Mechanical arm control method and device based on IVY-BP model and medium

Abstract

The invention relates to a mechanical arm control method and device based on an IVY-BP model and a medium. The method comprises the following steps that electromyographic signals of a user are collected through electromyographic collection equipment; preprocessing the electromyographic signals to obtain preprocessed data; dividing a time window for the preprocessed data; taking the time window as a unit, extracting an activity section according to the activity threshold value, and obtaining an action signal; performing feature extraction on the action signal, and inputting the obtained features into an IVY-BP neural network model to obtain a control action; and an action instruction for controlling the mechanical arm is generated according to the control action. Compared with the prior art, the IVY-BP neural network model is adopted to identify the electromyographic signals, so that the method has the advantages of high control accuracy, low operation delay and the like. Meanwhile, the artificial intelligence technology experience of the myoelectricity control mechanical arm can be provided, the control process of the mechanical arm is visually experienced, and the artificial intelligence robot technology is experienced while entertainment is conducted.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence control technology, and in particular to a robot arm control method, device and medium based on an IVY-BP model. Background Art

[0002] In the entertainment industry, claw machines, as a popular leisure and entertainment equipment, have long occupied a prominent position in the market.

[0003] Traditional claw machines mostly rely on manual levers, buttons or touch screens as the main control methods, which have certain limitations in terms of operation complexity, operation delay and user experience services. The lever requires users to have certain skills and experience. Due to the physical limitations of the operating equipment, it is difficult for users to accurately control the position of the gripper. Although touch screen control simplifies the operation process, it lacks an intuitive experience.

[0004] Therefore, it is very important to develop a control method with more intuitive operation and more accurate interaction. Summary of the invention

[0005] The purpose of the present invention is to provide a robot arm control method, device and medium based on the IVY-BP model in order to overcome the defects of low accuracy and poor intuitiveness in the above-mentioned prior art.

[0006] The purpose of the present invention can be achieved by the following technical solutions:

[0007] A robotic arm control method based on an IVY-BP neural network model comprises the following steps:

[0008] S1: Collect the user's electromyographic signal through electromyographic acquisition equipment;

[0009] S2: preprocessing the electromyographic signal to obtain preprocessed data;

[0010] S3: Divide the preprocessed data into time windows;

[0011] S4: Extract activity segments based on activity thresholds in time windows to obtain action signals;

[0012] S5: Extract features from the action signal and input the obtained features into the IVY-BP neural network model to obtain the control action; the IVY-BP neural network model uses the weights, biases and growth factors of the BP neural network model as individuals, calculates the optimal individual with the highest fitness, and updates the parameters of the remaining individuals based on the optimal individual to generate new individuals, thereby iteratively updating to obtain the optimal individual;

[0013] S6: Generate action instructions for controlling the robot arm according to the control action.

[0014] Furthermore, the training process of the IVY-BP neural network model is as follows:

[0015] A1: Initialize the weights, biases and growth factors of the IVY-BP neural network model, take a set of weights, biases and growth factors as individuals, calculate the fitness of the individuals, and sort them to get the optimal individuals;

[0016] A2: For individuals other than the best individual, select the first update method or the second update method according to their respective fitness to generate new individuals;

[0017] A3: Calculate the fitness of new individuals, sort and exclude them, and update the best individual;

[0018] A4: Repeat A2-A3 until the preset termination condition is reached, and use the optimal individual at this time as the weight, bias and growth factor of the IVY-BP neural network model for model training.

[0019] Furthermore, it is characterized in that the first updating method or the second updating method is selected according to their respective fitness:

[0020] If the individual fitness is greater than or equal to the preset threshold, the first update method is executed. The calculation expression of the first update method is:

[0021]

[0022] In the formula, X j For another individual in the population, Represents random numbers that follow a standard normal distribution, used to introduce randomness and diversity, GV i is the growth factor of the current individual i, X new For the updated individual;

[0023] If the individual fitness is less than the preset threshold, the second update method is executed. The calculation expression of the second update method is:

[0024]

[0025] In the formula, rand is a random number.

