Master hand device hand-release identification method for master-slave control medical robot and related device

Abstract

The invention belongs to the technical field of master-slave control medical robots, and discloses a master-hand equipment hand release identification method for a master-slave control medical robot and a related device. The master hand device hand-free identification method comprises the following steps: acquiring encoder data of a pitching Pitch axis and a yaw Yaw axis of a master hand device, performing state identification judgment by using a trained hand-free identification model, and outputting a master hand device hand-free identification result, the hand-release identification model comprises a feature extraction unit used for inputting encoder data, performing feature extraction on three levels of position, angular velocity and angular acceleration, and outputting a multi-dimensional motion feature vector; the discrimination network is used for inputting a multi-dimensional motion feature vector and outputting a probability result; and the judgment output unit is used for inputting a probability result and outputting a master hand equipment hand release identification result. According to the technical scheme disclosed by the invention, the method can better adapt to the recognition of the falling condition of the main hand equipment, and is higher in reliability.

Description

Technical Field

[0001] This invention belongs to the field of master-slave control medical robot technology, and specifically relates to a method and related device for master hand device detachment recognition in master-slave control medical robots. Background Technology

[0002] In master-slave controlled medical robots, the operator (typically a doctor) holds the master hand device. The robot controller maps the operator's three-dimensional hand movements to the slave hand robot according to certain motion control rules. The slave hand robot then performs surgical procedures based on the mapped motion information. The master hand device typically consists of a handle and a series of links. The handle facilitates the operator's gripping, while the links, while ensuring the handle's degrees of freedom, detect the handle's current position and orientation via multiple encoders mounted on the link connecting shafts and transmit kinematic information to the control system. Further, master-slave controlled medical robots have unique requirements for patient safety. Their design must consider a special scenario: due to operator error, the operator may unexpectedly release the handle (e.g., failing to release the enable pedal), causing the master hand device's handle to detach and fall. This could lead to unexpected movement of the slave hand robot's end effector and potentially cause a serious medical accident.

[0003] Given the above, it is necessary to effectively identify cases where the master device is detached and dropped, and to use the identification results to lock onto the slave robot to avoid medical accidents. However, existing drop detection methods heavily rely on acceleration signals, using them as the criterion. Some improved methods merely refine the acceleration signal processing flow; the fundamental technical approach remains unchanged, making them unsuitable for identifying cases where the master device is detached and dropped. Specifically, master devices generally do not contain inertial measurement units or accelerometers. Adding these sensors to the master device would increase system redundancy and complexity. Considering the specific application of master-slave controlled medical robots, existing detection schemes that heavily rely on acceleration signals are not well-suited for identifying cases where the master device is detached and dropped. Summary of the Invention

[0004] The purpose of this invention is to provide a method and related apparatus for identifying when a master hand device is accidentally released from the operator's grasp in a master-slave controlled medical robot, thereby solving one or more of the aforementioned technical problems. The technical solution disclosed in this invention, based on machine learning technology, relies solely on information from the roll, yaw, and pitch joint encoders in the master hand device to accurately determine whether the operator has unexpectedly released the master hand device. It is well-suited for identifying situations where the master hand device is accidentally released from the operator's grasp and has high reliability.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for identifying the removal of a master hand device from the hand of a master-slave controlled medical robot, comprising the following steps: Based on the selected master-slave controlled medical robot and the set sampling frequency, the encoder data of the pitch and yaw axes of the master hand device are acquired. Based on the acquired encoder data, the trained hands-off recognition model is used to perform state recognition and judgment, and output the hands-off recognition result of the master device. The hands-free recognition model includes: The feature extraction unit is used to input encoder data, perform feature extraction at three levels: position, angular velocity, and angular acceleration, and output a multi-dimensional motion feature vector. A discriminant network is used to take the multidimensional motion feature vector as input, perform inference and prediction, and output probability results; The determination output unit is used to input the probability result, compare and determine based on a preset determination threshold, and output the main hand device removal recognition result.

