Mechanical arm human-computer interaction method and system

Abstract

The application provides a mechanical arm human-computer interaction method and system, the method comprising: acquiring force data of an external force applied to the end of the mechanical arm collected by a force sensor; preprocessing the force data, the preprocessing comprising: performing time domain feature extraction, and performing linear fitting on the time domain features; inputting the preprocessed force data into a pre-trained random forest model, and outputting the category of human motion as a motion intention recognition result; and the random forest model being trained with force data corresponding to multiple categories of human motion as input and the category of human motion as output. The mechanical arm human-computer interaction method provided in the embodiment of the application adopts the random forest method to construct a human upper limb subjective motion intention recognition algorithm based on the force sensor, recognizes multiple human upper limb motion behaviors, has a certain predictability, can recognize human upper limb motion intention in advance, and improves the real-time performance of the human-computer interaction system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of human-robot interaction of mechanical arm, in particular to a human-robot interaction method and system of mechanical arm. BACKGROUND

[0002] With the transformation and upgrading of manufacturing industry and the innovative development of robot technology, the working environment and task of mechanical arm are more and more complex and diversified. Many complex tasks cannot be effectively completed by mechanical arm alone, and the participation of human is needed to complete the task. Human-robot collaboration (HRC) is a working mode in which mechanical arm and human collaborator share the working space and closely cooperate. The robot can undertake repetitive, high-load or high-precision work, and the human is responsible for tasks requiring creativity, judgment or decision-making ability. The advantages of the two are complementary, which improves the overall productivity and achieves the common goal through coordination.

[0003] Human-robot collaboration combines the advantages of high precision of mechanical arm and creativity of human, and efficiently completes complex task. In the process of human-robot collaboration, human and mechanical arm need physical contact to complete interactive task. In order to ensure that the mechanical arm can complete the task with high precision, the real-time performance, stability and safety of the system are particularly important. High real-time system can improve the precision of mechanical arm operation, and high stability can improve the operation time of mechanical arm system. At the same time, the mechanical arm does not need complex programming, and can be guided by hand to teach, and can cooperate with human in real time in highly dynamic working environment, and adapt to changing task requirements.

[0004] The human-robot interaction system of mechanical arm puts forward higher and higher requirements on the task execution precision, stability and safety of mechanical arm. The existing human-robot interaction scheme cannot meet the needs of mechanical arm to execute complex and diversified tasks. SUMMARY

[0005] To solve the above problems, the embodiment of the present application provides a human-robot interaction method of mechanical arm, comprising: acquiring force data of external force applied on the end of the mechanical arm collected by a force sensor; preprocessing the force data, the preprocessing comprising: performing time domain feature extraction, and performing linear fitting on the time domain feature; inputting the preprocessed force data into a pre-trained random forest model to output the type of human motion as a motion intention recognition result; the random forest model is trained to obtain the type of human motion as input and output, and the type of human motion includes at least one of the following: stretching, flexing and rotating.

[0006] The mechanical arm human-computer interaction method provided by the embodiment of the application adopts a random forest method to construct a human upper limb subjective motion intention recognition algorithm based on a force sensor, recognizes various human upper limb motion behaviors, has a certain predictability, can recognize human upper limb motion intention in advance, and improves the real-time performance of a human-computer interaction system.

[0007] Optionally, the time domain feature extraction comprises: time domain feature extraction on the force data of a single axis, and detection on a time domain rising edge and a falling edge; if it is detected that the time domain feature corresponding curve maintains a rising edge or a falling edge state within a preset time length, then the rising edge or the falling edge is subjected to linear fitting.

[0008] In the embodiment of the application, the force data is subjected to time domain feature extraction and rising edge or falling edge linear fitting, so as to avoid vibration errors caused by noise.

[0009] Optionally, the preprocessing further comprises: filtering processing of the force data by using Kalman filtering; or, end execution gravity contained in the force data is removed through gravity calibration.

[0010] In the embodiment of the application, the force data is subjected to filtering and gravity calibration, so as to improve the accuracy of the force data.

[0011] Optionally, the decision tree of the random forest model adopts a classification and regression tree algorithm; the classification and regression tree algorithm adopts a minimum mean square error as a division principle; and the classification and regression tree algorithm adopts a Pearson correlation coefficient to measure vector similarity.

