Six-dimensional force compensation method and device for distal center motion mechanism

Abstract

The present invention provides a six-dimensional force compensation method for a remote center motion mechanism, the method comprising: obtaining the center of gravity of an RCM mechanism, the speed increment of the end of the RCM mechanism, and the position and posture of the end of the RCM mechanism relative to a base; performing inertia compensation for the torque and gravity caused by inertia of a target robot during movement according to the center of gravity, the speed increment, and the position and posture of the end of the RCM mechanism relative to the base, respectively, to obtain compensation torque and compensation force; performing reverse compensation for the compensation torque and compensation force according to the filter signals corresponding to the center of gravity, the torque caused by inertia, and gravity, respectively, to obtain target compensation torque and target compensation force. The method of the present invention can also eliminate the influence of the gravity of the robotic arm and the end effector when the external force on the robotic arm changes and the position and posture of the robotic arm are not fixed, thereby improving the accuracy and efficiency of gravity compensation for the six-dimensional force sensor.

Description

Technical Field

[0001] The present invention relates to the field of gravity compensation technology, and in particular to a six-dimensional force compensation method and device for a distal central motion mechanism. Background Art

[0002] Medical assistive devices (e.g., microsurgery robots) are equipped with force sensors at the end of their arms or at the base of their robots to measure external force information. In fact, affected by the robot's distal center of motion (RCM) mechanism, the gravity of the end load, and the posture of the RCM mechanism, when the robot is stationary or moving, even if the force sensor is not subject to external force, the sensor's reading will change, affecting the judgment of the external force.

[0003] In the related art, the gravity compensation method of the six-dimensional force sensor at the base is to use DH parameters and force coordinate transformation of different coordinate systems to obtain the force vector and torque vector represented by the end load and its own gravity in the force sensor coordinate system, and then use the Newton-Euler method to analyze the relationship between the position, velocity, acceleration and torque of each joint of the robot arm, so as to eliminate the influence of the gravity of the robot body and the end load on the six-dimensional force sensor during movement. However, due to the influence of force and torque brought by the zero-point drift of the force sensor and the inertia of the robot during movement, this method has poor effect on gravity compensation; in addition, the static gravity compensation method at the base is only applicable to static or quasi-static force measurement. When the measured force has dynamic changes, the static compensation effect becomes inaccurate or invalid; and the above methods all require relatively complex structural analysis, and assume that the center of gravity position of each rod is known, and the measurement is performed in a static state, and real-time measurement during movement cannot be guaranteed. Summary of the invention

[0004] The present invention provides a six-dimensional force compensation method and device for a remote center motion mechanism, which are used to solve the problems of zero-point drift of the gravity compensation force sensor in the prior art and the influence of force and torque caused by inertia during robot movement, and the need to know the center of gravity position of each rod, resulting in poor gravity compensation effect during robot movement and complex calculation, thereby improving the accuracy and efficiency of gravity compensation for the six-dimensional force sensor.

[0005] The present invention provides a six-dimensional force compensation method for a remote center motion mechanism, which is applied to a target robot. The target robot includes a remote center motion RCM mechanism and a base. A six-dimensional force sensor is installed on the RCM mechanism, including:

[0006] Acquire the center of gravity of the RCM mechanism, the velocity increment of the end of the RCM mechanism, and the position and posture of the end of the RCM mechanism relative to the base; wherein the center of gravity is determined based on the position and posture of the end of the RCM mechanism relative to the six-dimensional force sensor;

[0007] The inertia compensation method based on deep deterministic policy gradient performs inertia compensation on the torque and gravity caused by inertia during the movement of the target robot according to the center of gravity, the velocity increment, and the position and posture of the end of the RCM mechanism relative to the base, and obtains compensation torque and compensation force;

[0008] The compensation torque and the compensation force are respectively reversely compensated according to the center of gravity, the torque caused by inertia, and the filtered signal corresponding to gravity to obtain a target compensation torque and a target compensation force; wherein the target compensation torque and the target compensation force are used to ensure that the reading of the six-dimensional force sensor is zero when the target robot is not subjected to external force.

[0009] According to the six-dimensional force compensation method of the remote center motion mechanism provided by the present invention, the center of gravity of the RCM mechanism is obtained by the following steps:

[0010] The motion data and center of gravity position data of the RCM cantilever under different operating conditions are collected by sensors; the motion data include joint angles, speeds and accelerations, and the center of gravity position data include the force and torque output by the six-dimensional force sensor and the center of gravity position change data of the RCM cantilever under different postures;

[0011] Cleaning the motion data and the center of gravity position data to obtain cleaned data; standardizing and normalizing the cleaned data to obtain processed data;

[0012] Extracting key features from the processed data, and predicting the future state of the target robot based on the key features based on a long short-term memory (LSTM) network to obtain a future state quantity of the target robot;

[0013] The key features include the joint angle, velocity, acceleration and sensor output force and torque of the RCM mechanism. The LSTM network includes two LSTM layers and one fully connected layer. The first LSTM layer is set to return a sequence, and the second LSTM layer is set to not return a sequence.

[0014] The LSTM network predicts the future state quantity by the following formula:

[0015] ;

[0016] in, represents the current state of the system, which includes the joint angles, joint velocities, end effector positions, end effector postures, center of gravity positions, end effector velocities, end effector accelerations, and outputs of six-dimensional force sensors of the target robot in the current state. represents a control input, wherein the control input includes a joint velocity instruction, a desired position of the end effector, a desired posture of the end effector, a desired velocity of the end effector, and a desired acceleration of the end effector, represents the model parameters, Represents a predictive model, learned and optimized through machine learning algorithms;

[0017] Calculating the center of gravity of the RCM mechanism according to the future state quantity based on a PSO-BP neural network to obtain the center of gravity;

[0018] The PSO-BP neural network is determined based on a particle swarm optimization PSO algorithm and a back propagation BP neural network; the PSO-BP neural network is based on the positions of multiple rotational degrees of freedom, multiple pitch degrees of freedom, and multiple feed degrees of freedom of the RCM mechanism as inputs of the BP neural network, and the positions of multiple forces, corresponding moments, and multiple feed degrees of freedom of the center of gravity of the RCM mechanism detected by the six-dimensional force sensor as outputs of the BP neural network, and is trained using the following formula as the fitness function of the PSO algorithm:

[0019] ;

[0020] in, t j For the j The expected output of a neuron is p j For the j The actual output of a neuron is n is the number of output units of the PSO-BP neural network.