[0006] A further improvement to the technical solution of the present invention is that the training process of the off-hand recognition model includes: Based on the set sampling frequency, obtain the training sample dataset; Based on the acquired training sample dataset, the drop-hand recognition model is trained and its parameters are updated. Once the training is complete, the trained drop-hand recognition model is obtained. The training samples in the training sample dataset include: state labels representing static, normal movement, or drop, and encoder data samples of the Pitch and Yaw axes obtained in the corresponding states. The discriminant network in the off-hand recognition model is built based on a decision tree. During training and parameter updates, the decision tree outputs the probability of each state label occurring, and the label with the highest probability is used as the final result. Training ends when the model's prediction performance reaches a preset performance threshold.

[0007] A further improvement to the technical solution of the present invention is that the calculation expression for the model prediction performance is as follows: ; In the formula, To improve the model's prediction performance; False alarm rate; This represents the underreporting rate.

[0008] A further improvement to the technical solution of this invention lies in that, in the feature extraction unit, the process of extracting features from the input encoder data, performing feature extraction at three levels—position, angular velocity, and angular acceleration—and outputting a multidimensional motion feature vector includes: Based on the input encoder data, a sliding buffer is established for the position, velocity, and acceleration of the Pitch and Yaw axes, respectively. The peak-to-peak value and net displacement of the encoder position signals of the Pitch and Yaw axes within the sliding buffer are calculated, as are the mean, peak-to-peak value, and maximum value of the differential velocity signals of the Pitch and Yaw axes within the sliding buffer. The mean, variance, peak-to-peak value, proportion of zero acceleration, and the difference between the first and last data points of the sliding buffer are also calculated for the second-order differential acceleration signals of the Pitch and Yaw axes within the sliding buffer, resulting in a 20-dimensional motion feature vector.

[0009] A further improvement of the technical solution of the present invention is that the process by which the determination output unit performs the input of the probability result, compares and determines based on a preset determination threshold, and outputs the result of the main hand device being removed from the hand includes: Within each sampling frame, jitter cancellation processing is performed based on the input probability results to obtain the jitter-cancelled probability results; If, within multiple consecutive sampling frames, the predicted probability of a drop after jitter elimination exceeds the preset threshold, the main device will output a drop recognition result as a drop.

[0010] A further improvement to the technical solution of this invention is that, after the main hand device outputs the result of the hand-drop recognition as "dropped from hand," it also includes: The control decouples the slave robot from the master device, brakes the slave robot, and triggers an alarm.

[0011] A further improvement of the technical solution of the present invention is that the encoder data is specifically the encoder's radian return value.

[0012] A second aspect of the present invention provides a master-slave control device removal recognition system for a master-slave controlled medical robot, comprising: The data acquisition module is used to acquire encoder data of the pitch and yaw axes of the master hand device based on the selected master-slave controlled medical robot and the set sampling frequency. The recognition and prediction module is used to perform state recognition and judgment based on the acquired encoder data and the trained hands-off recognition model, and output the hands-off recognition result of the master device. The hands-free recognition model includes: The feature extraction unit is used to input encoder data, perform feature extraction at three levels: position, angular velocity, and angular acceleration, and output a multi-dimensional motion feature vector. A discriminant network is used to take the multidimensional motion feature vector as input, perform inference and prediction, and output probability results; The determination output unit is used to input the probability result, compare and determine based on a preset determination threshold, and output the main hand device removal recognition result.

[0013] In a third aspect, the present invention provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the master hand device off-hand recognition method for master-slave control of a medical robot as described in any one of the first aspects of the present invention.

[0014] In a fourth aspect, the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the master hand device detachment identification method for master-slave control of a medical robot as described in any one of the first aspects of the present invention.

