurgery robot arm according to claim 1, wherein: The generating and processing of the compensation control signal by the tissue-protective super-twisted sliding mode controller according to the compensation strategy parameters to obtain the control output signal includes: Performing ratio calculation on the compensation strategy parameters and the real-time tissue contact force to obtain a tissue protective sliding surface function; Based on the tissue protective sliding surface function, a nonlinear constraint transformation process is performed through a hyperbolic tangent function to obtain a tissue safety boundary limit coefficient; The tissue safety boundary limit coefficient is processed by a control law calculation using a super-distortion algorithm for vascular pulsation frequency compensation to obtain an intermittent control component; Performing smoothing processing on the discontinuous control component through an exponential decay integrator to obtain a continuous compensation control signal; The continuous compensation control signal and the intracranial pressure constraint parameter are subjected to safety verification processing to obtain a control output signal.

6. The method for adaptive anti-shake control of a neurosurgery robot arm according to claim 5, characterized in that: The control law calculation and processing of the tissue safety boundary limit coefficient by a super-distortion algorithm of vascular pulsation frequency compensation to obtain a discontinuous control component includes: Performing cosine function modulation processing on the tissue safety boundary limit coefficient and the blood vessel pulsation frequency to obtain a control gain coefficient for blood vessel pulsation compensation; Performing a square root operation on the sliding surface function based on the control gain coefficient of the vascular pulsation compensation to obtain a first control item of the super-distortion algorithm; Performing direction determination processing on the first control item of the super-twist algorithm through a sign function to obtain a direction control flag bit; Performing an integral operation according to the direction control flag using a calculation formula for a second control term of a super-twisted algorithm to obtain a second control term of the super-twisted algorithm; The first control item of the super-distortion algorithm and the second control item of the super-distortion algorithm are subjected to vector synthesis processing to obtain a discontinuous control component.

7. The method for adaptive anti-shake control of a neurosurgery robot arm according to claim 1, characterized in that: The step of inputting the control output signal into an intracranial constraint flexible actuator for performing jitter compensation processing to implement adaptive anti-shake control of the end of the robotic arm in the intracranial space includes: Decomposing the control output signal into X, Y, and Z axis components and performing feasibility verification processing on the skull drilling point constraint kinematic model to obtain a constraint satisfaction flag; Performing current modulation activation processing on the shape memory alloy driver based on the constraint satisfaction flag to obtain a fine-tuning compensation displacement vector; The fine-tuning compensation displacement vector is subjected to damping coefficient adjustment processing by controlling the excitation current of the magnetorheological damper to obtain an adaptive damping parameter; Performing compensation trajectory generation processing through 7-DOF redundant manipulator zero-space motion planning according to the adaptive damping parameters to obtain intracranial space compensation motion instructions; The intracranial space compensation movement instruction is input into the robot arm joint controller for real-time execution processing to obtain the anti-shake compensation action of the robot arm end.

8. An adaptive anti-shake control system for a neurosurgery robot arm, characterized in that: The method for implementing the adaptive anti-shake control method for a neurosurgical robot arm according to any one of claims 1 to 7, wherein the adaptive anti-shake control system for the neurosurgical robot arm comprises: An acquisition module is used to collect and process signals of the surgical environment contact status through a multimodal sensor array to obtain tissue contact force, intracranial pressure change, and vascular pulsation frequency; a fusion module, configured to perform multi-source signal fusion processing based on the tissue contact force, intracranial pressure change, and vascular pulsation frequency using a neurosurgery-specific extended Kalman filter algorithm to obtain a jitter feature vector; A prediction module, configured to input the jitter feature vector into a neurosurgery scenario adaptive prediction model to perform jitter trend prediction processing and obtain compensation strategy parameters; a generating module, configured to generate a compensation control signal through a tissue-protective hypertwisted sliding mode controller according to the compensation strategy parameters to obtain a control output signal; The execution module is used to input the control output signal into the intracranial constraint flexible actuator for jitter compensation execution processing, thereby realizing adaptive anti-shake control of the end of the robotic arm in the intracranial space.

9. An adaptive anti-shake control device for a neurosurgery robot arm, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the computer program, the method for adaptive anti-shake control of a neurosurgical robot arm according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the processor is caused to execute the adaptive anti-shake control method for a manipulator arm of a neurosurgical robot according to any one of claims 1 to 7.

Images