[0021] According to the six-dimensional force compensation method of the remote center motion mechanism provided by the present invention, the center of gravity of the RCM mechanism is calculated based on the PSO-BP neural network according to the future state quantity, and the center of gravity is obtained including:

[0022] Initializing a particle swarm according to the future state quantity, and setting the size of each particle in the particle swarm; wherein each particle is used to represent a possible solution of the BP neural network, and the possible solution includes a weight and a threshold;

[0023] Initialize the particle position by:

[0024] ;

[0025] in, is the initial position of particle 𝑖 in dimension 𝑗, is the reference value of the future state output by the prediction model on dimension 𝑗, is a random number between [0,1], and are the maximum and minimum values ​​of the 𝑗th dimension respectively;

[0026] Initialize the particle velocity by:

[0027] ;

[0028] in, is the initial position of particle 𝑖 in dimension 𝑗, is a random number between [0,1], and are the maximum and minimum values ​​of the 𝑗th dimension respectively;

[0029] Update the velocity and position of each particle using the following formula:

[0030] ;

[0031] in, is the velocity of particle i in dimension j, is the position of particle i in dimension j, is the inertia weight, and is the learning factor, and is a random number, is the historical best position of particle i, and g is the global best position;

[0032] The center of gravity is determined based on the fitness function according to the updated speed and position of each particle.

[0033] According to the six-dimensional force compensation method of the remote center motion mechanism provided by the present invention, the inertia compensation method based on the deep deterministic policy gradient performs inertia compensation on the torque and gravity caused by inertia during the movement of the target robot according to the center of gravity, the velocity increment, and the position and posture of the end of the RCM mechanism relative to the base, and obtains the compensation torque and compensation force, including:

[0034] Inputting the center of gravity, the velocity increment, the position and posture of the end relative to the base into a deep deterministic policy gradient network to obtain a compensation torque and a compensation force output by the deep deterministic policy gradient network;

[0035] Among them, the deep deterministic policy gradient network constructs the target state space of Markov decision based on the center of gravity of the sample RCM mechanism, the velocity increment of the end of the sample RCM mechanism, the position and posture of the end of the sample RCM mechanism relative to the base, and the torque and gravity caused by inertia, and constructs the action space of Markov decision based on the torque to be compensated and the gravity to be compensated, and updates the network weights corresponding to the target state space and the action space according to the sparse reward function.

[0036] According to the six-dimensional force compensation method of the distal center motion mechanism provided by the present invention, the sparse reward function includes:

[0037] ;

[0038] in, is the force threshold and is the torque threshold; f is gravity, T is the torque; m is a real number greater than 0.

[0039] According to the six-dimensional force compensation method of the distal center motion mechanism provided by the present invention, the filtering signal corresponding to the torque and gravity caused by inertia is determined by the following steps:

[0040] Obtaining the torque and gravity of the six-dimensional force sensor in a zero-point drift and fluctuation state;

[0041] Perform multiple Fourier linear filtering on the torque and the gravity to obtain the filtered signal.

[0042] According to the six-dimensional force compensation method of the distal center motion mechanism provided by the present invention, the performing of multiple Fourier linear filtering on the torque and the gravity to obtain the filtered signal comprises:

[0043] Compensating and suppressing the torque and the gravity based on a recursive least squares band-limited multiple Fourier linear combination RLS-BMFLC network to obtain the filtered signal;

[0044] The RLS-BMFLC network is obtained based on a recursive least squares algorithm and a band-limited multiple Fourier transform linear combination BMFLC network; the BMFLC network calculates the filtered signal by the following formula:

[0045] ;

[0046] in, Y k For the k The filtered signal at sampling time, An array of weight parameters expressed as a linear combination, It is expressed as a group of sine and cosine functions at each frequency point.

[0047] The present invention also provides a six-dimensional force compensation device for a distal central motion mechanism, comprising:

[0048] A data acquisition module, used to acquire the center of gravity of the remote center motion RCM mechanism of the target robot, the velocity increment of the end of the RCM mechanism, and the position and posture of the end of the RCM mechanism relative to the base of the target robot; wherein the center of gravity is determined based on the position and posture of the end of the RCM mechanism relative to the six-dimensional force sensor installed on the RCM mechanism;

[0049] A first compensation module is used for performing inertia compensation on the torque and gravity caused by inertia during the movement of the target robot according to the center of gravity, the velocity increment, and the position and posture of the end of the RCM mechanism relative to the base based on the inertia compensation method of the deep deterministic policy gradient, so as to obtain a compensation torque and a compensation force;

[0050] The second compensation module is used to reversely compensate the compensation torque and the compensation force according to the center of gravity, the torque caused by inertia, and the filtered signal corresponding to gravity to obtain a target compensation torque and a target compensation force; wherein the target compensation torque and the target compensation force are used to ensure that the reading of the six-dimensional force sensor is zero when the target robot is not subjected to external force.

[0051] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, a six-dimensional force compensation method for a distal center motion mechanism as described above is implemented.

[0052] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the six-dimensional force compensation method of the distal center motion mechanism as described in any one of the above is implemented.

[0053] The six-dimensional force compensation method and device of the remote center motion mechanism provided by the present invention respectively perform inertia compensation for the torque and gravity caused by inertia during the movement of the target robot through the inertia compensation method of deep deterministic policy gradient, and then reversely compensate the compensation torque and compensation force according to the filter signals corresponding to the center of gravity, the torque caused by inertia, and the gravity, so as to ensure that when the target robot is not subjected to external force, the reading of the six-dimensional force sensor is zero, and when the external force on the robot arm is a variable force and the posture of the robot arm is not fixed, the influence of the gravity of the robot arm itself and the end effector can be eliminated, thereby improving the accuracy and efficiency of gravity compensation for the six-dimensional force sensor. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. 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 paying creative work.