[0015] Compared with the prior art, the present invention has the following beneficial effects: To address the technical challenge of existing drop detection methods heavily relying on acceleration signals, while master handpieces typically lack accelerometers, and adding one would increase system redundancy and complexity, this invention specifically discloses a drop detection method for master handpieces in master-slave controlled medical robots. In data acquisition, it does not rely on acceleration signals but instead acquires encoder data for the pitch and yaw axes of the master handpiece based on the selected master-slave controlled medical robot and a set sampling frequency. This avoids the problems caused by adding an accelerometer and does not increase system complexity. Subsequently, based on the acquired encoder data, a trained drop detection model is used for state recognition and judgment. The feature extraction unit of this drop detection model extracts features from the encoder data at three levels: position, angular velocity, and angular acceleration, outputting a multi-dimensional motion feature vector, enriching the feature information and providing a foundation for accurate recognition. The discriminant network performs inference and prediction on the multi-dimensional motion feature vector, outputting probability results, and improving recognition accuracy through nonlinear inference. The judgment output unit compares and judges the probability results based on a preset threshold, outputting the final recognition result. In summary, this invention discloses a novel solution based on machine learning technology. Relying solely on the joint encoder information of the master hand device, it can accurately determine whether an operator has accidentally released the master hand device. This solution is well-suited for identifying cases where the master hand device is dropped, demonstrating high reliability. Further illustratively, the reliable determination result can be used to lock the slave robot, ensuring medical safety during surgical procedures.

[0016] In the preferred embodiment of the present invention, the feature vector is specifically extended to 20 dimensions and combined into a nonlinear decision surface, which improves the precision of the judgment range and increases the accuracy. In particular, it can accurately distinguish between the rapid movement of the doctor's main hand and the abnormal release and drop scenario, and reduce the false judgment rate. Moreover, the dimension of the input vector can be flexibly adjusted without affecting the essential implementation of the method, further demonstrating the versatility of the present invention. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating a method for identifying the removal of a master hand device from the hand in a master-slave controlled medical robot, as described in an embodiment of the present invention. Figure 2 This is a schematic diagram of the model training process in an embodiment of the present invention; Figure 3 This is a schematic diagram of the process of using a trained model to identify and determine whether the main hand device is removed from the hand in a specific application embodiment of the present invention; Figure 4 This is a schematic diagram of a master hand device detachment recognition system for master-slave control of a medical robot, as described in an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention; obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0020] Based on the technical solutions disclosed in the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this invention. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. 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 explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or devices.

[0021] Please see Figure 1This invention provides a method for identifying the removal of the master handpiece from the hand in a master-slave controlled medical robot. Specifically, it is a master handpiece removal recognition algorithm based on encoders and machine learning applied in a master-slave controlled medical robot, comprising the following steps: Step 1: Based on the selected master-slave controlled medical robot and the set sampling frequency, acquire the encoder data of the pitch axis and yaw axis of the master hand device; Step 2: Based on the encoder data obtained in Step 1, use the trained hands-off recognition model to perform state recognition and judgment, and output the hands-off recognition result of the master device. The hands-free recognition model includes: The feature extraction unit is used to input encoder data, perform feature extraction at three levels: position, angular velocity, and angular acceleration, and output a multi-dimensional motion feature vector. A discriminant network is used to take the multidimensional motion feature vector as input, perform inference and prediction, and output probability results; The determination output unit is used to input the probability result, compare and determine based on a preset determination threshold, and output the main hand device removal recognition result.

[0022] The technical solutions disclosed in this invention are applicable to master hand devices connected by mechanical linkages. Typical master hand devices (such as ForceDimension's Omega series or SensAble's Touch series) belong to this category. In addition, this invention only declares a method for detecting master hand drop based on master hand encoder signal measurement, and does not involve improvements to the master hand device itself. Therefore, the structure of the master hand device and its underlying algorithms are not within the scope of improvement of this invention.

[0023] In the technical solution disclosed in this invention, the first part is to train the drop detection model, and the second part is to use the trained drop detection model to detect drops and obtain the detection result of the main hand drop state. Explained in principle, the technical solution of this invention inputs the encoder position signal, the differential velocity signal, and the second-order differential acceleration signal obtained by SG sliding window processing into the drop detection model. The drop detection model outputs the probability and maximum value of discrete signals 0 (stationary), 1 (normal movement), and 2 (main hand dropped).

[0024] In a specific exemplary technical solution of the present invention, the part of training the off-hand recognition model specifically includes: constructing an off-hand recognition model based on a decision tree; specifically, by establishing a decision tree with a maximum depth of 10 layers and a maximum leaf capacity of 64, training it using 80% of the labeled collected data as the training set, and validating it using 20% ​​of the labeled collected data as the test set; the labeled collected data is obtained by calculating and processing the encoder data of the two axes, Pitch (tilt, rotation around the horizontal axis) and Yaw (yaw, rotation around the vertical axis), and performing feature processing.

