Sensorless control method and device of pneumatic soft body actuator, pneumatic soft body actuator and electronic equipment

Abstract

The invention provides a sensorless control method and device for a pneumatic soft actuator, the pneumatic soft actuator and electronic equipment, and belongs to the field of soft robots. The method comprises the steps that the current injected air quality and pressure are obtained through a detection module arranged on an air pipe between an air pump module and an actuator body; the mass and the pressure intensity of the injected air are input into a preset actuation estimation model, and the current actuation estimation quantity of the actuator body is obtained through the actuation estimation model; obtaining an error value according to the actuation estimation quantity and the target control quantity; and the controller adjusts the gas injection amount of the gas pump module based on the error value so as to adjust the actuating amount of the actuator body. Thus, a large-size sensor does not need to be installed on the actuator body, accurate control over the actuator body can be achieved only according to the quality and the pressure intensity of the injected air, the size and the gravity of the actuator body are reduced, the actuator body is not afraid of collision and moisture, and therefore the fault risk is reduced, and the flexibility, the comfort and the portability are improved.

Description

Technical Field

[0001] The present invention relates to the field of soft robots, and in particular to a sensorless control method and device for a pneumatic soft actuator, a pneumatic soft actuator and electronic equipment. Background Art

[0002] Wearable robotic exoskeletons are revolutionizing human capabilities, with applications ranging from enhancing strength and endurance to assisting rehabilitation and improving daily activities. These devices are particularly effective by reducing physical fatigue, improving weight-bearing capacity, and helping individuals with physical disabilities regain mobility and independence.

[0003] Soft actuators are widely used in rehabilitation exoskeleton manipulators due to their advantages such as lightness, low cost and simple drive to assist patients with hand dysfunction in rehabilitation training and daily life. However, the current soft actuator control methods all require the installation of large-volume sensors such as angle sensors and torque sensors on the soft actuator body, resulting in poor flexibility, comfort and portability. In addition, the electronic components and circuits of the sensors require special anti-collision and moisture-proof treatment, resulting in a high risk of failure. Summary of the invention

[0004] In view of this, an object of the present invention is to provide a sensorless control method and device for a pneumatic soft actuator, a pneumatic soft actuator and an electronic device, which can achieve control without installing a large-volume sensor on the actuator body, greatly improving flexibility, comfort and portability, and reducing the risk of failure.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0006] In a first aspect, the present invention provides a sensorless control method for a pneumatic soft actuator, wherein the pneumatic soft actuator comprises a controller, an air pump module, a detection module and an actuator body, wherein the air pump module is connected to the actuator body through an air pipe, and the detection module is arranged on the air pipe, and the method comprises:

[0007] Obtaining current measurement values ​​of the detection module; wherein the measurement values ​​include injected air mass and pressure;

[0008] Inputting the injected air mass and the pressure into a preset actuation estimation model, and using the actuation estimation model to obtain a current actuation estimation amount of the actuator body; wherein the actuation estimation model is a model obtained after model training using experimental data of the actuator body, and characterizes the relationship between the input amount and the actuation amount of the actuator body;

[0009] Obtaining an error value according to the actuation estimation amount and the target control amount;

[0010] The error value is input into the controller, so that the controller adjusts the gas injection amount of the air pump module based on the error value to adjust the actuation amount of the actuator body.

[0011] Optionally, the method for acquiring the actuation estimation model includes:

[0012] Acquire a plurality of experimental data of the actuator body; wherein the experimental data include input quantities and output quantities, the output quantities include angles and / or torques, and the input quantities include injected air mass and pressure;

[0013] Using the output as a label, labeling each of the experimental data to obtain a sample set;

[0014] The sample set and Bayesian regularization are used to iteratively train a double-layer feedforward neural network to obtain an actuation estimation model.

[0015] Optionally, the sample set includes a first sample set with angle as a label, and a second sample set with torque as a label;

[0016] The step of iteratively training a double-layer feedforward neural network using the sample set and Bayesian regularization to obtain an actuation estimation model comprises:

[0017] Using Bayesian regularization, and using the first sample set, iteratively training the double-layer feedforward neural network to obtain an angle estimation model;

[0018] The double-layer feedforward neural network is iteratively trained using the second sample set by adopting Bayesian regularization to obtain a torque estimation model.

[0019] Optionally, the detection module includes a gas flow sensor and a pressure sensor;

[0020] The step of obtaining the current measurement value of the detection module includes:

[0021] Sampling the gas flow sensor and the pressure sensor to obtain real-time gas flow rate and pressure;

[0022] The target time period is taken from the starting time to the current time period, and the gas flow rates of each gas in the target time period are integrated in combination with the average molar mass of the gas output by the air pump to obtain the injected air mass at the current time period.

[0023] Optionally, the actuation estimation includes a first angle and a first torque; when the target control amount is an angle, the step of obtaining an error value according to the actuation estimation and the target control amount includes:

[0024] Inputting the first torque and the pressure into a preset angle prediction model, and using the angle prediction model to generate a second angle; wherein the angle prediction model is a model obtained by training the model with experimental data of the actuator body;

[0025] The first angle value and the second angle value are combined to obtain a current angle value;

[0026] The difference between the current angle value and the target control amount is taken as the error value.

[0027] Optionally, the actuation estimation includes a first angle and a first torque; when the target control amount is torque, the step of obtaining an error value according to the actuation estimation and the target control amount includes:

[0028] Inputting the first angle and the pressure into a preset torque prediction model, and using the torque prediction model to generate a second torque; wherein the torque prediction model is a model obtained by training the model with experimental data of the actuator body;

[0029] The first torque value and the second torque value are combined to obtain a current torque value;

[0030] The difference between the current torque value and the target control amount is taken as an error value.

[0031] Optionally, the actuator body includes a flexible integrated board and a plurality of airbags arranged on the flexible integrated board;

[0032] One end of the airbag is arranged on the flexible integrated board, and the distance between the arrangement positions of every two adjacent airbags is equal to a preset value;

[0033] Each of the airbags is provided with an air hole, and the airbag is connected with the trachea through the air hole.

[0034] In a second aspect, the present invention provides a sensorless control device for a pneumatic soft actuator, the pneumatic soft actuator comprising a controller, an air pump module, a detection module and an actuator body, the air pump being connected to the actuator body through an air pipe, the detection module being arranged on the air pipe, the device comprising a detection module, an estimation module, an error acquisition module and a control and adjustment module;

[0035] The detection module is used to obtain the current measurement value of the detection module; wherein the measurement value includes the injected air mass and pressure;

[0036] The estimation module is used to input the injected air mass and the pressure into a preset actuation estimation model, and obtain the current actuation estimation amount of the actuator body by using the actuation estimation model; wherein the actuation estimation model is a model obtained after model training with experimental data of the actuator body, and represents the relationship between the input amount and the actuation amount of the actuator body;

[0037] The error acquisition module is used to obtain an error value according to the actuation estimation amount and the target control amount;

[0038] The control and adjustment module is used to input the error value into the controller, so that the controller adjusts the gas injection amount of the air pump module based on the error value to adjust the actuation amount of the actuator body.