[0055] Figure 1 It is one of the flow charts of the six-dimensional force compensation method of the distal center motion mechanism provided by the present invention.

[0056] Figure 2 It is a schematic diagram of the deep deterministic policy gradient operation mechanism provided by the present invention.

[0057] Figure 3 It is a structural schematic diagram of the RCM cantilever mechanism provided by the present invention.

[0058] Figure 4 It is a flow chart of the PSO-BP algorithm provided by the present invention.

[0059] Figure 5 This is the second flow chart of the six-dimensional force compensation method of the distal center motion mechanism provided by the present invention.

[0060] Figure 6 It is a structural schematic diagram of the six-dimensional force compensation device of the distal center motion mechanism provided by the present invention.

[0061] Figure 7 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0063] Combine the following Figure 1-Figure 5 The six-dimensional force compensation method and device of the distal center motion mechanism of the present invention are described.

[0064] Figure 1 It is one of the flow charts of the six-dimensional force compensation method of the distal center motion mechanism provided by the present invention, such as Figure 1As shown, the six-dimensional force compensation method of the remote center motion mechanism is applied to a target robot, the target robot includes a remote center motion RCM mechanism and a base, and a six-dimensional force sensor is installed on the RCM mechanism, including the following steps:

[0065] Step 110, obtaining the center of gravity of the RCM mechanism, the velocity increment of the end of the RCM mechanism, and the position and posture of the end of the RCM mechanism relative to the base; wherein the center of gravity is determined based on the position and posture of the end of the RCM mechanism relative to the six-dimensional force sensor.

[0066] In this step, the target robot is a medical auxiliary device, for example, the target robot may be a microsurgery robot.

[0067] In this step, the six-dimensional force sensor is installed at the end cantilever of the remote center of motion (RCM) mechanism of the microsurgical robot (specifically located at the shoulder of the robot, that is, the rotating arm of the middle joint) to measure the external force and torque applied to the robot during operation in real time, thereby achieving precise force control and force feedback.

[0068] It should be noted that the six-dimensional force sensor is a sensor used to measure the force and torque of an object in six degrees of freedom. It is used to sense the force of an object in the X, Y, and Z axis directions and the torque around the X, Y, and Z axes.

[0069] In this embodiment, during the movement of the target robot, the center of gravity and the speed at the end of the RCM mechanism will change due to the influence of its own gravity and the end load of the RCM mechanism. The posture change of the RCM mechanism can be obtained according to the posture sensor, and the position of the center of gravity of the RCM mechanism can be calculated by determining the position of the end of the RCM mechanism relative to the six-dimensional force sensor according to the position sensor.

[0070] In this embodiment, the speed change of the RCM mechanism can be obtained by a speed sensor installed on the RCM mechanism.

[0071] Step 120, the inertia compensation method based on deep deterministic policy gradient performs inertia compensation for the torque and gravity caused by inertia during the movement of the target robot according to the center of gravity, velocity increment, position and posture of the end of the RCM mechanism relative to the base, and obtains compensation torque and compensation force.

[0072] In this step, the inertia compensation method of the deep deterministic policy gradient is to compensate for the torque and gravity caused by the inertia of the target robot during the movement process through the reinforcement learning model.

[0073] In this embodiment, the gravity and torque of the target robot during movement can be compensated by constructing a deep deterministic policy gradient (DDPG) network.

[0074] Specifically, the inertia compensation method based on deep deterministic policy gradient performs inertia compensation for the torque and gravity caused by inertia during the movement of the target robot according to the center of gravity, velocity increment, position and posture of the end of the RCM mechanism relative to the base, and obtains the compensation torque and compensation force, including: inputting the center of gravity, velocity increment, position and posture of the end relative to the base into the deep deterministic policy gradient network, and obtaining the compensation torque and compensation force output by the deep deterministic policy gradient network; wherein the deep deterministic policy gradient network constructs a target state space of Markov decision based on the center of gravity of the sample RCM mechanism, the velocity increment of the end of the sample RCM mechanism, the position and posture of the end of the sample RCM mechanism relative to the base, the torque and gravity caused by inertia, constructs an action space of Markov decision based on the torque to be compensated and the gravity to be compensated, and updates the network weights corresponding to the target state space and the action space according to the sparse reward function.

[0075] In this embodiment, the construction of the DDPG model is achieved through the following steps: During the network learning process, the state space needs to contain more information to find the optimal strategy, that is, the state space used for learning is expressed as:

[0076] ;

[0077] Among them, according to the compensation torque and compensation force to be determined, the action space of the DDPG model is set as follows:

[0078] ;

[0079] In this embodiment, the sparse reward function includes:

[0080] ;

[0081] in, is the force threshold and is the torque threshold; f is gravity, T is the torque; m is a real number greater than 0.

[0082] In this embodiment, the value of m can be set according to user needs. For example, m can be set to 5.

[0083] In this embodiment, the sparse reward function is set to evaluate and reward previous states and actions, thereby updating the weights of the Actor and Critic networks. When the number of iterations is met, a trained DDPG network is obtained for subsequent testing.

[0084] In this embodiment, during the DDPG model test, a new state space for Markov decision making is constructed using the above-mentioned center of gravity, velocity increment, and the position and posture of the end of the RCM mechanism relative to the base as shown below:

[0085] ;

[0086] In one embodiment, Input into the trained DDPG model, and use the long short-term memory network (LSTM) to establish the robot state prediction model. The next moment state output by the prediction model is used as the input of the DDPG model, and the output corresponds to gravity compensation and torque compensation, which is expressed as , realize feedforward control, reduce compensation delay and shorten control cycle.

[0087] In this example, with the help of reinforcement learning methods, the need for complex robot modeling can be avoided, and experience can be given to the robot more quickly, thereby improving the compensation effect.