[0025] For example, the process of acquiring labeled data is as follows: A 64ms sliding buffer is established for the position, velocity, and acceleration of the Pitch and Yaw encoders using the acquired encoder position data; the peak-to-peak value and net displacement of the Pitch and Yaw encoder position signals within the sliding buffer are calculated, resulting in 4-dimensional data; the mean, peak-to-peak value, and maximum value of the Pitch and Yaw differential velocity signals within the sliding buffer are calculated, resulting in 6-dimensional data; the mean, variance, peak-to-peak value, zero acceleration percentage, and the difference between the beginning and end of the buffer data are calculated for the second-order differential acceleration signals obtained by SG sliding window processing of the Pitch and Yaw signals within the sliding buffer, resulting in 10-dimensional data; In summary, a total of 20 dimensions of data are used as input values ​​for the decision tree. Labeling involves designing operations during data acquisition to perform actions such as the main hand being stationary, normal pen holding with slow and fast movements in different directions, and releasing the pen at different positions, and assigning corresponding status labels of 0 (stationary), 1 (normal movement), and 2 (released from hand) to the corresponding data.

[0026] In the specific exemplary technical solution of this invention, the part of detecting dropped hands using the trained drop recognition model specifically includes: when calling the trained decision tree for judgment, the decision tree needs to be embedded in the application code, the encoder data of the main hand device acquired in real time is window-featured according to the trained 20-dimensional input vector and input into the decision tree, the decision tree will return the probability and maximum value of each category, after performing linear regression on the nonlinear output, 0.5 is taken as the threshold to determine whether the main hand has been dropped.

[0027] In summary, this invention discloses a novel drop detection scheme, which differs fundamentally from existing conventional zero-acceleration and acceleration threshold methods. It expands the feature vector to 20 dimensions (which can be optimized and adjusted according to specific circumstances) and combines multi-dimensional features into a nonlinear decision surface, improving the precision of the judgment range and increasing the accuracy of the judgment. It is important to emphasize that this invention distinguishes between scenarios involving rapid movement of the primary hand device operated by a doctor and scenarios involving drop detection, effectively reducing the false positive rate. Further illustratively, the 20-dimensional input vector is only used as an example; similar methods for determining drop detection or free fall, which only reduce or increase the dimension of the input vector, should be considered similar or even identical technical solutions to the method of this invention.

[0028] In a specific exemplary technical solution of this invention, LightGBM decision tree is used as the model framework, and the master encoder is sampled at a frequency of 1kHz. Interpretably, LightGBM decision tree is lightweight; taking a decision tree with a maximum depth of 10 layers and a maximum leaf capacity of 64 as an example, the time consumed for one inference cycle is 1~2 seconds. μs This satisfies the computational efficiency requirements for drop detection. Further, the core of the drop detection method proposed in this embodiment is to train a decision tree on existing labeled data to achieve nonlinear discrimination of the drop state of the main hand using multi-dimensional feature data. Therefore, the LightGBM decision tree is not within the scope of improvement of this embodiment, and the method of this invention does not depend on a specific machine learning model. Using other models to achieve similar functions to the method of this invention should also be regarded as similar or even the same technical solution.

[0029] Please see Figure 2 In this embodiment of the invention, the specific implementation steps for decision tree training are as follows: Step T1, raw data collection, includes: the host controller reads the radian return values ​​of the Yaw and Pitch two-axis encoders from the host device at a rate of 1kHz via USB or SPI acquisition card; it should be noted that since gravity compensation is performed in this embodiment of the invention, it is not necessary to read the value in the Z-axis direction, and whether or not data is sampled for other degrees of freedom is not the core of measuring whether it is similar to or even the same as the method of this invention.

[0030] Step T2, labeling data, includes: During the offline sampling phase, operators can label different events using a foot switch, graphical interface shortcut keys, or other triggering methods: for example, 0 represents stillness, 1 represents normal movement, and 2 represents dropping. In the preferred technical solution, the program records the labeling time and generates the same label for the data within a 200ms window before and after the labeling to avoid label misalignment caused by perceived lag.

