A rope-driven soft actuator platform for driving the movement of a soft body and a control method thereof

Abstract

A rope-driven soft actuator platform for driving soft base body movement and a control method thereof, the control method comprising: collecting input and deformation state data pairs of the soft base body and inputting them into a nonlinear physical information neural network to generate optimized input and deformation state data pairs and train a proxy model; using Koopman operator theory, converting the trained proxy model into a linear state space model through an extended dynamic mode decomposition method, extracting state space model parameters and dimensionality increasing functions; designing a model predictive controller to solve optimal control variables on-line and complete driving of a servo motor; and feeding back system states in real time through a data feedback sensor and feeding back to the model predictive controller after Kalman filter estimation to constitute a closed-loop optimal control and realize tracking of a desired trajectory of the soft base body and disturbance suppression of the soft base body under movement along a rotating shaft. The application has high control precision and provides a feasible control scheme for research on soft robots.

Description

Technical Field

[0001] This invention relates to the field of soft robot technology, and in particular to a rope-driven soft actuator platform and its control method for driving the movement of a soft substrate. Background Technology

[0002] Rope-driven soft robots are a type of soft robot that uses ropes for propulsion. Due to their excellent dynamic response, high load-bearing capacity, and high compatibility with various soft materials, they have become an important part of the soft robotics field. Rope-driven operation uses micro-motors or servo mechanisms to adjust the length of flexible ropes embedded in or attached to the soft robot, thereby achieving motion control. Rope-driven soft robots are currently being used in exploration, agriculture, and medicine, and show great promise.

[0003] However, traditional cable-driven soft actuator platforms often restrict the soft substrate to a fixed spatial position, meaning the soft substrate can only achieve point-to-point motion, which limits its application range to some extent. The soft substrate of traditional cable-driven soft actuator platforms is based on idealized assumptions of physical and geometric modeling, which are even more difficult to satisfy in motion, thus having limitations. Furthermore, data-driven modeling has relatively low interpretability and is difficult to optimize in later stages. Summary of the Invention

[0004] This invention provides a rope-driven software actuator platform and its control method for driving the movement of a software substrate, in order to solve the technical problems mentioned in the background art.

[0005] To achieve the above objectives, the technical solution of the present invention is implemented as follows:

[0006] This invention provides a control method for a rope-driven software actuator platform, comprising the following steps:

[0007] S1. Collect input and deformation state data pairs of the soft substrate, train a nonlinear physical information neural network using the input and deformation state data pairs, input the input and deformation state data pairs into the nonlinear physical information neural network to generate optimized input and deformation state data pairs, and train the surrogate model using the optimized input and deformation state data pairs.

[0008] S2. Utilize a nonlinear physical information neural network to generate a pair of predicted input and deformation state data for the next step. Combine this with the Koopman operator theory, and based on the pair of input and deformation state data for the next step, transform the trained surrogate model into a linear state-space model using the extended dynamic mode decomposition method. Then, extract the parameters and the upscaling function of the linear state-space model.

[0009] S3. A model predictive controller is designed based on a linear state-space model. The optimal control quantity is solved online by using the increased-dimensional vector generated by the increased-dimensional function as the prediction basis. The control command corresponding to the optimal control quantity is transmitted to the embedded microcontroller to control the motor drive module to complete the drive of the servo motor.

[0010] S4. The system status is fed back in real time through the data feedback sensor. The system status is the deformation state of the soft substrate. The system status is estimated by the Kalman filter and fed back to the model predictive controller. The model predictive controller outputs the optimal control command to drive the servo motor. This constitutes closed-loop optimization control, realizing the tracking of the desired trajectory of the soft substrate and the suppression of disturbances when the soft substrate moves along the axis of rotation.

[0011] Furthermore, step S1 specifically includes the following steps:

[0012] S11. Collect input and deformation state data pairs of the software substrate;

[0013] S12. Select a physical information neural network based on a long short-term memory neural network framework, and design a first loss function based on the selected physical information neural network.