[0088] Figure 2 is a schematic diagram of the deep deterministic policy gradient operation mechanism provided by the present invention. Figure 2 In the illustrated embodiment, the DDPG model includes an action Actor network and an evaluation Ctitic network. During the DDPG model training process, DDPG stores and utilizes historical data through experience replay techniques. Specifically, the data (state, action, reward, next state) generated by the surgical robot system is stored in an experience replay pool. During each training, a batch of data is randomly sampled from the experience replay pool to calculate the loss function and update the network parameters. In terms of parameter updating, DDPG adopts the Actor-Critic idea, in which the Actor network is responsible for generating actions, while the Critic network is responsible for evaluating the value of actions. By calculating the mean square error between the target value and the current value, the loss function is constructed and the gradient ( , ) is updated and the optimal strategy is finally determined.

[0089] Step 130, reversely compensate the compensation torque and compensation force according to the filter signals corresponding to the center of gravity, the torque caused by inertia, and gravity, to obtain the target compensation torque and target compensation force; wherein the target compensation torque and target compensation force are used to ensure that the reading of the six-dimensional force sensor is zero when the target robot is not subjected to external force.

[0090] In this step, during the actual movement of the target robot, in order to reduce the impact of the zero-point drift and fluctuation of the six-dimensional force sensor, the collected torque and gravity caused by inertia should be filtered and analyzed. For example, Fourier analysis is used to filter the torque and gravity caused by inertia to obtain the corresponding filtered signal.

[0091] In this embodiment, by compensating for the center of gravity deviation caused by the RCM structural gap, more reliable compensation torque and compensation force can be obtained.

[0092] In this embodiment, the numerical matrix corresponding to the above-mentioned filtered signal, the numerical matrix corresponding to the center of gravity position obtained by the above-mentioned calculation, and the numerical matrices corresponding to the compensation torque and the compensation force can be added to obtain the numerical matrix corresponding to the target compensation torque and the numerical matrix corresponding to the target compensation force, respectively; finally, the gravity compensation of the six-dimensional force sensor is realized according to the target compensation torque and the target compensation force to ensure that the reading of the six-dimensional force sensor on the RCM structure is always zero when the target robot is in motion or stationary process and is not subject to external force.

[0093] The six-dimensional force compensation method of the remote center motion mechanism provided by the present invention performs inertia compensation for the torque and gravity caused by inertia of the target robot during the movement by an inertia compensation method of a deep deterministic policy gradient, and then reversely compensates the compensation torque and compensation force according to the filter signals corresponding to the center of gravity, the torque caused by inertia, and the gravity, so as to ensure that when the target robot is not subjected to external force, the reading of the six-dimensional force sensor is zero, and when the external force applied to the robot arm is a variable force and the posture of the robot arm is not fixed, the influence of the gravity of the robot arm itself and the end effector can be eliminated, thereby improving the accuracy and efficiency of gravity compensation for the six-dimensional force sensor.

[0094] In some embodiments, the center of gravity of the RCM mechanism is obtained by the following steps: collecting motion data and center of gravity position data of the RCM cantilever under different operating conditions through sensors; the motion data includes joint angles, speeds and accelerations, and the center of gravity position data includes the force, torque and center of gravity position change data of the RCM cantilever under different postures output by the six-dimensional force sensor; cleaning the motion data and center of gravity position data to obtain cleaned data; standardizing and normalizing the cleaned data to obtain processed data; extracting key features from the processed data, and predicting the future state of the target robot based on the key features based on the long short-term memory network LSTM to obtain the future state quantity of the target robot; wherein the key features include the joint angles, speeds, accelerations and force and torque output by the sensors of the RCM mechanism; the LSTM network includes two LSTM layers and one fully connected layer; the first LSTM layer is set to return a sequence, and the second LSTM layer is set to not return a sequence; the LSTM network predicts the future state quantity by the following formula:

[0095] ;

[0096] in, Represents the current state of the system, which includes the joint angle, joint velocity, end effector position, end effector posture, center of gravity position, end effector velocity, end effector acceleration and the output of the six-dimensional force sensor of the target robot in the current state. represents the control input, which includes the joint velocity command, the desired position of the end effector, the desired posture of the end effector, the desired velocity of the end effector, and the desired acceleration of the end effector. represents the model parameters, Represents a prediction model, which is learned and optimized through a machine learning algorithm; the center of gravity of the RCM mechanism is calculated based on the future state quantity based on the PSO-BP neural network to obtain the center of gravity; wherein, the PSO-BP neural network is determined based on the particle swarm optimization PSO algorithm and the back propagation BP neural network; the PSO-BP neural network is based on the positions of multiple rotational degrees of freedom, multiple pitch degrees of freedom, and multiple feed degrees of freedom of the RCM mechanism as the input of the BP neural network, and the multiple forces detected by the six-dimensional force sensor, the corresponding torques, and the positions of the center of gravity of the RCM mechanism and multiple feed degrees of freedom are outputs of the BP neural network, and the fitness function of the PSO algorithm is trained using the following formula:

[0097] ;

[0098] in, t j For the j The expected output of a neuron is pj For the j The actual output of a neuron is n is the number of output units of the PSO-BP neural network.

[0099] In this embodiment, after data collection is completed, missing values ​​and outliers in the data are cleaned to ensure the integrity and accuracy of the data; the data is standardized and normalized to unify the data scales of different features to facilitate subsequent model training.

[0100] In this embodiment, in order to improve the key feature representation capability, the key features can be reduced in dimension by using the principal component analysis (PCA) method to generate feature vectors, reduce redundant information, and improve the model training efficiency and prediction accuracy.

[0101] In this embodiment, before model training, the collected sample data is divided into a training set, a validation set, and a test set; the division ratio of each sample set can be set according to user needs, for example, the division ratio can be 70% for training, 15% for validation, and 15% for testing; through the above sample set division method, the performance of the model on different data sets can be ensured to prevent the model from overfitting; in the training stage, after the data is divided into a training set, a validation set, and a test set, the training set data is used to train the LSTM model, the number of training rounds (epochs) and the amount of data processed in each batch (batch size) are set, and the model performance is monitored on the validation set, and the learning process of the model is observed by drawing the training and validation loss curves to ensure that the model is not overfitting; after the training is completed, the validation set data is used to evaluate the model performance, and indicators such as the mean square error (MSE) and the mean absolute error (MAE) are calculated. According to the evaluation results, the model structure and parameters are adjusted to improve the prediction accuracy of the model; for example, the number of units in the LSTM layer can be increased, the learning rate can be modified, and the number of training rounds can be increased. Through multiple adjustments, an LSTM model that performs well on the validation set is finally obtained.