[0026] Further, it is characterized in that the features extracted include the absolute average value of time domain features, simple square integral, maximum amplitude, waveform length, average power frequency and peak power;

[0027] The calculation expression of the absolute average value MAV is:

[0028]

[0029] Where N is the length of the electromyographic signal, X i represents the amplitude of the electromyographic signal at the i-th sample point;

[0030] The calculation expression of the simple square integral SSI is:

[0031]

[0032] The calculation expression of the waveform length WL is:

[0033]

[0034] The calculation expression of average power frequency MPF is:

[0035]

[0036] Where SD(f) is the power spectral density function, and f is the frequency;

[0037] The calculation expression of peak power PF is:

[0038] PF=max(p1,p2,...,p N )

[0039] In the formula, p N is the power of the Nth frequency component obtained through frequency domain analysis.

[0040] Further, it is characterized in that the pre-processing comprises the following steps:

[0041] A 20-350Hz fourth-order Butterworth bandpass filter and a 50Hz notch filter are used for filtering. The expression of the fourth-order Butterworth bandpass filter is:

[0042]

[0043] Where, |H(ω)| 2 is a fourth-order Butterworth bandpass filter, ω is the frequency of the input signal, is the center frequency of the passband, B = ω H -ω L is the bandwidth of the passband, ω L =20 and ω H =350 are the lower and upper cut-off frequencies of the passband, respectively, and n=4 is the order of the filter;

[0044] The notch filter expression is:

[0045]

[0046] In the formula, ω0=2πf0 is the notch frequency, f0 is the power frequency, and Q is the quality factor, which is used to control the bandwidth.

[0047] Furthermore, it is characterized in that dividing the time window comprises the following steps:

[0048] Set the time window length and sliding step, and divide the time window according to the time window length and sliding step;

[0049] Calculate the root mean square value of the time window. The calculation expression of the root mean square value is:

[0050]

[0051] Where N represents the time window length, X i represents the amplitude of the electromyographic signal at the i-th sample point;

[0052] The activity segment extraction based on the activity threshold includes the following steps:

[0053] Compare the root mean square value with the preset activity threshold, and take the time window greater than or equal to the activity threshold as the action signal. Furthermore, the IVY-BP neural network model includes an input layer, a hidden layer, and an output layer; the input layer has 24 neurons; the hidden layer is 1 layer, with 64 neurons, and the output layer has 8 neurons; the activation function from the input layer to the hidden layer is ReLU, the activation function from the hidden layer to the output layer is Softmax, and the optimizer is Adam.

[0054] In a second aspect of the present invention, a robotic arm control device comprises a robotic arm, an electromyographic signal acquisition device and a host computer. The electromyographic signal acquisition device acquires the electromyographic signal of the user and inputs it into the host computer. When the host computer is running, any one of the above-mentioned robotic arm control methods based on the IVY-BP model is executed.

[0055] Furthermore, it also includes an interaction module, which is connected to a host computer. The host computer generates description information based on the current motion instruction of the robot arm. The interaction module is used to set the operating speed of the robot arm and display the description information.

[0056] A third aspect of the present invention is a storage medium, comprising a program stored therein, and when the program is run, any one of the above-mentioned robot arm control methods based on the IVY-BP model is executed.

[0057] Compared with the prior art, the present invention has the following beneficial effects:

[0058] 1) The present invention proposes a robotic arm control method based on the IVY-BP neural network model, which collects the electromyographic signal of the user's arm as input, uses the electromyographic signal processing method to eliminate errors, and uses an improved machine learning algorithm technology to recognize the user's actions in real time, and controls the movement and grasping actions of the robotic arm accordingly according to the actions. This method not only simplifies the operation, but also improves the accuracy and intuitiveness of the operation.

[0059] 2) The present invention is applied to the control of a robotic arm, which not only overcomes the control delay and control instability of traditional methods such as manual operating levers, but also realizes a more intuitive and accurate operating experience by performing personalized settings and receiving feedback through an interactive module.

[0060] 3) The present invention can also be applied to the control interaction of virtual environments to provide users with a more personalized and immersive operating experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 The present invention is a flow chart of the control method.

[0062] Figure 2 This is a diagram of the offline training and online recognition process of the IVY-BP neural network model of this embodiment.

[0063] Figure 3 This is a diagram of the training process of the IVY-BP neural network.

[0064] Figure 4 This is the robot arm action instruction diagram.