[0031] Step T3, raw data feature extraction, includes: after sampling, firstly, the angle sequences of Pitch and Yaw are differentially calculated to obtain angular velocity; then, angular acceleration is calculated using a 9-point 3rd order SG sliding window to ensure the real-time performance of acceleration while denoising; subsequently, within the continuous data range of the same label (400ms), with a sliding window length of 64 frames (64ms), the mean, variance, peak-to-peak value, proportion of zero acceleration, and the difference between the beginning and end of the buffer data, the mean, peak-to-peak value, and maximum value of velocity, the peak-to-peak value of position, and the net displacement are calculated for each window. A total of 20-dimensional motion features of the two axes of Pitch and Yaw are stored in a two-dimensional matrix for training. This step transforms the information in the time domain into numerical information, enabling the decision tree to capture the dynamic pattern of the master hand without adding input dimensions to the time series.

[0032] Step T4, training set / test set partitioning, includes: randomly rearranging samples according to the window length as the unit length, and partitioning the training set and validation set according to an 8:2 ratio.

[0033] Step T5, LightGBM decision tree training, includes: training using the GBDT algorithm of the LightGBM framework; wherein, in this embodiment of the invention, a depth of 10 and 64 leaves are used exemplary (interpretatively, the actual depth can be adjusted between 6 to 12 layers and 32 to 128 leaves depending on the amount of data and embedded resources), column sampling and row sampling are enabled to suppress overfitting, loss information is output synchronously during the training process, and automatic truncation is set to ensure a balance between model capacity, running performance and generalization ability.

[0034] Step T6: Evaluate the training results, including: During the evaluation of the training results, first, calculate the false positive rate F and the false negative rate L. The decision tree output will show the probability of each label occurring, and the label with the highest probability will be used as the final result. Among all results, let the number of times the correct judgment of a drop be denoted as . Let the total number of times the error detection resulted in the item slipping and falling be . Let the total number of times the drop was not detected be . The false alarm rate can be calculated. and underreporting rate Obviously, when Theoretically optimal training time is used, but in practice... Time calculation Evaluate the level of training results; for example, require G>99% to ensure the reliability of the results.

[0035] Step T7: Export the model, which includes saving the model as a txt file, translating it into a pure C source file using the Treelite toolchain, and then embedding it into the application for use.

[0036] Please see Figure 3 In the specific example of the technical solution of this invention, the application steps of the trained off-hand recognition model are as follows: Step S1 is encoder angle sampling, including: ensuring a sampling frequency of 1kHz, ensuring that the unit of the returned radian value of the encoders on the Yaw and Pitch axes is consistent with that during training and ensuring that the data timing is correct, so as to ensure that the amplitude range of the subsequent first-order and second-order differentials is consistent with the training data.

[0037] Steps S2 and S3 involve calculating angular velocity and acceleration, using formulas consistent with those used in step T3 of the training process. The SG coefficients are loaded using the same parameter file as in the training process.

[0038] Step S4 is to update the buffer, which includes: pushing the position, calculated angular velocity and angular acceleration into three corresponding 64-frame buffers respectively. The buffers follow the first-in-first-out principle. When a new frame enters, the earliest frame data is automatically discarded. The window length is consistent with the window length in the training phase to ensure the consistency of the model feature distribution.

[0039] Step S5 is feature extraction, which includes: consistent with the calculation formula of the feature extraction part in step T3 of the training process, to obtain a total of 20-dimensional motion feature vectors of the Pitch and Yaw axes in three dimensions: position, angular velocity and angular acceleration.

[0040] Step S6 is LightGBM inference, which includes: the model file is completely identical to the model exported in step T7, and the model outputs a probability vector. .

[0041] Step S7 is jitter elimination, which includes: taking a weighted average of the probability result of the current drop judgment and the result of the previous drop judgment and caching it to the next window, so as to eliminate the impact of single judgment caused by data jitter to a certain extent.

[0042] Step S8 is the drop detection, including: the typical threshold value is 0.5; in practice, it can be adjusted between 0.4 and 0.6 according to recall / false alarm requirements, if after jitter elimination If the threshold (48ms) is exceeded for 48 consecutive sampling frames, it is determined that the main hand has been dropped. If the main hand has not been dropped, the process returns to step S1 to continue the next round of judgment.