[0102] In this embodiment, the trained LSTM model is used to predict the future state; the new input data is passed into the model to obtain the future system state, and these predicted future states are used to initialize the particle swarm, thereby optimizing the initial weights and thresholds of the PSO-BP neural network.

[0103] Figure 3 is a schematic diagram of the structure of the RCM cantilever mechanism provided by the present invention. Figure 3 In the illustrated embodiment, the six-axis force sensor is mounted on an RCM cantilever mechanism, wherein the cantilever mechanism includes a plurality of degrees of freedom, and the degrees of freedom corresponding to the direction away from the six-axis force sensor are deflection degree of freedom, pitch degree of freedom and feed degree of freedom.

[0104] In this embodiment, the center of gravity of the RCM mechanism is calculated based on the future state quantity based on the PSO-BP neural network, and the center of gravity is obtained, including: initializing the particle swarm according to the future state quantity, and setting the size of each particle in the particle swarm; wherein each particle is used to represent a possible solution of the BP neural network, and the possible solution includes a weight and a threshold; initializing the particle position by the following formula:

[0105] ;

[0106] in, is the initial position of particle 𝑖 in dimension 𝑗, is the reference value of the future state output by the prediction model on dimension 𝑗, is a random number between [0,1], and are the maximum and minimum values ​​of the 𝑗th dimension respectively; the particle velocity is initialized by the following formula:

[0107] ;

[0108] in, is the initial position of particle 𝑖 in dimension 𝑗, is a random number between [0,1], and are the maximum and minimum values ​​of the 𝑗th dimension respectively; the speed and position of each particle are updated by the following formula:

[0109] ;

[0110] in, is the velocity of particle i in dimension j, is the position of particle i in dimension j, is the inertia weight, and is the learning factor, and is a random number, is the historical best position of particle i, and g is the global best position. The center of gravity is determined based on the fitness function according to the updated speed and position of each particle.

[0111] In this embodiment, by introducing the prediction model and feedforward control strategy, the response speed and compensation accuracy of the system are further enhanced. In actual operation, real-time sensor data is input into the prediction model to obtain future interference prediction results, and the required compensation amount is calculated through the feedforward control model to adjust the control parameters of the RCM cantilever and the six-dimensional force sensor in advance. The feedforward control strategy is integrated with the feedback control system to form a feedforward-feedback closed-loop control system, using feedforward compensation to reduce response delays and feedback control to correct errors.

[0112] In this embodiment, the particle swarm is initialized and the size of the particle swarm is set by predicting the future state of the model output, that is, each particle represents a possible solution of the BP neural network, including a weight and a threshold.

[0113] In this embodiment, the future state output by the LSTM prediction model can be used to initialize the position and velocity of the particle swarm, thereby optimizing the initial weight and threshold of the PSO-BP neural network, improving the model training speed and prediction accuracy, achieving accurate calculation of the center of gravity position of the RCM mechanism, and improving the accuracy and efficiency of the gravity compensation of the six-dimensional force sensor.

[0114] In this embodiment, in order to solve the problem that the center of gravity coordinates of the manipulator body and the end effector of the moving target robot are difficult to establish accurately, this embodiment uses PSO-BP to adjust the center of gravity offset caused by the RCM posture change, thereby improving the model training speed and prediction accuracy.

[0115] Specifically, the use of PSO-BP online calibration targets can eliminate the impact of the change in the center of gravity of the robot arm on the six-dimensional force sensor when the robot arm RCM mechanism moves; in this process, the initial weights and thresholds of the back propagation (BP) neural network are first optimized by Particle Swarm Optimization (PSO), where the fitness function selects the error between the predicted value and the actual value of the BP neural network; then the fitness function and the particle swarm algorithm are iteratively updated, and the final optimal parameters are used to initialize the BP neural network weights and thresholds; finally, the optimally initialized BP neural network is obtained, and the network is trained to obtain the PSO-BP neural network, which will greatly improve the convergence speed and prediction accuracy. Figure 4 It is a flow chart of the PSO-BP algorithm provided by the present invention. Figure 4 In the embodiment shown, during the PSO algorithm operation phase, the particle velocity, position, inertia factor, iteration number and other parameters are initialized by using the initialized BP neural network structure and parameters, and then the velocity and position of particle i are updated by combining equations (1) and (2), and the fitness value f(i), individual optimal value f(pBest) and global optimal value f(gBest) of each particle are calculated by equation (3) corresponding to the fitness function mentioned above; the results are compared and pBest and gBest are updated, and the global optimal particle position is output when the iteration conditions are met; during the BP neural network training phase, the BP neural network weights and thresholds are initialized using the global optimal particle position, and the BP neural network training is performed. When the termination conditions are met, the PSO-BP neural network is obtained, and subsequent tests are performed.

[0116] In this embodiment, during the training process of the BP neural network, ReLU is used as the activation function. During the training, the three degrees of freedom positions of the RCM mechanism should all be involved in the model training. A large amount of sample data is collected in advance. No external force is applied but only the positions of the degrees of freedom are changed. The collected data is grouped as the degrees of freedom position-actual measured values ​​of the sensor. The deviation between the network output and the actual measured value of the sensor is used to update the parameters between the neurons of the BP neural network using stochastic gradient descent.

[0117] Specifically, the particle swarm algorithm in this embodiment is as follows:

[0118] (1) Initialize the particle swarm dimensions and parameters;

[0119] (2) Calculate the current fitness value of each particle using the fitness function;

[0120] (3) Find the optimal individual particle pBest and the global optimal gBest;

[0121] (4) Update the speed and position of each particle. The update formula is as follows:

[0122] ;

[0123] in, is the inertia weight, which is generally between [0.5, 1.5], , is called the acceleration constant, and is generally taken as [1,4], , is a random number distributed in the interval [0,1]. is the current iteration number, , are the current particle speed and position, is the individual optimal particle position, is the global optimal particle position;

[0124] (5) Repeat steps (2) to (4) until the termination condition is reached.