[0012] In the embodiment of the application, the random forest model is used for motion intention recognition, and the classification and regression tree algorithm is specifically used, so as to improve the real-time performance of the human-computer interaction system.

[0013] Optionally, the force data comprises multi-axis force data, and each-axis force data is used for training of the random forest model and motion intention recognition.

[0014] In the embodiment of the application, each-axis data of multi-dimensional force data is used for model training and recognition, so as to improve training and recognition efficiency.

[0015] Optionally, the method further comprises: a force sensor zero drift compensation algorithm constructed by using a Pi-Sigma fuzzy neural network is used to compensate for the force sensor zero drift error.

[0016] In the embodiment of the application, the force sensor zero drift is compensated for by using a Pi-Sigma fuzzy neural network to construct a compensation algorithm, the influence of zero drift on force data is reduced, the single-start use time length of the six-dimensional force sensor is prolonged, and the stability of the human-computer interaction system is improved.

[0017] Optionally, the Pi-Sigma fuzzy neural network uses the zero-point drift error of the force sensor as a result of changes in usage time, force magnitude, and temperature data as a training set, uses the usage time, force magnitude, and temperature data of the force sensor as input, and uses the zero-point drift compensation of the force sensor as output.

[0018] In this embodiment of the invention, a Pi-Sigma fuzzy neural network is used for training. The Pi-Sigma fuzzy neural network whose training error reaches the error performance index is taken as the complete model.

[0019] Optionally, the Pi-Sigma fuzzy neural network includes an input layer, a fuzzification layer, a fuzzy inference layer, and an output layer;

[0020] The output of the blurring layer is:

[0021]

[0022] In the formula Let represent the membership degree, where i is the set of membership functions for the i-th input, and j is the j-th membership function for the i-th input. b is the center of the j-th membership function of the i-th input. j The width of the membership function;

[0023] Preliminary reasoning for rule construction:

[0024]

[0025] In the formula ω k Let ω be the k-th weight;

[0026] The output of each inference rule is: The reasoning conclusion for each rule is: l represents the number of preliminary reasoning layers;

[0027] The output layer result is:

[0028]

[0029] After constructing the forward channel of the Pi-Sigma fuzzy neural network, a back-learning algorithm is built, and the output layer weights are adjusted as follows:

[0030]

[0031]

[0032] The learning algorithm for weight p is as follows:

[0033]

[0034] wherein, η is a learning rate, and α is a momentum factor.

[0035] The embodiment of the present application provides the specific structure of the Pi-Sigma fuzzy neural network and the expression defined by each structure, and can realize zero-point drift error compensation of a force sensor.

[0036] The embodiment of the present application provides a mechanical arm human-computer interaction system for executing the method.

[0037] Optionally, the random forest model is trained by taking force data corresponding to a plurality of human body motion categories as input and taking the category of the human body motion as output, and the category of the human body motion includes at least one of the following: stretching, bending and rotating.

[0038] The mechanical arm human-computer interaction system provided by the embodiment of the present application can achieve the same technical effect as the mechanical arm human-computer interaction method. BRIEF DESCRIPTION OF DRAWINGS

[0039] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiment or prior art description. Obviously, the drawings in the following description are only embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of the provided drawings.

[0040] Figure 1 The embodiment of the present application provides a structure block diagram of a human-computer interaction system;

[0041] Figure 2 The embodiment of the present application provides a flowchart of a mechanical arm human-computer interaction method.

[0042] Figure 3 The embodiment of the present application provides a Pi-Sigma fuzzy neural network structure diagram. DETAILED DESCRIPTION

[0043] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0044] The information acquisition of the existing human-computer interaction mostly uses force sensors, visual sensors, voice sensors and the like. The force sensor is widely used in the field of human-computer interaction due to its fast response speed, good stability and accurate data. Therefore, the embodiment of the present application selects the force sensor as the acquisition device of human behavior operation data in the human-computer interaction process.

[0045] The embodiment of the application is applied to a multi-degree-of-freedom mechanical arm terminal mounted force sensor and a terminal executor. The terminal executor serves as a contact medium between an operator and the mechanical arm. The operator applies force on the force sensor by holding the terminal executor, and the force sensor obtains force data in this way. Exemplarily, the embodiment of the application adopts a six-dimensional force sensor.