[0065] Figure 5 This is the confusion matrix result diagram.

[0066] Figure 6 This is a control flow chart of the claw machine in Example 2. DETAILED DESCRIPTION

[0067] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0068] Example 1

[0069] The present invention is a robot arm control method based on an IVY-BP neural network model, comprising the following steps:

[0070] S1: Collect the user's electromyographic signal through electromyographic acquisition equipment;

[0071] The electromyographic acquisition device includes a first electromyographic sensor, a second electromyographic sensor, a third electromyographic sensor and a fourth electromyographic sensor. The first electromyographic sensor is attached to the ulnar wrist extensor of the operator's forearm, the second electromyographic sensor is attached to the finger extensor of the operator's forearm, the third electromyographic sensor is attached to the ulnar wrist flexor of the operator's forearm, and the fourth electromyographic sensor is attached to the abductor pollicis longus of the operator's forearm. The acquisition actions are eight wrist actions and hand actions, namely, fist clenching, fist extension, wrist upcut, wrist downcut, palm left turn, palm right turn, thumb extension and index finger extension. The sampling rate is 1000Hz.

[0072] S2: preprocessing the electromyographic signal to obtain preprocessed data;

[0073] A 20-350Hz fourth-order Butterworth bandpass filter and a 50Hz notch filter are used for filtering. The expression of the fourth-order Butterworth bandpass filter is:

[0074]

[0075] Where, |H(ω)| 2 is a fourth-order Butterworth bandpass filter, ω is the frequency of the input signal, is the center frequency of the passband, B = ω H -ω L is the bandwidth of the passband, ω L =20 and ω H =350 are the lower and upper cut-off frequencies of the passband, respectively, and n=4 is the order of the filter;

[0076] The notch filter expression is:

[0077]

[0078] In the formula, ω0=2πf0 is the notch frequency, f0 is the power frequency, and Q is the quality factor, which is used to control the bandwidth.

[0079] Preprocessing to remove low frequency noise, baseline drift, motion artifacts and power frequency interference

[0080] S3: Divide the preprocessed data into time windows;

[0081] Since there are resting intervals in the process of collecting electromyographic signals, it is necessary to extract active segments from the collected data. The time window length is set to 150ms and the sliding step length is 75ms.

[0082] After dividing the time window, calculate the root mean square value of the time window. The calculation expression of the root mean square value is:

[0083]

[0084] Where N represents the time window length, X i represents the amplitude of the electromyographic signal at the i-th sample point;

[0085] S4: Extract activity segments based on activity thresholds in time windows to obtain action signals;

[0086] The RMS value is compared with the preset activity threshold, and the time window greater than or equal to the activity threshold is used as the action signal.

[0087] According to the experimental method, this embodiment selects RMS=0.06 as the activity threshold.

[0088] S5: Extract features of the action signal and input the obtained features into the IVY-BP model to obtain the control action;

[0089] Features are selected according to the F test method. The F test was originally used to evaluate whether a group of variables have common significance, and is now widely used in feature evaluation and selection in machine learning. Finally, a total of 6 time domain or frequency domain features, including the absolute mean value (MAV), simple square integral (SSI), maximum amplitude, waveform length (WL), mean power frequency (MPF) and peak power (PF), are selected as the electromyographic signal features in this embodiment.

[0090] The calculation expression of the absolute average value MAV is:

[0091]

[0092] Where N is the length of the electromyographic signal, X i represents the amplitude of the electromyographic signal at the i-th sample point;

[0093] The calculation expression of the simple square integral SSI is:

[0094]

[0095] The calculation expression of the waveform length WL is:

[0096]

[0097] The calculation expression of average power frequency MPF is:

[0098]

[0099] Where SD(f) is the power spectral density function, and f is the frequency;

[0100] The calculation expression of peak power PF is:

[0101] PF=max(o1,o2,...,pn )

[0102] In the formula, p N is the power of the Nth frequency component obtained through frequency domain analysis.

[0103] The classification model used in this embodiment is the IVY-BP neural network model.

[0104] Among them, the classification model is a BP neural network model, including an input layer, a hidden layer and an output layer; the input layer has 24 neurons; the hidden layer is 1 layer, with 64 neurons, and the output layer has 8 neurons, corresponding to the 8 categories in the acquisition action of this embodiment; the activation function from the input layer to the hidden layer is ReLU, the activation function from the hidden layer to the output layer is Softmax, and the optimizer is Adam.