[0039] In a third aspect, the present invention provides a pneumatic soft actuator, comprising a controller, an air pump module, a detection module and an actuator body, wherein the actuator body comprises a flexible integrated board and a plurality of air bags arranged on the flexible integrated board;

[0040] One end of the airbag is arranged on the flexible integrated board, and the distance between the arrangement positions of every two adjacent airbags is equal to a preset value, and each airbag is provided with an air hole;

[0041] The air pump module is connected to the air hole of the airbag through an air pipe, and the detection module is arranged on the air pipe;

[0042] The detection module is used to continuously detect the gas injected into the actuator body by the air pump module to obtain a real-time measurement value;

[0043] The controller is used to implement the sensorless control method of the pneumatic soft actuator as described in the first aspect.

[0044] In a fourth aspect, the present invention provides an electronic device comprising a processor and a memory, wherein the memory stores a computer program that can be executed by the processor, and the processor can execute the computer program to implement the sensorless control method of the pneumatic soft actuator as described in the first aspect.

[0045] In a fifth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the sensorless control method of the pneumatic soft actuator as described in the first aspect.

[0046] The sensorless control method, device, pneumatic soft actuator and electronic device of the pneumatic soft actuator provided by the embodiment of the present invention include: obtaining the current injected air mass and pressure through the detection module arranged on the air pipe between the air pump module and the actuator body; inputting the injected air mass and pressure into the preset actuation estimation model, and using the actuation estimation model to obtain the current actuation estimation of the actuator body, wherein the actuation estimation model is a model obtained after model training with the experimental data of the actuator body, and characterizes the relationship between the input amount and the actuation amount of the actuator body; according to the actuation estimation amount and the target control amount, an error value is obtained; the error value is input into the controller, so that the controller adjusts the gas injection amount of the air pump module based on the error value to adjust the actuation amount of the actuator body. In this way, there is no need to install a large volume sensor on the actuator body, and the actuator body can be accurately controlled only according to the injected air mass and pressure, which greatly reduces the volume and gravity of the actuator body, thereby improving the flexibility, comfort and portability of the actuator body. At the same time, the limitations of sensor weight and size on the force output of the soft actuator are reduced or even eliminated, reducing the risk of mechanical and electronic failure.

[0047] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0049] Figure 1 A schematic structural diagram of a pneumatic soft actuator provided in an embodiment of the present invention is shown.

[0050] Figure 2 A schematic structural diagram of an actuator body provided by an embodiment of the present invention is shown.

[0051] Figure 3 A schematic structural diagram of an electronic device provided by an embodiment of the present invention is shown.

[0052] Figure 4 One of the flow charts of the sensorless control method provided by the embodiment of the present invention is shown.

[0053] Figure 5 Shows Figure 4 Schematic diagram of the process flow of some sub-steps of step 11.

[0054] Figure 6 The second flowchart of the sensorless control method provided by the embodiment of the present invention is shown.

[0055] Figure 7 A schematic structural diagram of a measurement experiment system provided by an embodiment of the present invention is shown.

[0056] Figure 8 One of the control system architecture diagrams provided by an embodiment of the present invention is shown.

[0057] Fig. 9 A second control system architecture diagram provided by an embodiment of the present invention is shown.

[0058] Fig.10 Shows Figure 4 One of the flowcharts of some sub-steps of step 15.

[0059] Fig.11 Shows Figure 4 The second flowchart of some sub-steps of step 15.

[0060] Fig.12 A dynamic simulation model of a pneumatic soft brake is shown.

[0061] Fig.13 The figure shows the simulation results when the actuator body is set at the ankle joint.

[0062] Fig.14 The figure shows the simulation results when the actuator body is set at the knee joint.

[0063] Fig.15 A block diagram of a sensorless control device provided by an embodiment of the present invention is shown.

[0064] Explanation of the reference numerals: 10 - pneumatic soft actuator; 110 - controller; 120 - air pump; 130 - solenoid valve; 140 - actuator body; 141 - flexible integrated board; 142 - air bag; 143 - air duct joint; 150 - air flow sensor; 160 - pressure sensor; 20 - electronic equipment; 210 - memory; 220 - processor; 230 - communication module; 30 - measurement experiment system; 310 - support platform; 320 - lever; 330 - moving column; 340 - tensioning wheel; 350 - soft rope; 360 - first inclinometer; 370 - second inclinometer; 380 - tension meter; 390 - barometer; 391 - thermometer; 40 - sensorless control device; 410 - detection module; 420 - estimation module; 430 - error acquisition module; 440 - control and adjustment module. DETAILED DESCRIPTION

[0065] The following will be combined with the accompanying drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0066] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present invention.

[0067] It should be noted that relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0068] Currently, the actuators in wearable robotic exoskeletons usually have complex structures. For example, in order to facilitate control, sensors for measuring torque, angle, etc. are set on the actuator body, which increases the weight and size of the actuator body. This not only leads to poor flexibility, comfort and lightness of the actuator body, but also limits the force output, bringing the risk of mechanical and electronic failure.

[0069] In response to the above problems, an embodiment of the present invention provides a sensorless control method and device for a pneumatic soft actuator, a pneumatic soft actuator and an electronic device, which can achieve control without installing a large-volume sensor on the actuator body, greatly reducing the mass and size of the actuator body, thereby improving the above problems.

[0070] Please refer to Figure 1 and Figure 2 An embodiment of the present invention provides a pneumatic soft actuator 10, including a controller 110, an air pump module, a detection module 410 and an actuator body 140, wherein the actuator body 140 includes a flexible integrated board 141 and a plurality of airbags 142 arranged on the flexible integrated board 141.

[0071] Reference Figure 2 One end of the airbag 142 is arranged on the flexible integrated board 141, and the distance between the arrangement positions of every two adjacent airbags 142 is equal to a preset value, and each airbag 142 is provided with an air hole. That is, the multiple airbags 142 are arranged at equal intervals on the flexible integrated board 141, and the sizes and shapes of these airbags 142 are consistent.

[0072] The air pump module is connected to the air hole of the air bag 142 through the air pipe, and the detection module 410 is arranged on the air pipe.