[0046] Different from existing research technologies, the embodiment of the application identifies subjective motion intention of a human upper limb based on force data of the six-dimensional force sensor, and predicts a zero-point drift mechanism of the six-dimensional force sensor. A mechanical arm human-machine interaction method based on the six-dimensional force sensor is designed and implemented, and the human-machine interaction method is applied to the mechanical arm with high reliability and high real-time performance.

[0047] Figure 1 A structure block diagram of a human-machine interaction system provided by the embodiment of the application is shown. In terms of hardware, the system mainly includes a mechanical arm, a lower computer, an upper computer, a force sensor and a terminal executor. The mechanical arm and the lower computer, the lower computer and the upper computer, and the upper computer and the force sensor all adopt Ethernet communication.

[0048] The upper computer includes a human-machine interaction main program, which can obtain force data, mechanical arm states and zero-point drift compensation of a Pi-Sigma fuzzy neural network, and perform motion intention identification, trajectory planning and compliant control. The lower computer can perform motion instruction receiving, mechanical arm state obtaining and mechanical arm motion control. The multi-degree-of-freedom mechanical arm is connected with the lower computer and is controlled by the lower computer, and performs position response and tracking. Exemplarily, the six-dimensional force sensor is mounted on a robot terminal executor, and a human upper limb acting force acts on the terminal executor, and the six-dimensional force sensor can collect corresponding force data.

[0049] In the embodiment of the application, a random forest method is adopted to construct a human upper limb subjective motion intention identification algorithm based on the force sensor, a plurality of human upper limb motion behaviors are identified, and the algorithm has a certain predictability, can identify human upper limb motion intention in advance, and improves real-time performance of the human-machine interaction system.

[0050] Further, in the embodiment of the application, a Pi-Sigma fuzzy neural network is adopted to construct a force sensor zero-point drift compensation algorithm, the influence of zero-point drift on force data is reduced, the single-start use duration of the force sensor is prolonged, and the stability of the human-machine interaction system is improved.

[0051] The embodiment of the application provides a mechanical arm human-machine interaction method, referring to a flowchart of the mechanical arm human-machine interaction method shown in Figure 2 The method mainly includes the following steps:

[0052] S202, force data of an external force applied on a terminal end of a mechanical arm collected by a force sensor is obtained.

[0053] S204, pre-process the force data. The pre-processing includes: time domain feature extraction, and linear fitting of the time domain features.

[0054] The time domain feature extraction in this embodiment includes: time domain feature extraction of the force data of a single axis, and detection of the rising edge and falling edge in the time domain; if the curve corresponding to the detected time domain feature remains in the rising edge or falling edge state within a preset time length, the rising edge or falling edge is linearly fitted.

[0055] The force data of each axis of the multi-dimensional force sensor needs to be subjected to time domain feature extraction, and the rising edge and falling edge are fitted therefrom.

[0056] Further, the pre-processing can further include: filtering the force data by Kalman filtering; or removing the end execution gravity contained in the force data by gravity calibration.

[0057] In this embodiment, the initial data can be processed by Kalman filtering, the end execution gravity is removed by gravity calibration, and then the filtered force data is subjected to time domain feature extraction, the rising edge and falling edge in the time domain are detected, and the least square method is used to fit the rising edge and falling edge of the force data in combination with the residual sum of squares and the partial derivative function, and a variable-length moving time window is set to avoid distortion of the rising edge and falling edge fitting.

[0058] The rising edge and falling edge of the single-axis data of the filtered force sensor data are fitted to avoid vibration errors caused by noise, and according to the force sensor data, the rising or falling edge state of the force data curve within a period of time is detected, and then fitting is performed.

[0059] Exemplarily, the least square fitting method is used in this embodiment. The residual sum of squares of the dependent variable values on the fitted straight line and the actual values is minimized as the optimization target, and the correlation coefficient of the fitted straight line is determined.

[0060] S206, input the pre-processed force data into a pre-trained random forest model, and output the category of human motion as the motion intention recognition result.

[0061] The random forest model is trained by taking the force data corresponding to multiple categories of human motions as input and the category of human motion as output, and the category of human motion includes at least one of the following: stretching, flexing, and rotating.