[0014] S13. The physical information neural network is trained using the input and deformation state data pairs and the first loss function to obtain a nonlinear physical information neural network.

[0015] S14. Then, the collected input and deformation state data pairs are input into a nonlinear physical information neural network to generate optimized input and deformation state data pairs.

[0016] S15. Select a surrogate model and design a second loss function. Train the surrogate model based on the optimized input and deformation state data and the second loss function to obtain the trained surrogate model.

[0017] Furthermore, in S12, the first loss function is the sum of data loss and physical loss, whereby physical loss includes energy loss and geometric compatibility loss; the specific formula for calculating the first loss function is as follows:

[0018] ;

[0019] in, This represents data loss, specifically the error between the expected data of the physical neural network and the actual measured data. This represents the energy loss, i.e., the residual that minimizes energy. This refers to the geometric compatibility loss, which is the calculation error of the geometric compatibility between the drive rope and the soft matrix; , and These are the weighted averages of the errors for each item;

[0020] The formulas for calculating each error are as follows:

[0021] ;

[0022] ;

[0023] ;

[0024] in, N This represents the total number of steps in the future. M The number of drive ropes; L This is the total length of the soft substrate; and The physical neural network at the 1st n The predicted output at step +1 and the actual data value Arc length of the centerline of the soft matrix predicted by a physical neural network The function, its second derivative is the centerline curvature vector of the soft matrix; and The i-th driving rope is in the... n The predicted cumulative length and the actual cumulative length of the step. Represents the norm.

[0025] Furthermore, step S2 specifically includes the following steps:

[0026] S21. In offline training, first, the input quantity of the servo motor in the current step is... The deformation state of the current step software substrate The input is fed into a nonlinear physical information neural network to generate a predicted input for the next servo motor step. And the predicted deformation state of the next soft matrix. ;

[0027] S22, Input of the next servo motor based on prediction And the predicted deformation state of the next soft matrix. Using Koopman operator theory, the trained nonlinear surrogate model is transformed into an equivalent linear state-space model through an extended dynamic mode decomposition method, and the parameters of the linear state-space model are extracted. A , B , C With increasing dimensionality functions.

[0028] Furthermore, step S3 specifically includes the following steps:

[0029] S31. Design a model predictive controller based on a linear state-space model, and construct an optimization strategy for the model predictive controller;

[0030] S32. The model predictive controller uses the increased-dimensional vector output by the increased-dimensional function as the basis for prediction and solves the optimal control quantity online in a rolling manner.

[0031] S33. Convert the optimal control quantity into control instructions and transmit the control instructions to the embedded microcontroller to control the motor drive module to complete the servo motor drive.

[0032] Furthermore, step S4 specifically includes the following steps:

[0033] S41. Data feedback sensor collects real-time data on the deformation state of the current step software substrate. The data is then transmitted to a Kalman filter for optimal estimation to obtain the filtered deformation state. ;

[0034] S42. Filter the deformation state The input is fed into the dimension-upgrading function to calculate the corresponding dimension-upgrading vector. ;

[0035] S43, Model Predictive Controller with Up-Dimensional Vector As initial conditions, and based on the parameters of the state-space model A , B , C The optimal control problem is solved online, along with the desired trajectory of the soft matrix, to obtain the optimal control command for the current control cycle. The desired trajectory of the soft matrix includes the target bending angle, the target deflection angle, and the target curvature.

[0036] S44, Optimal control command The data is transmitted from the host computer to the embedded microcontroller, which then processes the optimal control commands. Converted into drive signals for the motor drive module;

[0037] S45. The motor drive module controls the servo motor to move according to the drive signal, thereby reducing the change in length of the drive rope. The soft substrate moves to its actual trajectory; the actual trajectory of the soft substrate includes the actual bending angle. Actual deflection angle and actual curvature ;

[0038] S46. Repeat steps S41 to S45, continuously optimize the model predictive controller using the optimization strategy of the model predictive controller until the error between the expected trajectory and the actual trajectory of the soft substrate is within the set range, thereby achieving tracking of the expected trajectory of the soft substrate and suppression of disturbances when the soft substrate moves along the rotation axis.