[0073] The actuator body 140 is used to be set at any part of the application object, for example, it can be set at the knee joint of the human body, and can also be set at the ankle joint of the human body.

[0074] The detection module 410 is used to continuously detect the gas injected into the actuator body 140 by the air pump module to obtain real-time measurement values.

[0075] The controller 110 is used to implement the sensorless control method of the pneumatic soft actuator 10 provided in the embodiment of the present invention.

[0076] Reference Figure 2 In order to facilitate the inflation of the airbag 142 and avoid air leakage, an air guide pipe joint 143 can be provided on the air hole of the airbag 142, and the air guide pipe joint 143 is fixed to the airbag 142 by TPU glue. In this way, the air guide pipe joint 143 can be stably fixed and air leakage between the air hole of the airbag 142 and the air guide pipe joint 143 can be avoided.

[0077] In addition, refer to Figure 1 The air pump module includes an air pump 120 and a solenoid valve 130. The air pump 120 is connected to the actuator body 140 (ie, the air guide pipe joint 143 of each air bag 142) through an air pipe, and the solenoid valve 130 is arranged on the air pipe.

[0078] Reference Figure 1 The detection module 410 includes an air flow sensor 150 (which may also be a volume flow sensor) and a pressure sensor 160, which are arranged on the air pipe between the solenoid valve 130 and the actuator body 140. Thus, the air flow sensor 150 can measure the flow rate of the gas injected into the actuator body 140 in real time. The air pipe and the actuator body 140 are in a connected state, so the pressure sensor 160 can measure the pressure of the gas in the actuator body 140.

[0079] Furthermore, it may also include a computer device (which may be a server or a server cluster) that is communicatively connected to the controller 110 of the pneumatic soft actuator 10, which is used to train an actuation estimation model based on experimental data of the pneumatic soft actuator 10 and deploy the actuation estimation model to the controller 110.

[0080] It should be noted that the controller 110 may be, but is not limited to, a single chip microcomputer, a programmable logic controller 110, a field programmable gate array, a personal computer, etc. The material of the airbag 142 of the actuator body 140 may be TPU coated nylon or any other material with light weight, high strength and tear resistance, and the flexible integrated board 141 may be a cloth of any material.

[0081] Please refer to Figure 3 , is a block diagram of an electronic device 20, which may be Figure 1 The controller 110 in the pneumatic soft actuator 10 is shown. The electronic device 20 includes a memory 210, a processor 220 and a communication module 230. The memory 210, the processor 220 and the communication module 230 are electrically connected to each other directly or indirectly to achieve data transmission or interaction. For example, these components can be electrically connected to each other through one or more communication buses or signal lines.

[0082] The memory 210 is used to store programs or data and can be, but not limited to, a random access memory, a read-only memory, a programmable read-only memory, an erasable read-only memory, an electrically erasable read-only memory, and the like.

[0083] The processor 220 is used to read / write data or programs stored in the memory 210 and execute corresponding functions. For example, Figure 1 In the pneumatic soft actuator 10 shown, the processor 220 of the controller 110 reads the computer program stored in the memory 210 to execute the sensorless control method of the pneumatic soft actuator 10 provided in the embodiment of the present invention.

[0084] The communication module 230 is used to establish a communication connection between the electronic device 20 and other communication terminals through wired or wireless means, and is used to send and receive data through the network. Figure 1 In the pneumatic soft actuator 10 shown, the controller 110 communicates with the air pump 120, the air flow sensor 150, the pressure sensor 160 and the solenoid valve 130 respectively through the communication module 230 to send and receive data.

[0085] It should be understood that Figure 3 The structure shown is only a schematic diagram of the structure of the electronic device 20. The electronic device 20 may also include Figure 3 More or fewer components as shown, or with Figure 3 Different configurations shown. Figure 3 Each component shown in the figure can be implemented by hardware, software or a combination thereof.

[0086] Please refer to Figure 4The embodiment of the present invention provides a sensorless control method for a pneumatic soft actuator 10, comprising steps 11 to 17. Thus,: Figure 1 In the pneumatic soft actuator 10 shown, the controller 110 is Figure 2 The structure shown implements steps 11 to 17 when the processor 220 executes the computer program stored in the memory 210.

[0087] Step 11, obtaining the current measurement value of the detection module.

[0088] Among them, the measured values ​​include the injected air mass and pressure.

[0089] Step 13, inputting the injected air mass and pressure into a preset actuation estimation model, and using the actuation estimation model to obtain the current actuation estimation value of the actuator body.

[0090] The actuation estimation model is a model obtained after model training using experimental data of the actuator body, and represents the relationship between the input amount and the actuation amount of the actuator body.

[0091] Step 15, obtaining an error value according to the actuation estimation amount and the target control amount.

[0092] Step 17, inputting the error value into the controller, so that the controller adjusts the gas injection amount of the air pump module based on the error value to adjust the actuation amount of the actuator body.

[0093] Here, the target control amount may be a target torque or a target angle.

[0094] For example, in combination Figure 1 The pneumatic soft actuator 10 shown, after the actuator body 140 is worn to the knee joint or ankle joint of the human body and enters the working mode, the air flow sensor 150 measures the current gas flow rate in real time, the pressure sensor 160 measures the current gas pressure in real time, and the controller 110 samples the air flow sensor 150 and the pressure sensor 160 to obtain the real-time injected air mass and pressure.

[0095] The controller 110 inputs the current injected air mass and pressure into the preset actuation estimation model, uses the actuation estimation model to obtain the current actuation estimation of the actuator body 140, and obtains the error value according to the actuation estimation and the target control amount. Thus, the controller 110 adopts the PID control principle, based on the current error value, outputs the control voltage corresponding to the error value to the solenoid valve 130, adjusts the opening of the solenoid valve 130 in the air pump module, so as to adjust the gas flow rate and / or pressure (i.e., gas injection amount) injected into the actuator body 140, and then adjusts the actuation amount (i.e., angle or output torque) of the actuator body 140, so that the actuation amount of the actuator body 140 is closer to the target actuation amount.

[0096] Compared with the existing control method of the pneumatic soft actuator 10, the above sensorless control method can achieve precise control of the actuator body 140 only based on the measurement value (i.e., the injected air mass and pressure) of the detection module 410 arranged on the trachea, and there is no need to install sensors for measuring torque, angle, etc. on the actuator body 140, which greatly reduces the volume and gravity of the actuator body 140, thereby improving the flexibility, comfort and portability of the actuator body 140. At the same time, the limitation of the weight and size of the sensor on the force output of the soft actuator is reduced or even eliminated, reducing the risk of mechanical and electronic failures.