[0105] The classification model of the present invention adopts the IVY algorithm during training, which is a biological heuristic algorithm that simulates the growth pattern of ivy. When the IVY algorithm is used for weight and bias optimization of the BP neural network, the IVY-BP neural network model uses the weight, bias and growth factor of the BP neural network model as individuals, calculates the optimal individual with the highest fitness, and updates the parameters of the remaining individuals based on the optimal individual to generate new individuals, thereby iteratively updating to obtain the optimal individual. The algorithm initializes the parameters of the neural network by searching for high-quality solutions to improve the training speed and avoid falling into the local optimum.

[0106] The offline training and online recognition process of the IVY-BP neural network model in this embodiment is as follows: Figure 2 As shown in the figure, the weights and biases of the BP neural network can be regarded as variables in the solution space of an optimization problem. The IVY algorithm calculates candidate solutions for these parameters and updates them in each iteration. The goal is to find a solution that minimizes the network error function. The objective function corresponds to the loss function of the BP neural network, the mean square error MSE, and its expression is:

[0107]

[0108] Where y i is the real data of the i-th sample, is the fitting data of the ith sample, and n is the total number of samples.

[0109] like Figure 3 As shown, the specific training process is:

[0110] Initialize the weights, biases and growth factors of the BP neural network model. Each set of weights, biases and growth factors is regarded as an individual, and all individuals constitute a population.

[0111] The expressions for initializing weights and biases are:

[0112]

[0113] Where lb and ub are the weight and bias bounds, respectively.

[0114] Growth factor GV for each individual i Calculated according to the following formula:

[0115]

[0116] This growth factor controls the individual's ability to explore the solution space.

[0117] The calculation expression for calculating individual fitness Cost is:

[0118] Cost = fobj(X new )

[0119] Where fobj is the objective function, that is, the mean square error MSE of the above loss function.

[0120] If the individual fitness is greater than or equal to the threshold, the first update method is executed. The calculation expression of the first update method is:

[0121]

[0122] In the formula, X j For another individual in the population, Represents random numbers that follow a standard normal distribution, used to introduce randomness and diversity, GV i is the growth factor of the current individual i;

[0123] If the individual fitness is less than the threshold, the second update method is executed. The calculation expression of the second update method is:

[0124]

[0125] In the formula, rand is a random number;

[0126] Calculate individual fitness again, perform competition elimination and sorting, and update the optimal individual;

[0127] Competitive elimination and sorting means sorting individuals according to fitness and retaining the top N individuals with the lowest fitness to ensure that the optimal solution is continuously retained. The termination condition is judged based on the best retained individuals: if the termination condition is met, the iteration is stopped and the best individual parameters, that is, the best initial weights and biases are output; if the termination condition is not met, it is updated again based on fitness and the population is iterated for the next time.

[0128] Repeat the previous step until the preset termination condition is reached, output the optimal initial weights and biases at this time to train the BP neural network model, and complete the training process.

[0129] In this embodiment, during the training process:

[0130] When collecting electromyographic signals, the collector collects a set of movements for 3 seconds and rests for 3 seconds, and repeats this process for a total of 2 minutes. After collecting a set of movements, the collector rests for 3 minutes to prevent muscle fatigue. The number of individuals in the population is 60, and the maximum number of iterations is 50.

[0131] S6: Generate action instructions for controlling the robot arm according to the control action.

[0132] like Figure 4 As shown, it is a robot arm motion instruction diagram of an embodiment of the present invention. The classification result is mapped into a control signal of the robot arm, wherein the action "clenching fist" corresponds to "closing the robot claw", the action "extending fist" corresponds to "opening the robot claw", the action "cutting wrist upward" corresponds to "moving the robot arm upward", the action "cutting wrist downward" corresponds to "moving the robot arm downward", the action "turning palm left" corresponds to "moving the robot arm left", the action "turning palm right" corresponds to "moving the robot arm right", the action "extending thumb" corresponds to "moving the robot arm forward", the action "extending index finger" corresponds to "moving the robot arm backward", and "no action" corresponds to "moving the robot arm still".