[0125] The six-dimensional force compensation method of the remote center motion mechanism provided by the present invention obtains the center of gravity position of the target robot during the motion process through the particle swarm optimization BP neural network dynamic compensation method, which provides reliable data support for the subsequent calculation of the target compensation torque and target compensation force.

[0126] In some embodiments, the filtered signals corresponding to the torque and gravity caused by inertia are determined by the following steps: obtaining the torque and gravity of the six-dimensional force sensor under zero drift and fluctuation states; performing multiple Fourier linear filtering on the torque and gravity to obtain filtered signals.

[0127] In this embodiment, a multiple Fourier linear combiner can be constructed through preliminary experiments to perform multiple Fourier linear filtering on the torque and gravity, thereby improving data processing efficiency.

[0128] Specifically, performing multiple Fourier linear filtering on the torque and gravity to obtain the filtered signal includes: compensating and suppressing the torque and gravity based on the recursive least squares band-limited multiple Fourier linear combination RLS-BMFLC network to obtain the filtered signal; wherein the RLS-BMFLC network is obtained based on the recursive least squares algorithm and the band-limited multiple Fourier linear combination BMFLC network; the BMFLC network calculates the filtered signal by the following formula:

[0129] ;

[0130] in, Y k For the k The filtered signal at sampling time, An array of weight parameters expressed as a linear combination, It is expressed as a group of sine and cosine functions at each frequency point.

[0131] In this embodiment, when applying a multiple band-limited Fourier linear combiner in a six-dimensional force sensor of a surgical robot with an RCM mechanism, the frequency range of the Fourier linear combiner is first limited. , and the "band-limited" bandwidth of the combiner, the form of the BMFLC combining analog signals using sine and cosine trigonometric functions is as follows:

[0132] ;

[0133] in, is the estimated signal at sampling time k, and is the frequency point at time k The combined weight parameter at , the corresponding filtered signal can be expressed as:

[0134] .

[0135] In this embodiment, a recursive least squares band-limited multiple Fourier linear combiner (RLS-BMFLC) is used to compensate and suppress the zero drift and fluctuation of the six-bit force sensor.

[0136] Specifically, the recursive least squares algorithm updates the weights of the BMFLC network by the following formula:

[0137] ;

[0138] in, For the k The recursive least squares gain at sampling time, is the recursive least squares factor, The value range is , For the k The recursive least squares error at sampling time, k For the k The recursive least squares error covariance at sampling time, For the k The weight parameter at each sampling moment.

[0139] The six-dimensional force compensation method of the remote center motion mechanism provided by the present invention performs Fourier linear filtering on torque and gravity by recursive least squares band-limited multiple Fourier linear combination to obtain a filtered signal, which can reduce the influence of zero drift and fluctuation of the six-dimensional force sensor and improve the measurement accuracy of the six-dimensional force sensor.

[0140] Figure 5 This is the second flow chart of the six-dimensional force compensation method of the distal center motion mechanism provided by the present invention. Figure 5 In the embodiment shown, the current state quantity of the surgical robot system is , control input and model parameters , input into the robot state prediction model based on LSTM, and output the next moment ( t +1) Status As the input of the PSO-BP network, the relative force sensor posture of the RCM end , RCM end relative to the force sensor position , the position of the cantilever end relative to the robot base output by the surgical robot system P , the posture of the cantilever end relative to the robot base R and terminal velocity increment , respectively input into the PSO-BP network and DDPG network for inertia compensation, and obtain the compensation torque and compensation force; the force sensor feedback force output by the surgical robot system is fed back by RLS-BMFLC f and feedback torque T Filter to obtain the filtered signal, and use the filtered signal and the center of gravity O Perform reverse compensation on the compensation torque and compensation force to obtain the corresponding compensation force and compensation torque ; Finally, the and The six-dimensional force sensor acting on the surgical robot system can ensure that the reading of the six-dimensional force sensor is always zero when the surgical robot system is not subjected to external force.

[0141] The six-dimensional force compensation device of the distal center motion mechanism provided by the present invention is described below. The six-dimensional force compensation device of the distal center motion mechanism described below and the six-dimensional force compensation method of the distal center motion mechanism described above can be referenced to each other.

[0142] Figure 6 Schematic diagram of the structure of the six-dimensional force compensation device of the distal center motion mechanism provided by the present invention, such as Figure 6 As shown, the six-dimensional force compensation device of the distal central motion mechanism includes: a data acquisition module 610, a first compensation module 620 and a second compensation module 630.

[0143] The data acquisition module 610 is used to acquire the center of gravity of the remote center motion RCM mechanism of the target robot, the velocity increment of the end of the RCM mechanism, and the position and posture of the end of the RCM mechanism relative to the base of the target robot; wherein the center of gravity is determined based on the position and posture of the end of the RCM mechanism relative to the six-dimensional force sensor installed on the RCM mechanism;

[0144] A first compensation module 620 is used to perform inertia compensation on the target robot based on the inertia compensation method of the deep deterministic policy gradient according to the center of gravity, the velocity increment, the position and posture of the end of the RCM mechanism relative to the base, and the torque and gravity caused by the inertia during the movement of the target robot to obtain the compensation torque and compensation force;

[0145] The second compensation module 630 is used to perform reverse compensation on the compensation torque and compensation force according to the filter signals corresponding to the center of gravity, the torque caused by inertia, and gravity, to obtain the target compensation torque and target compensation force; wherein, the target compensation torque and target compensation force are used to ensure that the reading of the six-dimensional force sensor is zero when the target robot is not subjected to external force.

[0146] The six-dimensional force compensation device of the remote center motion mechanism provided by the present invention performs inertia compensation for the torque and gravity caused by inertia of the target robot during the movement by an inertia compensation method of a deep deterministic policy gradient, and then reversely compensates the compensation torque and compensation force according to the filter signals corresponding to the center of gravity, the torque caused by inertia, and the gravity, so as to ensure that when the target robot is not subjected to external force, the reading of the six-dimensional force sensor is zero, and when the external force applied to the robot arm is a variable force and the position of the robot arm is not fixed, the influence of the gravity of the robot arm itself and the end effector can be eliminated, thereby improving the accuracy and efficiency of gravity compensation for the six-dimensional force sensor.