[0097] When the detection module 410 includes an air flow sensor 150 and a pressure sensor 160, the air flow sensor 150 obtains in real time the current gas flow rate in the injection actuator body 140. At this time, the method of obtaining the measured value (i.e., the injected air mass and pressure) can be flexibly selected, for example, the injected air mass can be estimated by using an estimation model, or the injected air mass can be calculated by using a preset calculation formula, and the implementation method is not limited.

[0098] In order to obtain a more accurate measurement value, the gas flow rate is integrated in the process of obtaining the measurement value in step 11 to obtain the concept of the injected air mass of the actuator body 140 at the current time. Figure 5 In step 11, the measured value is obtained by following steps 111 to 113.

[0099] Step 111, sampling the gas flow sensor and the pressure sensor to obtain real-time gas flow rate and pressure.

[0100] Step 113, taking the target time period from the start time to the current time period, integrating the gas flow rates of each gas in the target time period in combination with the average molar mass of the gas output by the air pump, and obtaining the injected air mass at the current time period.

[0101] Here, the starting time refers to the time when the air pump module starts to work, and the injected air mass at the current time refers to the total mass of air injected into the actuator body at the current time.

[0102] Specifically, the average air pressure during each sampling period is obtained based on the atmospheric pressure and the pressure measured by the pressure sensor. Then, the average air pressure and the average molar mass of the gas output by the air pump are combined to integrate the gas flow rates of each target period to obtain the injected air mass at the current moment.

[0103] The above-mentioned process of obtaining the injected air mass can be expressed by the formula: Where M represents the injected air mass, σ represents the gas flow rate at the nth sampling moment, μ represents the average molar mass of air, R represents the molar gas constant, and T represents the room temperature. represents the average air pressure during the sampling period, n represents the total number of sampling moments before the current moment, and Δt represents the sampling period (i.e., the interval between two adjacent sampling moments). P d Represents atmospheric pressure, P n Represents the pressure measured by the pressure sensor 160 at the nth sampling moment.

[0104] It should be noted that the average molar mass of air, room temperature, and atmospheric pressure are all known set values.

[0105] Through the above steps 111 to 113, the mass of air injected into the actuator body 140 is obtained by integration, and the pressure measured by the pressure sensor 160 during each sampling period, room temperature, atmospheric pressure and gas flow rate are considered in the integration process, so that the calculated injected air mass is more accurate.

[0106] After the injected air mass is calculated in the above manner, step 13 is performed to input the input air mass and pressure into a preset actuation estimation model to obtain the current actuation estimation value of the actuator body 140 .

[0107] The actuation estimation amount may be a first angle amount and / or a first torque amount.

[0108] The actuation estimation model characterizes the relationship between the input quantity (i.e., the injected air mass and pressure) of the actuator body 140 and the output quantity (i.e., the angle and / or torque) of the actuator. The actuation estimation model can be a model obtained by regressing the experimental data, or a model obtained by training a neural network model using the experimental data, and the acquisition method is not limited.

[0109] In order to make the error of the actuation estimation model smaller and more accurate, and at the same time improve the training speed and calculation speed of the actuation estimation model, the double-layer feedforward neural network is iteratively trained using experimental data to obtain the actuation estimation model. Figure 6 , including steps 21 to 25.

[0110] Step 21, obtaining a plurality of experimental data of the actuator body.

[0111] The experimental data include input and output quantities, the output quantities include angle and / or torque, and the input quantities include injected air mass and pressure.

[0112] Here, you can Figure 1 The actuator body 140 in the pneumatic soft actuator 10 is subjected to a simulation experiment to obtain experimental data.

[0113] For example, one can construct Figure 7 The measurement experiment system 30 shown in the figure includes a support platform 310, a lever 320, a mobile column 330, a tension wheel 340 and a soft rope 350. One end of the lever 320 is hinged on the support platform 310, the actuator body 140 is arranged between the support platform 310 and the lever 320, a slide rail is arranged on the mobile column 330, and the tension wheel 340 is arranged on the mobile column 330, and the tension wheel 340 and the slide rail are slidably matched. One end of the soft rope 350 is wound around the tension wheel 340, and the end of the soft rope 350 away from the tension wheel 340 is arranged on the lever 320. Therefore, by adjusting the position of the mobile column 330 and moving and fixing the tension wheel 340 to any position on the slide rail, the lever 320 can be fixed at any angle, thereby limiting the actuator body 140 at any angle.

[0114] The measurement experiment system 30 further includes a first inclinometer 360, a second inclinometer 370, a tension meter 380, an air flow sensor 150, a pressure sensor 160, a barometer 390, and a thermometer 391. The first inclinometer 360 and the tension meter 380 are arranged on the soft rope 350, and the second inclinometer 370 is arranged on the lever 320. The air flow sensor 150 and the pressure sensor 160 are arranged on the air pipe and are located between the actuator body 140 and the solenoid valve 130. The controller 110 is respectively connected to the air pump 120, the solenoid valve 130, the air flow sensor 150, the pressure sensor 160, the barometer 390, and the thermometer 391 for communication.

[0115] The torque of the actuator body 140 is measured and recorded by the tension meter 380, the angle of the actuator body is measured and recorded by the second inclinometer 370, and the angle measured by the first inclinometer 360 is used to determine whether the pulling direction of the soft rope 350 and the lever 320 are perpendicular to each other. The current atmospheric pressure is measured by the barometer 390, the current room temperature is measured by the thermometer 391, the gas flow rate injected into the actuator body 140 is measured by the air flow sensor 150, and the gas pressure of the pneumatic module 1 is measured by the pressure sensor 160.

[0116] Based on the above structure, the actuator body 140 can be quickly tested to obtain multiple groups of experimental data, each group of experimental data at least including the angle, torque, injected air mass and pressure of the actuator body 140.

[0117] Step 23, using the output quantity as a label, annotate each experimental data to obtain a sample set.

[0118] Step 25, using the sample set and Bayesian regularization, iteratively train the double-layer feedforward neural network to obtain an actuation estimation model.

[0119] The double-layer feedforward neural network includes a hidden layer with seven sigmoid transfer function neurons connected in sequence and an output layer with a linear transfer function.

[0120] When the training purpose is: the actuation estimation model takes the injected air mass and pressure as input and outputs torque and angle, the label of each sample in the sample set is torque + angle. When the training purpose is: the actuation estimation model takes the injected air mass and pressure as input and outputs torque, the label of each sample in the sample set is torque. When the training purpose is: the actuation estimation model takes the injected air mass and pressure as input and outputs angle, the label of each sample in the sample set is angle.