[0062] The embodiment is directed to the random forest (RF) machine learning method for pattern recognition of sensor six-axis force data (Fx, Fy, Fz, Mx, My, Mz), and various upper limb movement acts (extension, flexion, rotation) are used as a training set to identify the human upper limb movement intention, and finally the RF model is used to decode the movement intention, infer the sensor data in real time, and output the predicted upper limb movement intention.

[0063] The RF model adopts multiple decision trees for integration, and multiple decision trees are integrated. Multiple regression trees jointly constitute a random forest regression model, and the decision trees in the random forest model are independent of each other. Each decision tree in the model jointly determines the final output of the regression model.

[0064] Optionally, the decision tree of the random forest model adopts a classification and regression tree algorithm (CART). CART is a binary tree, and only yes or no can be selected when making a decision at each node. Each node to be split selects an optimal feature value from the data set as the splitting condition.

[0065] Optionally, the classification and regression tree algorithm adopts the minimum mean squared error (MSE) as the division principle. The classification and regression tree algorithm adopts the Pearson correlation to measure the vector similarity.

[0066] In the embodiment, various upper limb movement acts (extension, flexion, rotation) force data can be input into the RF model as a training set to construct a complete random forest model. The model takes six-dimensional force sensor data as input and human upper limb subjective movement intention as output.

[0067] It should be noted that the above force data can include multi-axis force data, and each axis force data can be used for training and movement intention recognition of the random forest model.

[0068] The mechanical arm human-computer interaction method provided by the embodiment adopts the random forest method to construct a human upper limb subjective movement intention recognition algorithm based on a force sensor, recognizes various human upper limb movement behaviors, has a certain predictability, can identify the human upper limb movement intention in advance, and improves the real-time performance of the human-computer interaction system.

[0069] The rising edge and falling edge of the single-axis data of the filtered force sensor data are fitted. In order to avoid the vibration error caused by noise, if it is detected that the force data curve remains in the rising or falling edge state for a period of time, fitting is performed. Exemplarily, the least square fitting method is adopted in this embodiment. The residual sum of squares of the dependent variable value on the fitted straight line and the actual value is minimized as the optimization target, and the correlation coefficient of the fitted straight line is determined.

[0070] The objective function is defined as follows:

[0071]

[0072] In the formula, x i is the value of the random variable x at i; y i is the value of the random variable y at i; and α and β are the fitting straight line coefficients.

[0073] The partial derivative of the objective function with respect to α and β is obtained as follows:

[0074]

[0075]

[0076] The fitting straight line coefficients are obtained by the zero point of the partial derivative as follows:

[0077]

[0078]

[0079] The random forest machine learning model is constructed by fitting the force sensor data in multiple axes (Fx, Fy, Fz, Mx, My, Mz).

[0080] The least mean square error MSE can be used as the CART decision tree division principle in this embodiment. The least mean square error is used to find the feature C corresponding to the division point S in the data sets D1 and D2:

[0081]

[0082] In the formula, c1 is the sample output mean of the data set D1; and c2 is the sample output mean of the data set D2.

[0083] The Pearson correlation coefficient is used to measure the similarity of the vectors:

[0084]

[0085] In the formula, x i is the value of the random variable x at i; y i is the value of the random variable y at i; is the mean of the random variable y; is the mean of the random variable x.

[0086] The Pearson coefficient ranges from (-1, 1), and when the correlation coefficient is 0, it indicates that there is no correlation between the two, and when it is negative, it indicates that the two are negatively correlated, and when it is positive, it indicates that the two are positively correlated. The correlation degree |p| is: 0.0-0.2 is extremely weak correlation or no correlation; 0.2-0.4 is weak correlation; 0.4-0.6 is moderate correlation; 0.6-0.8 is strong correlation; 0.8-1.0 is extremely correlated.

[0087] Various upper limb movement (extension, flexion, rotation) force data are input into the RF model as a training set to build a complete random forest model. The model takes six-dimensional force sensor data as input and human upper limb subjective movement intention as output. Through the six-dimensional force sensor data, the human upper limb movement intention can be identified in advance, and the real-time performance of the human-computer interaction system is improved.