[0147] Figure 7 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 7As shown, the electronic device may include: a processor (processor) 710 , a communication interface (Communications Interface) 720 , a memory (memory) 730 and a communication bus 740 , wherein the processor 710 , the communication interface 720 , and the memory 730 communicate with each other through the communication bus 740 . The processor 710 can call the logic instructions in the memory 730 to execute the six-dimensional force compensation method of the remote center motion mechanism, which is applied to the target robot. The target robot includes a remote center motion RCM mechanism and a base, and a six-dimensional force sensor is installed on the RCM mechanism. The method includes: obtaining the center of gravity of the RCM mechanism, the velocity increment of the end of the RCM mechanism, and the position and posture of the end of the RCM mechanism relative to the base; wherein the center of gravity is determined based on the position and posture of the end of the RCM mechanism relative to the six-dimensional force sensor; the inertia compensation method based on the deep deterministic policy gradient performs inertia compensation for the torque and gravity caused by inertia during the movement of the target robot according to the center of gravity, the velocity increment, and the position and posture of the end of the RCM mechanism relative to the base, and obtains the compensation torque and compensation force; the compensation torque and compensation force are reversely compensated according to the filter signals corresponding to the center of gravity, the torque caused by inertia, and the gravity, and the target compensation torque and target compensation force are obtained; wherein the target compensation torque and target compensation force are used to ensure that the reading of the six-dimensional force sensor is zero when the target robot is not subjected to external force.

[0148] In addition, the logic instructions in the above-mentioned memory 730 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform 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, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0149] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the six-dimensional force compensation method of the remote center motion mechanism provided by the above methods, which is applied to a target robot. The target robot includes a remote center motion RCM mechanism and a base, and a six-dimensional force sensor is installed on the RCM mechanism. The method includes: obtaining the center of gravity of the RCM mechanism, the velocity increment of the end of the RCM mechanism, and the position and posture of the end of the RCM mechanism relative to the base; wherein the center of gravity is based on the RCM The position and posture of the end of the mechanism relative to the six-dimensional force sensor are determined; the inertia compensation method based on deep deterministic policy gradient performs inertia compensation for the torque and gravity caused by inertia during the movement of the target robot according to the center of gravity, velocity increment, and the position and posture of the end of the RCM mechanism relative to the base, and obtains the compensation torque and compensation force; the compensation torque and compensation force are reversely compensated according to the filter signals corresponding to the center of gravity, the torque caused by inertia, and the gravity, and the target compensation torque and target compensation force are obtained; wherein, the target compensation torque and target compensation force are used to ensure that the reading of the six-dimensional force sensor is zero when the target robot is not subjected to external force.

[0150] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the six-dimensional force compensation method of the remote center motion mechanism provided by the above methods, and is applied to a target robot, wherein the target robot includes a remote center motion RCM mechanism and a base, and a six-dimensional force sensor is installed on the RCM mechanism, and the method includes: obtaining the center of gravity of the RCM mechanism, the velocity increment of the end of the RCM mechanism, and the position and posture of the end of the RCM mechanism relative to the base; wherein the center of gravity is determined based on the position and posture of the end of the RCM mechanism relative to the six-dimensional force sensor; an inertia compensation method based on a deep deterministic policy gradient performs inertia compensation for the torque and gravity caused by inertia of the target robot during movement according to the center of gravity, the velocity increment, and the position and posture of the end of the RCM mechanism relative to the base, and obtains a compensation torque and a compensation force; the compensation torque and the compensation force are reversely compensated according to the filter signals corresponding to the center of gravity, the torque caused by inertia, and the gravity, and obtain a target compensation torque and a target compensation force; wherein the target compensation torque and the target compensation force are used to ensure that when the target robot is not subjected to external force, the reading of the six-dimensional force sensor is zero.

[0151] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0152] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0153] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. 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 embodiments of the present invention.

Claims

1. A six-dimensional force compensation method for a remote center motion mechanism, applied to a target robot, wherein the target robot comprises a remote center motion RCM mechanism and a base, wherein a six-dimensional force sensor is installed on the RCM mechanism, and wherein: include: Acquire the center of gravity of the RCM mechanism, the velocity increment of the end of the RCM mechanism, and the position and posture of the end of the RCM mechanism relative to the base; wherein the center of gravity is determined based on the position and posture of the end of the RCM mechanism relative to the six-dimensional force sensor; The inertia compensation method based on deep deterministic policy gradient performs inertia compensation on the torque and gravity caused by inertia during the movement of the target robot according to the center of gravity, the velocity increment, and the position and posture of the end of the RCM mechanism relative to the base, and obtains compensation torque and compensation force; The compensation torque and the compensation force are respectively reversely compensated according to the center of gravity, the torque caused by inertia, and the filtered signal corresponding to gravity to obtain a target compensation torque and a target compensation force; wherein the target compensation torque and the target compensation force are used to ensure that the reading of the six-dimensional force sensor is zero when the target robot is not subjected to external force.