[0121] Furthermore, in order to make the estimation of the actuation estimation model more accurate, in step 25, an angle estimation model is introduced, which is trained with the injected air mass and pressure as input and the angle as output, and a torque estimation model is introduced, which is trained with the injected air mass and pressure as input and the torque as output.

[0122] At this time, the sample set obtained in step 23 includes a first sample set with angle as a label and a second sample set with torque as a label. The process of training the torque estimation model in step 25 is: using Bayesian regularization, iteratively training the double-layer feedforward neural network with the first sample set to obtain the angle estimation model; using Bayesian regularization, iteratively training the double-layer feedforward neural network with the second sample set to obtain the torque estimation model.

[0123] For example, when training the angle estimation model, a prior distribution is first defined for the model parameters of the two-layer feedforward neural network. The prior distribution can be a Gaussian distribution (corresponding to L2 regularization) or a Laplace distribution (corresponding to L1 regularization). When Gaussian distribution is selected as the prior distribution, its probability density function can be expressed as: α represents the regularization parameter, E w represents the sum of squares of model weights, Z w (α) represents the normalization factor, x i Represents the i-th parameter (such as weight or bias) of a two-layer feedforward neural network.

[0124] According to the first sample set, Bayes' theorem is used to update the prior distribution and obtain the posterior distribution. The posterior distribution can be expressed as: P(D|x,β,N) represents the likelihood function (i.e., the probability that the model observes data D given the parameter β and sample x), P(x|α,N) represents the prior distribution, P(D|α,β,N) represents the normalization factor, N represents a two-layer feedforward neural network (including the number of layers in the network and the number of neurons in each layer), and α and β represent hyperparameters (i.e., regularization parameters).

[0125] Then, batch samples are randomly selected from the first sample set, and the injected air mass and pressure in the batch samples are input into the double-layer feedforward neural network. According to the output of the double-layer feedforward neural network and the label of the batch samples, the loss value (i.e., error value) is calculated, and the data error term in the batch samples is determined, and then the posterior probability is maximized to estimate the model parameters, that is, the cost function is minimized: F(x) = βE D +E w α, E D Represents the data error term, and the hyperparameters are adjusted according to the cost function. After determining the hyperparameters, gradient descent or other optimization algorithms are used to minimize the cost function F(x) to complete an iterative training. Until the preset number of training times is reached or the convergence condition is met, the angle estimation model is obtained.

[0126] The process of training the torque estimation model is the same as the process of training the angle estimation model described above, and will not be repeated here.

[0127] The angle estimation model characterizes the relationship between the input quantity of the actuator body (i.e., the injected air mass and pressure) and the angle of the actuator, and the torque estimation model characterizes the relationship between the input quantity of the actuator body (i.e., the injected air mass and pressure) and the torque output by the actuator.

[0128] The angle estimation model can be expressed as: θ = f θ (M, P), the torque estimation model can be expressed as: τ = f τ (M,P), θ represents the angle, τ represents the torque, P represents the pressure, and M represents the injected air mass.

[0129] During the model training process, the mean square error loss function, the cross entropy loss function, or any other loss function can be used to calculate the difference between the sample label (ie, the true value) and the model's estimated value. You can also use any combination of several loss functions to calculate the difference between the sample label (ie, the true value) and the model's estimated value.

[0130] Through the above steps 21 to 25, prior knowledge is introduced into the model training through Bayesian regularization to prevent overfitting, improve the generalization ability of the model, and improve the accuracy of the angle estimation model and torque estimation model finally obtained. At the same time, a two-layer feedforward neural network with a simple structure is used as the basic model to improve the convergence speed of the model training, and the calculation speed of the model is improved by reducing the calculation amount of the model, so that it is convenient to use the model to quickly obtain the first angle quantity or the first torque quantity.

[0131] After the actuation estimation model (including the angle estimation model and the torque estimation model) is obtained in the above manner, in step 13, an adapted model can be selected according to the target control amount to generate the actuation estimation amount.

[0132] For example, when the target control amount is an angle, the angle estimation model is selected to generate the current first angle of the actuator body 140, and then in step 15 and step 17, the gas injection amount of the air pump module (i.e., the opening of the solenoid valve 130) is adjusted according to the error value between the target control amount and the first angle. At this time, the control system architecture diagram can be as follows Figure 8 shown.

[0133] When the target control amount is a torque, the torque estimation model is selected to generate the current first angle of the actuator body 140, and then in step 15 and step 17, the gas injection amount of the air pump module (i.e., the opening of the solenoid valve 130) is adjusted according to the error value between the target control amount and the first torque. At this time, the control system architecture diagram can be as follows Fig. 9 shown.

[0134] In order to improve the accuracy of the angle value and thus improve the precision of control, the idea of ​​obtaining the estimated values ​​of the angle and torque at the current moment (i.e., the first angle value and the first torque value) is introduced, one of the estimated values ​​is converted and fused with the other estimated value, so as to integrate the error between the obtained value and the target control value and adjust the opening of the solenoid valve 130.

[0135] The method of converting one of the estimators and fusing it with the other estimator can be flexibly set. For example, one of the estimators can be converted and then fused using a pre-trained conversion model, or one of the estimators can be converted and then fused according to a preset rule. The above method is only an example, and its implementation method is not limited.

[0136] When the target control amount is an angle, in order to ensure that there is no sensor on the actuator body and to improve the control accuracy, a pre-trained angle prediction model using torque and pressure as input to predict the angle is introduced in step 15, the first torque amount is converted into an angle, and then the error is calculated by fusing it with the first angle amount. Fig.10The process of obtaining the error value in step 15 includes steps 151A to 155A.

[0137] Step 151A, input the first torque and pressure into a preset angle prediction model, and use the angle prediction model to generate a second angle.

[0138] The angle prediction model is a model obtained after model training with the experimental data of the actuator body. The method of obtaining the angle prediction model is the same as the method of obtaining the angle estimation model mentioned above. That is, the training samples obtained after the experimental data are labeled (the training samples are the angles in a group of experimental data as labels, and the pressure and torque in the group of experimental data are used as input quantities), and the double-layer feedforward neural network is iteratively trained to obtain the angle prediction model, which can be expressed as θ = f θ (τ,P).

[0139] Step 153A, merge the first angle value and the second angle value to obtain the current angle value.

[0140] Here, the fusion method can be flexibly set, for example, the average value of the second angle value and the first angle value can be used as the current angle value, or the second angle value and the first angle value can be weighted fused to obtain the current angle value. The above fusion method is only an example, and the implementation method is not limited.

[0141] Step 155A, taking the difference between the current angle value and the target control amount as the error value.