[0088] Since the force sensor will have a zero drift phenomenon in actual use, in order to ensure the stability of the human-computer interaction system, the Pi-Sigma fuzzy neural network zero drift compensation algorithm is constructed according to the principle of force sensor zero drift. The algorithm can feedback and compensate the force sensor, reduce the influence of zero drift on force data, prolong the single startup use time of the force sensor, and improve the stability of the human-computer interaction system.

[0089] Based on this, the above method further includes the following steps: using the force sensor zero drift compensation algorithm constructed by the Pi-Sigma fuzzy neural network to compensate the force sensor zero drift error.

[0090] The Pi-Sigma fuzzy neural network takes the zero drift error of the force sensor changing with the use time, force size and temperature data as the training set, takes the use time, force size and temperature data of the force sensor as the input, and takes the force sensor zero drift compensation as the output.

[0091] Exemplarily, since the six-dimensional force sensor will have a zero drift phenomenon due to the use time, force size and temperature, in order to improve the stability of the human-computer interaction system and prolong the single startup use time of the six-dimensional force sensor, a six-dimensional force sensor zero drift compensation algorithm based on Pi-Sigma fuzzy neural network is constructed. The Pi-Sigma fuzzy neural network includes an input layer, a fuzzification layer, a fuzzy reasoning layer, and an output layer. The Pi-Sigma fuzzy neural network structure is as shown in Figure 3 .

[0092] The force sensor is used as the input layer of the fuzzy neural network with time, force and temperature data, and the six-dimensional force sensor zero drift compensation is used as the output layer. The fuzzy layer uses Gaussian function as the membership function, and the output of the fuzzy layer is:

[0093]

[0094] wherein is the membership degree, x i is the membership function set of the ith input, and j is the jth membership function of the ith input, is the center of the jth membership function of the ith input, b j is the membership function width.

[0095] The pre-reasoning of the rule is constructed as follows:

[0096]

[0097] wherein ω k is the kth weight ω.

[0098] The output of each reasoning rule is and the reasoning conclusion of each rule is l is the number of pre-reasoning layers, p is the weight, h is the neuron output, and is also the reasoning output.

[0099] The output layer result is:

[0100]

[0101] After the forward channel of the Pi-Sigma fuzzy neural network is constructed, the backward learning algorithm is constructed, and the output layer weight adjustment method is:

[0102]

[0103]

[0104] wherein E p is the error, which is the difference between the actual output and the expected output in the training process, and η is the learning rate;

[0105] The learning algorithm of the weight p is:

[0106]

[0107] wherein α is the momentum factor.

[0108] The actual six-dimensional force sensor zero drift error is taken as the training set of the Pi-Sigma fuzzy neural network according to the change values of the use time, force size and temperature data, and the Pi-Sigma fuzzy neural network when the training error reaches the error performance index E is taken as the complete model.

[0109] The force sensor is inputted into the fuzzy neural network with the use time, force size and temperature data as the fuzzy neural network, and the output of the fuzzy neural network is used to compensate the force sensor zero drift error, so as to reduce the influence of the zero drift on the force data, prolong the single start use time of the force sensor, and improve the stability of the human-computer interaction system.

[0110] The embodiment of the present application aims at the real-time and stability problems of the mechanical arm human-computer interaction system based on the six-dimensional force sensor, and constructs the random forest motion intention recognition algorithm and the Pi-Sigma fuzzy neural network zero drift compensation algorithm based on the subjective motion intention recognition of the human upper limbs and the zero drift mechanism of the force sensor, so as to realize the real-time intention recognition of the human-computer interaction system, improve the stability of the human-computer interaction system, and reach the leading level in the real-time and stability indexes.

[0111] The embodiment of the present application provides a mechanical arm human-computer interaction system for executing the steps of the method provided by the above-mentioned embodiment.

[0112] Optionally, the system comprises a random forest model, the random forest model is trained with the force data corresponding to multiple types of human motions as the input and the types of human motions as the output, and the types of human motions comprise at least one of the following: stretching, bending and rotating.

[0113] The above-mentioned mechanical arm human-computer interaction system provided by the embodiment of the present application has the same implementation principle and generated technical effects as the foregoing embodiment, and for brief description, the part of the system embodiment not mentioned can refer to the corresponding content in the foregoing method embodiment.