[0043] Optional technical solutions may also include: Step S9 is to set the dropFlag flag and lock the robot's movement, including: immediately decoupling the robot from the master hand's movement and braking the robot upon confirmation of a fall; and setting the dropFlag flag and triggering a UI alarm message to notify the operator.

[0044] Step S10 is manual confirmation and unlocking, including: a flashing red light on the UI, the operator pressing the confirmation button on the UI, the dropFlag being cleared, and the robot's movement being re-coupled with the main hand. This step adds a manual reset step to prevent secondary damage.

[0045] Step S11 is buffering and state reset, including: clearing the circular buffer data and clearing the feature vector sequence to prevent old data residue from affecting the next round of discrimination. Then the process returns to step S1 and re-enters the 1kHz real-time loop.

[0046] In the above technical solutions provided by the embodiments of the present invention, the online discrimination process maintains complete consistency with the training environment in terms of data format, time window, signal processing algorithm and coefficients, and model file.

[0047] In summary, Chinese invention application CN101111770A discloses a drop detection method and device. It measures the triaxial acceleration of the device and sets a threshold. The device is judged to be in a drop state or not by detecting the duration of acceleration exceeding this threshold. Similar detection schemes are found in Chinese invention applications CN111812356A and CN102542724A. US invention application US20150020571A1 proposes a method for detecting falls using machine learning. It still uses accelerometer measurements to obtain raw data, but employs machine learning to determine the fall state. Sensor data is input into an SVM supervised learning unit to detect user falls. Its core innovation lies in the acceleration threshold criterion. Therefore, the aforementioned existing solutions all heavily rely on acceleration signals as the criterion; the main innovation lies in the processing of acceleration signals. In contrast, this invention aims to propose a method for determining the state of the main hand device being released from the hands using only the original encoder signals of the main hand device. Relying solely on the information from the roll, yaw, and pitch joint encoders in the main hand device, it can accurately determine whether the operator has released the main hand device unexpectedly. This method is well-suited for identifying situations where the main hand device is released from the hands and has high reliability.

[0048] The following are embodiments of the apparatus of the present invention, which can be used to execute embodiments of the method of the present invention. For details not disclosed in the apparatus embodiments, please refer to the embodiments of the method of the present invention.

[0049] Please see Figure 4 In this embodiment of the invention, a master hand device removal recognition system for master-slave controlled medical robots is provided, comprising: The data acquisition module is used to acquire encoder data of the pitch and yaw axes of the master hand device based on the selected master-slave controlled medical robot and the set sampling frequency. The recognition and prediction module is used to perform state recognition and judgment based on the acquired encoder data and the trained hands-off recognition model, and output the hands-off recognition result of the master device. The hands-free recognition model includes: The feature extraction unit is used to input encoder data, perform feature extraction at three levels: position, angular velocity, and angular acceleration, and output a multi-dimensional motion feature vector. A discriminant network is used to take the multidimensional motion feature vector as input, perform inference and prediction, and output probability results; The determination output unit is used to input the probability result, compare and determine based on a preset determination threshold, and output the main hand device removal recognition result.

[0050] In one embodiment of the present invention, a computer device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in the computer storage medium to achieve a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used to execute operations for a master-slave control method for identifying the release of a master hand device in a medical robot.

[0051] In one embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the operating system of the terminal. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor, which can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM (Random Access Memory) or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the master hand device release recognition method for master-slave controlled medical robots in the above embodiments.

[0052] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, optical storage, etc.) containing computer-usable program code.

[0053] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0054] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0055] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0056] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for identifying the removal of the master hand device in a master-slave controlled medical robot, characterized in that, Includes the following steps: Based on the selected master-slave controlled medical robot and the set sampling frequency, the encoder data of the pitch and yaw axes of the master hand device are acquired. Based on the acquired encoder data, the trained hands-off recognition model is used to perform state recognition and judgment, and output the hands-off recognition result of the master device. The hands-free recognition model includes: The feature extraction unit is used to input encoder data, perform feature extraction at three levels: position, angular velocity, and angular acceleration, and output a multi-dimensional motion feature vector. A discriminant network is used to take the multidimensional motion feature vector as input, perform inference and prediction, and output probability results; The determination output unit is used to input the probability result, compare and determine based on a preset determination threshold, and output the main hand device removal recognition result.