[0142] Similar to the above steps 151A to 155A, when the target control amount is a torque, in order to ensure that there is no sensor on the actuator body and to improve the control accuracy, a pre-trained torque prediction model that takes angle and pressure as input to predict the angle is introduced in step 15, the first angle is converted into torque, and then the error is calculated by fusion with the first torque. Fig.11 The process of obtaining the error value in step 15 includes steps 151B to 155B.

[0143] Step 151B: input the first angle value and the pressure into a preset torque prediction model, and use the torque prediction model to generate a second torque value.

[0144] The torque prediction model is a model obtained after model training with the experimental data of the actuator body. The method of obtaining the torque prediction model is the same as the method of obtaining the torque estimation model mentioned above. That is, the training samples obtained after the experimental data are labeled (the training samples are the torque in a group of experimental data as labels, and the pressure and angle in the group of experimental data are used as input quantities), and the double-layer feedforward neural network is iteratively trained to obtain the torque prediction model, which can be expressed as τ = f τ (θ,P).

[0145] Step 153B, combining the first torque value and the second torque value to obtain a current torque value.

[0146] Here, the fusion method can be flexibly set. For example, the average value of the first torque amount and the second torque amount can be used as the current torque value, or the first torque amount and the second torque amount can be weighted and fused to obtain the current torque value. The above fusion method is only an example, and the implementation method is not limited.

[0147] Step 155B, taking the difference between the current torque value and the target control amount as the error value.

[0148] Through the above steps 151A to 155A and 151B to 155B, one of the estimates obtained by the actuation estimation model is converted into a value of the same type as the other estimate, and then fused to calculate the error, which can eliminate the error caused by the performance of a single model itself, thereby improving the overall control accuracy of the pneumatic soft actuator by improving the accuracy of the error. At the same time, there is no need to add additional sensors to the actuator body, ensuring the small size, flexibility, comfort, lightness and low failure rate of the actuator body while ensuring the high-precision control of the pneumatic soft actuator.

[0149] In order to improve the accuracy, step 15 can also introduce the conversion of the estimated torque value into an angle quantity based on the setting position, center of gravity and load of the application object of the actuator body 140, and then merge it with the first angle quantity to obtain the error value with the fused angle and the target control quantity.

[0150] For example, first, according to the center of gravity of the application object and the setting position of the actuator body on the application object, the distance between the center of gravity and the setting position is acquired.

[0151] Here, the center of gravity line refers to the line perpendicular to the ground where the center of gravity is located. The application object can be a human body, and the setting position of the actuator body 140 can be a knee joint, ankle joint, etc. of the human body. A stress sensor can be set on the application object (such as a human body), and an inertial sensor and a pressure sensor can be set at the setting position, so that the real-time center of gravity point is measured according to the stress sensor, and then the real-time center of gravity line is updated, and the setting position (coordinates) of the actuator body 140 is obtained in real time according to the measurement value of the inertial sensor, and the real-time negative weight is obtained according to the pressure sensor. Then, by projection, the distance between the center of gravity line and the setting position (vertical distance) can be calculated.

[0152] Next, the moment of inertia and load are obtained based on the load and distance.

[0153] Here, the load is obtained by multiplying the gravity acceleration, the weight and the distance. The moment of inertia is obtained by multiplying the weight and the square of the distance. The load can be obtained by the formula: Load = 9.8WL, and the moment of inertia can be obtained by the formula: J = WL 2 , L represents the distance between the center of gravity line and the setting position, and W represents the negative weight.

[0154] Next, based on the torque-angle relationship of the pneumatic soft actuator, as well as the first torque, moment of inertia and load, a conversion angle corresponding to the first torque is obtained.

[0155] Among them, the conversion angle can be obtained by the formula: τ e represents the first torque quantity, θ c represents the conversion angle, c and k are constant parameters, and s represents the frequency domain variable of the conversion. The above conversion formula is the formula obtained by Laplace transforming the dynamic Euler Lagrange equation of the pneumatic soft actuator 10.

[0156] Thus, the conversion angle value and the first angle value are integrated to obtain the current angle value.

[0157] The average of the converted angle value and the first angle value may be used as the current angle value, or the angle value and the first angle value may be weightedly fused to obtain the current angle value. The above method of obtaining the current angle value is only an example, and the implementation method is not limited.

[0158] Furthermore, the difference between the current angle value and the target control amount is taken as the error value.

[0159] In the above manner, according to the application object of the actuator body 140 and the relevant influence amount of the installation position, the first torque amount is converted into an angle, and then the converted angle amount and the estimated angle amount are fused to further reduce the angle error, and then the error value of the current control is calculated with the current angle amount and the target control amount (i.e., the target angle) obtained after fusion. This makes the error value more accurate, which helps to improve the control accuracy.

[0160] In order to verify the accuracy of the above-mentioned sensorless control method, the actuator body 140 of the pneumatic soft actuator 10 is respectively set on the knee joint and ankle joint of the human body for simulation verification.

[0161] When the pneumatic soft actuator 10 is applied to the ankle joint of a human body, the dynamic simulation model of the pneumatic soft actuator 10 is as follows: Fig.12 As shown, the angle is estimated using the angle estimation model, the torque is estimated using the torque estimation model, the PID controller simulates the controller 110 to control the injected air mass (i.e., the control quantity of the opening of the solenoid valve 130 is controlled by the PID principle output), and the load module is used to simulate the load at the ankle joint. The air mass output by the PID controller and the output pressure of the actuator simulation model are fed into the angle estimation model to estimate the angle. Then, this estimated angle is fed back to the PID controller to adjust the air mass injected into the actuator, thereby controlling the bending angle of the actuator. The load module is used to simulate the load (i.e., load torque) during the bending process of the actuator. The module takes the angle as input, outputs 0 load when the angle is less than 0°, and outputs a load of 21.2N·m when the angle exceeds 60°. Two oscilloscope modules are used to display the angle and torque during the entire simulation process.

[0162] The simulation parameters of the control system are: load = 21.2Nm; moment of inertia = 0.1037kg.m2; damping coefficient c = 1, stiffness coefficient k = 0.05. The bending angle range is 60-86°, the time delay of air diffusion in the simulation model of the actuator = 0.003s, and the simulation time = 0-20 seconds. The simulation results are as follows Fig.13 shown.