[0114] The embodiment of the present application provides an electronic device, which comprises a processor and a storage device, the storage device stores a computer program capable of running on the processor, and the processor implements the steps of the method provided by the above-mentioned embodiment when executing the computer program.

[0115] The embodiment of the present application provides a computer readable medium, wherein the computer readable medium stores computer executable instructions, and the computer executable instructions make the processor implement the method provided by the above-mentioned embodiment when the computer executable instructions are called and executed by the processor.

[0116] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by a computer degree to instruct a control device, and the program can be stored in a computer readable storage medium. The program can include the processes of the above-mentioned method embodiments when executed, and the storage medium can be a memory, a disk, an optical disk, etc.

[0117] In this document, relational terms such as first and second and the like can be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises a... " does not, without more constraints, exclude the existence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0118] The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The same or similar parts between the various embodiments can be mutually referred to.

[0119] The above description of disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A human-machine interaction method for a robotic arm, characterized in that, include: Acquire force data of the external force applied to the end of the robotic arm, collected by a six-axis force sensor; The external force is the force applied by the operator to the end effector of the robotic arm; The force data includes multi-axis force data; The force data is preprocessed, including: extracting time-domain features and fitting the time-domain features with a straight line; The preprocessed force data is input into a pre-trained random forest model. The random forest model is trained and motion intention recognition is performed using force data for each axis. The output is the type of human motion as the motion intention recognition result. The random forest model is trained by using force data corresponding to various types of human motion as input and the type of human motion as output. The type of human motion includes at least two of the following: extension, flexion, and rotation. The type of human motion is used for trajectory planning and compliant control. The temporal feature extraction includes: The force data of a single axis is subjected to time-domain feature extraction, and the rising and falling edges in the time domain are detected. If the curve corresponding to the time-domain feature is detected to remain in a rising or falling edge state within a preset time period, then a straight line fit is performed on the rising or falling edge.

2. The method according to claim 1, characterized in that, The preprocessing also includes: The force data is filtered using a Kalman filter; or, Gravity calibration removes the end-effector gravity contained in the force data.

3. The method according to claim 1, characterized in that, The decision tree of the random forest model uses a classification and regression tree algorithm; The classification and regression tree algorithm uses the minimum mean square error as the partitioning principle; the classification and regression tree algorithm uses the Pearson correlation coefficient to measure vector similarity.

4. The method according to claim 1, characterized in that, The method further includes: A zero-point drift compensation algorithm for a force sensor, constructed using a Pi-Sigma fuzzy neural network, is used to compensate for the zero-point drift error of a six-axis force sensor.

5. The method according to claim 4, characterized in that, The Pi-Sigma fuzzy neural network uses the zero-point drift error of the six-axis force sensor as a training set, which varies with usage time, force magnitude, and temperature data. It takes the usage time, force magnitude, and temperature data of the six-axis force sensor as input and the zero-point drift compensation of the six-axis force sensor as output.

6. The method according to claim 4, characterized in that, The Pi-Sigma fuzzy neural network includes an input layer, a fuzzification layer, a fuzzy inference layer, and an output layer. The output of the blurring layer is: In the formula For membership degree, i For the first i A set of membership functions for each input. j For the first i The first input j Membership function, For the first i The first input j The center of a membership function The width of the membership function; Preliminary reasoning for rule construction: In the formula For the first k Individual weights ; The output of each inference rule is: Then the reasoning conclusion of each rule is: , l This refers to the number of layers in the initial reasoning. p For weights, h For neuron output; The output layer result is: After constructing the forward channel of the Pi-Sigma fuzzy neural network, a back-learning algorithm is built, and the output layer weights are adjusted as follows: In the formula, The error is the difference between the actual output and the expected output during the training process. For learning rate; weight p The learning algorithm is as follows: In the formula, This is the momentum factor.

7. A robotic arm human-machine interaction system, characterized in that, Used to perform the method as described in any one of claims 1 to 6.

8. The system according to claim 7, characterized in that, This includes a random forest model, which is trained using force data corresponding to various types of human movements as input and the types of human movements as output. The types of human movements include at least two of the following: extension, flexion, and rotation.

Images