[0133] In this embodiment, the set_position instruction is used to control the motion of the robot arm: first, the operator's motion intention is decoded by the host computer to obtain the position coordinates of the robot arm, and the position coordinates of the robot arm are modified according to the motion intention. The modified coordinates are sent to the robot arm controller through the network cable to control the movement of the robot arm. For example, the current recognition action is "wrist cut". At this time, it is mapped to the robot arm moving in the "+z" direction, that is, moving upward. The motion instruction to control the robot arm includes the motion direction, the motion speed, and whether to wait for the execution of the preceding motion instruction. The motion direction is the six-dimensional coordinate of the robot arm, at least one dimension or one clamp action in the clamp opening and clamp closing. The six-dimensional coordinate includes the horizontal axis direction, the vertical axis direction, the sagittal axis direction, the rotation around the horizontal axis direction, the rotation around the vertical axis direction, and the rotation around the sagittal axis direction. In this embodiment, every time the robot arm receives a new motion instruction, it will interrupt the current motion instruction.

[0134] The present invention is tested and verified. Figure 5 The confusion matrix shown in the figure shows that according to the confusion matrix results, the final prediction accuracy is 99.19%, which has a high accuracy. Under the environment of sampling rate of 1000Hz and action threshold of 0.06, the time from real-time reading of data into the window buffer to online output of classification results is about 130ms, which meets the real-time requirements.

[0135] Example 2

[0136] like Figure 6 As shown, the present invention provides a robotic arm device based on Example 1, including a robotic arm, an electromyographic signal acquisition device and a host computer. The electromyographic signal acquisition device collects the electromyographic signal of the user and inputs it to the host computer via Bluetooth or a wireless network. When the host computer is running, a method for controlling the robotic arm with an electromyographic signal provided by the present invention is adopted to generate a control signal of the robotic arm. Through the TCP / IP protocol, the robotic arm controller inputs the control signal of the robotic arm to the robotic arm to control the movement of the robotic arm.

[0137] Example 2 can be used for the robotic arm device of a claw machine. Compared with the traditional joystick-controlled claw machine, the robotic arm controlled by electromyography signals has a stronger sense of interaction with the operator, thereby improving the accuracy of grasping and the user experience.

[0138] Preferably, an interactive module can also be provided in the robot arm device, the interactive module is connected to the host computer, the host computer generates description information based on the current motion instructions of the robot arm, and the interactive module is used to set the running speed of the robot arm and display the description information. Through the interactive module, users can adjust parameters according to their preferences and needs, such as the movement speed and grasping mode of the robot arm. At the same time, the information generated based on the current motion instructions of the robot arm is displayed to realize user feedback, which can be used to help students experience the interactivity of artificial intelligence technology, enhance the user experience in the process of entertainment, and further improve the intuitiveness of the robot arm control process.

[0139] Example 3

[0140] Based on the embodiment 1 of the present invention, it can also be stored in a machine-readable storage medium and run on a computer device, and can be applied to a VR device to control a virtual arm structure, or to control an intelligent prosthesis, or to remotely control a robotic arm, thereby realizing the multi-purpose application of the present invention.

[0141] In the context of the present invention, a machine-readable storage medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable storage medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0142] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A robot arm control method based on IVY-BP model, characterized in that: The following steps are involved: S1: Collect the user's electromyographic signal through electromyographic acquisition equipment; S2: preprocessing the electromyographic signal to obtain preprocessed data; S3: Divide the preprocessed data into time windows; S4: Extract activity segments based on activity thresholds in time windows to obtain action signals; S5: extracting features from the action signal, and inputting the obtained features into the IVY-BP neural network model to obtain the control action; the IVY-BP neural network model uses the weight, bias and growth factor of the BP neural network model as individuals, calculates the optimal individual with the highest fitness, and updates the parameters of the remaining individuals based on the optimal individual to generate a new individual, thereby iteratively updating to obtain the optimal individual; S6: Generate action instructions for controlling the robot arm according to the control action.