[0163] At this time, the simulation is divided into four stages. In the first stage (0-3.7 seconds), the actuator is not subjected to load, the output torque is zero, and the pressure inside the actuator is zero, and the bending angle begins to increase. When the injected gas mass is 1.2g, the angle reaches 60 degrees at 3.7 seconds. In the second stage (3.7-5 seconds), when the actuator contacts the tibial cortex, it encounters resistance. As the injected air mass increases to 2.1g, the pressure rises to 23kPa, the torque increases to 21.2N·m, and the bending angle stabilizes at 60°. In the third stage (5-10 seconds), as the air mass and pressure continue to increase until the actuator torque exceeds the load torque, the acceleration generated by this torque difference drives the knee angle to 86°. In the final stage (after 10 seconds), the actuator torque is equal to the load torque of 21.2N·m, the bending angle remains at 86°, the air mass is 2.8g, and the pressure is 27kPa. The air mass flow rate during the simulation is less than 1 g / sec; the volumetric air flow rate is less than 30 L / min.

[0164] When the pneumatic soft actuator 10 is applied to the human knee joint, the simulation system is the same as the system applied to the ankle joint. The difference is that the load module is used to simulate the load at the ankle joint. At this time, the control system computer simulation parameters are: load variation range 0-54.7Nm; moment of inertia variation range 0-0.69kg.m2; damping coefficient c = 2.5, stiffness coefficient k = 0.02. Bending angle range 40-157.1°, time delay of air diffusion in the actuator model = 0.003s, simulation time: 0-20 seconds. The simulation results are as follows Fig.14 shown.

[0165] At this time, the simulation is also divided into four stages. In the first stage (0-2.5 seconds), the load on the actuator is zero, the output torque is zero, and the bending angle begins to increase. At 2.5 seconds, the angle reaches 40°, and the mass of injected air is 0.6g. In the second stage (2.5-3.7 seconds), when the actuator contacts the outer skin of the tibia, it encounters resistance, the actuator torque increases, and the bending angle stabilizes at 40°. When the mass of injected air increases to 2.2g, the pressure rises from 0 to 48kPa, and the torque increases from 0 to 54.7N·m. In the third stage (3.7-9 seconds), the further increase in air mass and pressure causes the actuator torque to exceed the load torque. The acceleration generated by this torque difference drives the knee angle to 157.1°. In the final stage (after 9 seconds), the bending angle remains stable at 157.1°, the air mass is 2.7g, the pressure is 4.2kPa, and the torque is 0N·m. Throughout the simulation, the air mass flow rate varied between 1.6 g / sec and -1.5 g / sec, where negative flow rates indicate air is being drawn out of the actuator. The volumetric air flow rate range varied between 62 L / min and -43 L / min.

[0166] As can be seen from the above simulation results, when the above sensorless control method is adopted, the angle sensorless pneumatic soft actuator 10 can effectively support an individual's transition from a squatting position to a standing position, even with an additional weight of 90 kg (such as a barbell). The actuator body 140 operates at an internal gas pressure of less than 50 kPa, minimizing discomfort to the wearer while improving overall usability and comfort.

[0167] Based on the same concept as the sensorless control method of the pneumatic soft actuator 10, refer to Fig.15 The embodiment of the present invention also provides a sensorless control device 40, which can be applied to Figure 1 The controller 110 of the pneumatic soft actuator 10 shown includes a detection module 410 , an estimation module 420 , an error acquisition module 430 and a control and adjustment module 440 .

[0168] The detection module 410 is used to obtain the current measurement value of the detection module 410.

[0169] Among them, the measured values ​​include the injected air mass and pressure.

[0170] The estimation module 420 is used to input the injected air mass and pressure into a preset actuation estimation model, and obtain the current actuation estimation value of the actuator body 140 using the actuation estimation model.

[0171] The actuation estimation model is a model obtained by training the model with experimental data of the actuator body 140 , and represents the relationship between the input amount and the actuation amount of the actuator body 140 .

[0172] The error acquisition module 430 is used to obtain an error value according to the actuation estimation amount and the target control amount.

[0173] The control and adjustment module 440 is used to input the error value into the controller 110 so that the controller 110 adjusts the gas injection amount of the air pump module based on the error value to adjust the actuation amount of the actuator body 140 .

[0174] In addition, the sensorless control device 40 also includes a model acquisition module, which can be deployed on a computer device that is communicatively connected to the controller 110 of the pneumatic soft actuator 10 .

[0175] The model acquisition module is used to: acquire multiple experimental data of the actuator body 140; wherein the experimental data include input and output, the output includes angle and / or torque, and the input includes injected air mass and pressure; label each experimental data with the output as a label to obtain a sample set; use the sample set and Bayesian regularization to iteratively train a two-layer feedforward neural network to obtain an actuation estimation model.

[0176] The sensorless control device 40 of the above-mentioned pneumatic soft actuator 10 can realize precise control of the actuator body 140 only based on the measurement value (i.e., the injected air mass and pressure) of the detection module 410 arranged on the trachea under the coordinated action of the detection module 410, the estimation module 420, the error acquisition module 430 and the control and adjustment module 440, without installing sensors for measuring torque, angle, etc. on the actuator body 140, which greatly reduces the volume and gravity of the actuator body 140, thereby improving the flexibility, comfort and portability of the actuator body 140. At the same time, the limitation of the weight and size of the sensor on the force output of the soft actuator is reduced or even eliminated, reducing the risk of mechanical and electronic failures.

[0177] For the specific implementation and effect of the sensorless control device 40, please refer to the description of the implementation of the sensorless control method above. For example, for the specific implementation and effect of the detection module 410, please refer to the description of the relevant content of step 11 above, for the specific implementation and effect of the estimation module 420, please refer to the description of the relevant content of step 13 above, for the specific implementation and effect of the error acquisition module 430, please refer to the description of the relevant content of step 15 above, and for the specific implementation and effect of the control and adjustment module 440, please refer to the description of the relevant content of step 17 above, which will not be repeated here.

[0178] In addition, each module of the sensorless control device 40 can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of the processor 220 in the electronic device 20 in the form of hardware, or can be stored in the memory 210 of the electronic device 20 in the form of software, so that the processor 220 can call and execute the operations corresponding to each module to implement the sensorless control method of the pneumatic soft actuator 10 provided above.

[0179] An embodiment of the present invention also provides an electronic device 20, including a processor 220 and a memory 210, wherein the memory 210 stores a computer program that can be executed by the processor 220, and the processor 220 can execute the computer program to implement a sensorless control method of the pneumatic soft actuator 10 as proposed in the embodiment of the present invention.

[0180] The embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by the processor 220, the sensorless control method of the pneumatic soft actuator 10 proposed in the embodiment of the present invention is implemented.

[0181] In summary, the sensorless control method and device of the pneumatic soft actuator, the pneumatic soft actuator and the electronic device provided by the embodiments of the present invention have at least the following beneficial effects:

[0182] (1) There is no need to install a large volume sensor on the actuator body. The actuator body can be precisely controlled based on the mass and pressure of the injected air, which greatly reduces the volume and gravity of the actuator body, thereby improving the flexibility, comfort and portability of the actuator body;

[0183] (ii) The limitation of sensor weight and size on the force output of soft actuators is improved. Sensorless actuators are not afraid of collision and moisture, and can even be used in water, reducing the risk of mechanical and electronic failures.