2. The method for identifying the removal of the master hand device in a master-slave controlled medical robot according to claim 1, characterized in that, The training process of the off-hand recognition model includes: Based on the set sampling frequency, obtain the training sample dataset; Based on the acquired training sample dataset, the drop-hand recognition model is trained and its parameters are updated. Once the training is complete, the trained drop-hand recognition model is obtained. The training samples in the training sample dataset include: state labels representing static, normal movement, or drop, and encoder data samples of the Pitch and Yaw axes obtained in the corresponding states. The discriminant network in the off-hand recognition model is built based on a decision tree. During training and parameter updates, the decision tree outputs the probability of each state label occurring, and the label with the highest probability is used as the final result. Training ends when the model's prediction performance reaches a preset performance threshold.

3. The method for identifying the removal of the master hand device in a master-slave controlled medical robot according to claim 2, characterized in that, The formula for calculating the model's predictive performance is as follows: ； In the formula, To improve the model's prediction performance; False alarm rate; This represents the underreporting rate.

4. The method for identifying the removal of the master hand device in a master-slave controlled medical robot according to claim 1, characterized in that, The feature extraction unit performs feature extraction on three levels—position, angular velocity, and angular acceleration—from the input encoder data, and outputs a multi-dimensional motion feature vector. This process includes: Based on the input encoder data, a sliding buffer is established for the position, velocity, and acceleration of the Pitch and Yaw axes, respectively. The peak-to-peak value and net displacement of the encoder position signals of the Pitch and Yaw axes within the sliding buffer are calculated, as are the mean, peak-to-peak value, and maximum value of the differential velocity signals of the Pitch and Yaw axes within the sliding buffer. The mean, variance, peak-to-peak value, proportion of zero acceleration, and the difference between the beginning and end data of the sliding buffer are calculated for the second-order differential acceleration signals of the Pitch and Yaw axes within the sliding buffer, respectively, to obtain a 20-dimensional motion feature vector.

5. A method for identifying the removal of a master hand device from a master-slave controlled medical robot according to claim 1, characterized in that, The process by which the determination output unit performs the input probability result, compares and determines based on a preset determination threshold, and outputs the result of the main hand device's removal from hand includes: Within each sampling frame, jitter cancellation processing is performed based on the input probability results to obtain the jitter-cancelled probability results; If, within multiple consecutive sampling frames, the predicted probability of a drop after jitter elimination exceeds a preset threshold, the main device will output a drop recognition result as a drop.

6. A method for identifying the removal of the master hand device in a master-slave controlled medical robot according to claim 1, characterized in that, The output of the main hand device's drop detection result after the device is released from the hand also includes: The control decouples the slave robot from the master device, brakes the slave robot, and triggers an alarm.

7. A method for identifying the removal of a master hand device from a master-slave controlled medical robot according to claim 1, characterized in that, The encoder data is specifically the encoder's radian return value.

8. A master hand device removal recognition system for master-slave controlled medical robots, characterized in that, include: The data acquisition module is used to acquire encoder data of the pitch and yaw axes of the master hand device based on the selected master-slave controlled medical robot and the set sampling frequency. The recognition and prediction module is used to perform state recognition and judgment based on the acquired encoder data and the trained hands-free recognition model, and output the hands-free recognition result of the master device. The hands-free recognition model includes: The feature extraction unit is used to input encoder data, perform feature extraction at three levels: position, angular velocity, and angular acceleration, and output a multi-dimensional motion feature vector. A discriminant network is used to take the multidimensional motion feature vector as input, perform inference and prediction, and output probability results; The determination output unit is used to input the probability result, compare and determine based on a preset determination threshold, and output the main hand device removal recognition result.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the master hand device detachment recognition method for master-slave control of a medical robot as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the master hand device detachment recognition method for master-slave control of a medical robot as described in any one of claims 1 to 7.

Images