2. According to claim 1, a robot arm control method based on IVY-BP model is characterized in that: The training process of the IVY-BP neural network model is specifically as follows: A1: Initialize the weights, biases and growth factors of the IVY-BP neural network model, take a set of weights, biases and growth factors as individuals, calculate the fitness of the individuals, and sort them to get the optimal individuals; A2: For individuals other than the best individual, select the first update method or the second update method according to their respective fitness to generate new individuals; A3: Calculate the fitness of new individuals, sort and exclude them, and update the best individual; A4: Repeat A2-A3 until the preset termination condition is reached, and use the optimal individual at this time as the weight, bias and growth factor of the IVY-BP neural network model for model training.

3. A robot arm control method based on IVY-BP model according to claim 2, characterized in that: The selection of the first updating mode or the second updating mode according to the respective fitness is specifically: If the individual fitness is greater than or equal to the preset threshold, the first update method is executed, and the calculation expression of the first update method is: Where, X j For another individual in the population, Represents random numbers that follow a standard normal distribution, used to introduce randomness and diversity, GV i is the growth factor of the current individual i, X new For the updated individual; If the individual fitness is less than the preset threshold, the second update method is executed. The calculation expression of the second update method is: In the formula, rand is a random number.

4. The method for controlling a robotic arm based on the IVY-BP model according to claim 1, characterized in that: The features extracted include the absolute average value of time domain features, simple square integral, maximum amplitude, waveform length, average power frequency and peak power; The calculation expression of the absolute average value MAV is: Where N represents the length of the electromyographic signal, X i represents the amplitude of the electromyographic signal at the i-th sample point; The calculation expression of the simple square integral SSI is: The calculation expression of the waveform length WL is: The calculation expression of the average power frequency MPF is: Where SD(f) is the power spectral density function, and f is the frequency; The calculation expression of the peak power PF is: PF=max(p1,p2,...,p N ) In the formula, p N is the power of the Nth frequency component obtained through frequency domain analysis.

5. The method for controlling a robotic arm based on the IVY-BP model according to claim 1, characterized in that: The pre-treatment comprises the following steps: A 20-350Hz fourth-order Butterworth bandpass filter and a 50Hz notch filter are used for filtering. The fourth-order Butterworth bandpass filter expression is: Where, |H(ω)| 2 is a fourth-order Butterworth bandpass filter, ω is the frequency of the input signal, is the center frequency of the passband, B = ω H -ω L is the bandwidth of the passband, ω L =20 and ω H =350 are the lower and upper cut-off frequencies of the passband, respectively, and n=4 is the order of the filter; The notch filter expression is: In the formula, ω0=2πf0 is the notch frequency, f0 is the power frequency, and Q is the quality factor, which is used to control the bandwidth.

6. A robot arm control method based on IVY-BP model according to claim 1, characterized in that: The time window division comprises the following steps: Set the time window length and sliding step, and divide the time window according to the time window length and sliding step; Calculate the root mean square value of the time window, the calculation expression of the root mean square value is: Where N represents the time window length, X i represents the amplitude of the electromyographic signal at the i-th sample point; The extracting of activity segments according to the activity threshold comprises the following steps: The RMS value is compared with the preset activity threshold, and the time window greater than or equal to the activity threshold is used as the action signal.

7. The method for controlling a robotic arm based on the IVY-BP model according to claim 1, characterized in that: The IVY-BP neural network model includes an input layer, a hidden layer and an output layer; the input layer has 24 neurons; the hidden layer is 1 layer with 64 neurons, and the output layer has 8 neurons; the activation function from the input layer to the hidden layer is ReLU, the activation function from the hidden layer to the output layer is Softmax, and the optimizer is Adam.

8. A robot arm control device, characterized in that: The invention comprises a robotic arm, an electromyographic signal acquisition device and a host computer. The electromyographic signal acquisition device acquires the electromyographic signal of the user and inputs it into the host computer. When the host computer is running, the robotic arm control method based on the IVY-BP model as described in any one of claims 1 to 7 is executed.

9. A robot arm control device according to claim 8, characterized in that: It also includes an interaction module, which is connected to a host computer. The host computer generates description information based on the current motion instruction of the robot arm. The interaction module is used to set the operating speed of the robot arm and display the description information.

10. A storage medium, characterized in that: The invention comprises a program stored therein, and when the program is run, a robot arm control method based on the IVY-BP model as described in any one of claims 1 to 7 is executed.

Images