[0184] (iii) The real-time first angle value and / or first torque value of the actuator body is obtained through the actuation estimation model, and the control is performed in combination with the PID control principle to achieve high-precision and high-response control.

[0185] In several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0186] In addition, the functional modules in the various embodiments of the present invention may be integrated together to form an independent part, or each module may exist independently, or two or more modules may be integrated to form an independent part.

[0187] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for 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 methods described in the various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory, a random access memory, a magnetic disk or an optical disk.

[0188] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A sensorless control method for a pneumatic soft actuator, characterized in that: The pneumatic soft actuator comprises a controller, an air pump module, a detection module and an actuator body, wherein the air pump module is connected to the actuator body through an air pipe, and the detection module is arranged on the air pipe. The method comprises: Obtaining current measurement values ​​of the detection module; wherein the measurement values ​​include injected air mass and pressure; Inputting the injected air mass and the pressure into a preset actuation estimation model, and using the actuation estimation model to obtain a current actuation estimation amount of the actuator body; wherein the actuation estimation model is a model obtained after model training using experimental data of the actuator body, and characterizes the relationship between the input amount and the actuation amount of the actuator body; Obtaining an error value according to the actuation estimation amount and the target control amount; The error value is input into the controller, so that the controller adjusts the gas injection amount of the air pump module based on the error value to adjust the actuation amount of the actuator body.

2. The sensorless control method of the pneumatic soft actuator according to claim 1, characterized in that: The method for acquiring the actuation estimation model includes: Acquire a plurality of experimental data of the actuator body; wherein the experimental data include input quantities and output quantities, the output quantities include angles and / or torques, and the input quantities include injected air mass and pressure; Using the output as a label, labeling each of the experimental data to obtain a sample set; The sample set and Bayesian regularization are used to iteratively train a double-layer feedforward neural network to obtain an actuation estimation model.

3. The sensorless control method of the pneumatic soft actuator according to claim 2, characterized in that: The sample set includes a first sample set with angle as label and a second sample set with torque as label; The step of iteratively training a double-layer feedforward neural network using the sample set and Bayesian regularization to obtain an actuation estimation model comprises: Using Bayesian regularization, and using the first sample set, iteratively training the double-layer feedforward neural network to obtain an angle estimation model; The double-layer feedforward neural network is iteratively trained using the second sample set by adopting Bayesian regularization to obtain a torque estimation model.

4. The sensorless control method of a pneumatic soft actuator according to any one of claims 1 to 3, characterized in that: The detection module includes a gas flow sensor and a pressure sensor; The step of obtaining the current measurement value of the detection module includes: Sampling the gas flow sensor and the pressure sensor to obtain real-time gas flow rate and pressure; The target time period is taken from the starting time to the current time period, and the gas flow rates of each gas in the target time period are integrated in combination with the average molar mass of the gas output by the air pump to obtain the injected air mass at the current time period.

5. The sensorless control method of a pneumatic soft actuator according to any one of claims 1 to 3, characterized in that: The actuation estimation quantity includes a first angle quantity and a first torque quantity; When the target control amount is an angle, the step of obtaining an error value according to the actuation estimation amount and the target control amount includes: Inputting the first torque and the pressure into a preset angle prediction model, and using the angle prediction model to generate a second angle; wherein the angle prediction model is a model obtained by training the model with experimental data of the actuator body; The first angle value and the second angle value are combined to obtain a current angle value; The difference between the current angle value and the target control amount is taken as the error value.

6. The sensorless control method for a pneumatic soft actuator according to any one of claims 1 to 3, characterized in that: The actuation estimation comprises a first angle quantity and a first torque quantity; When the target control amount is torque, the step of obtaining an error value according to the actuation estimation amount and the target control amount includes: Inputting the first angle and the pressure into a preset torque prediction model, and using the torque prediction model to generate a second torque; wherein the torque prediction model is a model obtained by training the model with experimental data of the actuator body; The first torque value and the second torque value are combined to obtain a current torque value; The difference between the current torque value and the target control amount is taken as an error value.

7. The sensorless control method for a pneumatic soft actuator according to any one of claims 1 to 3, characterized in that: The actuator body includes a flexible integrated board and a plurality of air bags arranged on the flexible integrated board; One end of the airbag is arranged on the flexible integrated board, and the distance between the arrangement positions of every two adjacent airbags is equal to a preset value; Each of the airbags is provided with an air hole, and the airbag is connected with the trachea through the air hole.

8. A sensorless control device for a pneumatic soft actuator, characterized in that: The pneumatic soft actuator comprises a controller, an air pump module, a detection module and an actuator body, the air pump is connected to the actuator body through an air pipe, the detection module is arranged on the air pipe, and the device comprises a detection module, an estimation module, an error acquisition module and a control and adjustment module; The detection module is used to obtain the current measurement value of the detection module; wherein the measurement value includes the injected air mass and pressure; The estimation module is used to input the injected air mass and the pressure into a preset actuation estimation model, and obtain the current actuation estimation amount of the actuator body by using the actuation estimation model; wherein the actuation estimation model is a model obtained after model training with experimental data of the actuator body, and represents the relationship between the input amount and the actuation amount of the actuator body; The error acquisition module is used to obtain an error value according to the actuation estimation amount and the target control amount; The control and adjustment module is used to input the error value into the controller, so that the controller adjusts the gas injection amount of the air pump module based on the error value to adjust the actuation amount of the actuator body.

9. A pneumatic soft actuator, characterized in that: It includes a controller, an air pump module, a detection module and an actuator body, wherein the actuator body includes a flexible integrated board and a plurality of air bags arranged on the flexible integrated board; One end of the airbag is arranged on the flexible integrated board, and the distance between the arrangement positions of every two adjacent airbags is equal to a preset value, and each airbag is provided with an air hole; The air pump module is connected to the air hole of the airbag through an air pipe, and the detection module is arranged on the air pipe; The detection module is used to continuously detect the gas injected into the actuator body by the air pump module to obtain a real-time measurement value; The controller is used to implement the sensorless control method of the pneumatic soft actuator as described in any one of claims 1 to 7.

10. An electronic device, characterized in that: It comprises a processor and a memory, wherein the memory stores a computer program that can be executed by the processor, and the processor can execute the computer program to implement the sensorless control method of the pneumatic soft actuator as described in any one of claims 1 to 7.

Images