2. The six-dimensional force compensation method of the distal center motion mechanism according to claim 1, characterized in that: The center of gravity of the RCM mechanism is obtained by the following steps: The motion data and center of gravity position data of the RCM cantilever under different operating conditions are collected by sensors; the motion data include joint angles, speeds and accelerations, and the center of gravity position data include the force and torque output by the six-dimensional force sensor and the center of gravity position change data of the RCM cantilever under different postures; Cleaning the motion data and the center of gravity position data to obtain cleaned data; standardizing and normalizing the cleaned data to obtain processed data; Extracting key features from the processed data, and predicting the future state of the target robot based on the key features based on a long short-term memory (LSTM) network to obtain a future state quantity of the target robot; The key features include the joint angle, velocity, acceleration and force and torque output by the sensor of the RCM mechanism; the LSTM network includes two LSTM layers and one fully connected layer; the first LSTM layer is set to return a sequence, and the second LSTM layer is set to not return a sequence; The LSTM network predicts the future state quantity by the following formula: x t+1 =f(x t ,u t ;θ); Among them, x t represents the current state of the system, which includes the joint angle, joint velocity, end effector position, end effector posture, center of gravity position, end effector velocity, end effector acceleration and output of the six-dimensional force sensor of the target robot in the current state, u t represents the control input, which includes the joint speed command, the desired position of the end effector, the desired posture of the end effector, the desired speed of the end effector, and the desired acceleration of the end effector, θ represents the model parameters, and f represents the prediction model, which is learned and optimized by the machine learning algorithm; Calculating the center of gravity of the RCM mechanism according to the future state quantity based on a PSO-BP neural network to obtain the center of gravity; The PSO-BP neural network is determined based on a particle swarm optimization PSO algorithm and a back propagation BP neural network; the PSO-BP neural network is based on the positions of multiple rotational degrees of freedom, multiple pitch degrees of freedom, and multiple feed degrees of freedom of the RCM mechanism as inputs of the BP neural network, and the positions of multiple forces, corresponding moments, and multiple feed degrees of freedom of the center of gravity of the RCM mechanism detected by the six-dimensional force sensor as outputs of the BP neural network, and is trained using the following formula as the fitness function of the PSO algorithm: Among them, t j is the expected output of the jth neuron, p j is the actual output of the jth neuron, and n is the number of output units of the PSO-BP neural network.

3. The six-dimensional force compensation method of the distal center motion mechanism according to claim 2, characterized in that: The PSO-BP neural network is used to calculate the center of gravity of the RCM mechanism according to the future state quantity, and the center of gravity is obtained including: Initializing a particle swarm according to the future state quantity, and setting the size of each particle in the particle swarm; wherein each particle is used to represent a possible solution of the BP neural network, and the possible solution includes a weight and a threshold; Initialize the particle position by: x i,j (0)=x ref,j +α(x max,j -x min,j ); Among them, x i,j (0) is the initial position of particle i in dimension j, x ref,j is the reference value of the future state output by the prediction model on dimension j, α is a random number between [0,1], x max,j and x min,j are the maximum and minimum values ​​of the j-th dimension respectively; Initialize the particle velocity by: v i,j (0)=β(v max,j -v min,j ); Among them, v i,j (0) is the initial position of particle i in dimension j, β is a random number between [0,1], v max,j and v min,j are the maximum and minimum values ​​of the j-th dimension respectively; Update the velocity and position of each particle using the following formula: Among them, v i,j is the velocity of particle i in dimension j, x i,j is the position of particle i in dimension j, ω is the inertia weight, c1 and c2 are learning factors, r1 and r2 are random numbers, p i is the historical best position of particle i, and g is the global best position; The center of gravity is determined based on the fitness function according to the updated speed and position of each particle.

4. The six-dimensional force compensation method of the distal center motion mechanism according to claim 1, characterized in that: The inertia compensation method based on deep deterministic policy gradient performs inertia compensation on the torque and gravity caused by inertia during the movement of the target robot according to the center of gravity, the velocity increment, and the position and posture of the end of the RCM mechanism relative to the base, and obtains the compensation torque and compensation force, including: Inputting the center of gravity, the velocity increment, the position and posture of the end relative to the base into a deep deterministic policy gradient network to obtain a compensation torque and a compensation force output by the deep deterministic policy gradient network; Among them, the deep deterministic policy gradient network constructs the target state space of Markov decision based on the center of gravity of the sample RCM mechanism, the velocity increment of the end of the sample RCM mechanism, the position and posture of the end of the sample RCM mechanism relative to the base, the torque and gravity caused by inertia, and constructs the action space of Markov decision based on the torque to be compensated and the gravity to be compensated, and updates the network weights corresponding to the target state space and the action space according to the sparse reward function.

5. The six-dimensional force compensation method of the distal center motion mechanism according to claim 4, characterized in that: The sparse reward function includes: Among them, f is the force threshold and T is the torque threshold; f is gravity, T is torque; m is a real number greater than 0.

6. The six-dimensional force compensation method of the distal central motion mechanism according to claim 1 or 4, characterized in that: The filtered signal corresponding to the torque and gravity caused by inertia is determined by the following steps: Obtaining the torque and gravity of the six-dimensional force sensor in a zero-point drift and fluctuation state; Perform multiple Fourier linear filtering on the torque and the gravity to obtain the filtered signal.

7. The six-dimensional force compensation method of the distal central motion mechanism according to claim 6, characterized in that: The performing of multiple Fourier linear filtering on the torque and the gravity to obtain the filtered signal comprises: Compensating and suppressing the torque and the gravity based on the RLS-BMFLC network to obtain the filtered signal; The RLS-BMFLC network is obtained based on a recursive least squares algorithm and a band-limited multiple Fourier transform linear combination BMFLC network; the BMFLC network calculates the filtered signal by the following formula: Among them, Y k is the filtered signal at the kth sampling moment, An array of weight parameters expressed as linear combinations, X k It is expressed as a group of sine and cosine functions at each frequency point.

8. A six-dimensional force compensation device for a distal central motion mechanism, characterized in that: include: A data acquisition module, used to acquire the center of gravity of the remote center motion RCM mechanism of the target robot, the velocity increment of the end of the RCM mechanism, and the position and posture of the end of the RCM mechanism relative to the base of the target robot; wherein the center of gravity is determined based on the position and posture of the end of the RCM mechanism relative to the six-dimensional force sensor installed on the RCM mechanism; A first compensation module is used for performing inertia compensation on the torque and gravity caused by inertia during the movement of the target robot according to the center of gravity, the velocity increment, and the position and posture of the end of the RCM mechanism relative to the base based on the inertia compensation method of the deep deterministic policy gradient, so as to obtain a compensation torque and a compensation force; The second compensation module is used to reversely compensate the compensation torque and the compensation force according to the center of gravity, the torque caused by inertia, and the filtered signal corresponding to gravity to obtain a target compensation torque and a target compensation force; wherein the target compensation torque and the target compensation force are used to ensure that the reading of the six-dimensional force sensor is zero when the target robot is not subjected to external force.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the six-dimensional force compensation method of the distal central motion mechanism as described in any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the six-dimensional force compensation method of the distal central motion mechanism as described in any one of claims 1 to 7 is implemented.

Images