[0039] Furthermore, the input to the model predictive controller comprises three parts: the initial state, the reference trajectory, and the state-space model parameters. A , B , C :

[0040] The initial state is an up-dimensional vector. The reference trajectory is the future of the software matrix. N The expected trajectory of the step;

[0041] The optimization strategy of the model predictive controller is to seek the optimal control scheme by minimizing a multi-step prediction cost function while satisfying the constraints.

[0042] The cost function consists of two parts. The first part measures the tracking error between the system output and the desired trajectory, which is the weighted sum of squares of the difference between the predicted value of the model predictor and the desired predicted value at each step. The second part evaluates the cost of the control action, which is the weighted sum of squares of the input of the servo motor at each control step.

[0043] The cost function is calculated as follows: ;

[0044] in, Indicates the current cost; The model predicts the controller's first... n+i +1 step prediction value, Indicates the first n + i +1 step expected prediction value, Indicates the first n+i Input quantity of the servo motor; Q, R These represent the defined weighting matrices;

[0045] Constraints include state evolution constraints, state space mapping constraints, and physical implementation constraints;

[0046] The state evolution constraint states that the system state at any given step is determined by the system state at the previous step and the input quantity of the servo motor; the expression is as follows:

[0047] ;

[0048] in, Indicates the firstn + i +1 step: Deformation state of the soft substrate; Represents a function that increases dimensionality;

[0049] The state-space mapping constraint is that the system state at each step is obtained by mapping the actual deformation state of the soft matrix through an up-dimensional function, as shown in the following expression:

[0050] ;

[0051] in, The model predicts the controller's first... n+i The predicted value of the step; Show the first n+i Deformation state of the software substrate;

[0052] The physical implementation constraint is that the input values ​​of all servo motors must be within the feasible operating range of the servo motors, as expressed below:

[0053] ;

[0054] in, , These represent the minimum and maximum input values ​​for the servo motor, respectively.

[0055] In another aspect, the present invention provides a rope-driven soft actuator platform for driving the movement of a soft substrate, controlled by the above-described control method, including:

[0056] The platform fixing module includes a platform frame and component fixing modules mounted on the platform frame;

[0057] The rope drive module includes multiple servo motors mounted on the component fixing module, a rope transmission mechanism that is connected to the output end of the multiple servo motors, and a screw slide module mounted on the component fixing module. Several soft substrates are mounted on the movable part of the screw slide module, and the screw slide module is used to drive the movement of several soft substrates.

[0058] The platform control module is used to control the rope drive module.

[0059] Furthermore, the lead screw slide module includes a stepper motor and a lead screw slide assembly;

[0060] The lead screw slide assembly includes a slide rail, a slider slidably connected to the slide rail, and a lead screw that is drivenly connected to the slider; the slider is configured as the moving part of the lead screw slide module; and a stepper motor is drivenly connected to the lead screw on the lead screw slide assembly.

[0061] Furthermore, the component fixing module includes a base, a cable reel fixing component, a motor support frame, and a swivel bearing plate;

[0062] The base is fixedly connected to the slider of the lead screw slide module. The winding device is mounted on the top surface of the base. The motor support frame is mounted on the platform frame and has multiple partitions spaced apart to accommodate the servo motor. The swivel bearing plate is vertically fixed on the platform frame.

[0063] Furthermore, the rope transmission mechanism includes multiple flexible couplings, multiple horizontally spaced rotating shafts, multiple winding reels, multiple drive ropes, and several washers.

[0064] Multiple rotating shafts are connected laterally to the bearing plate of the component fixing module at one end, and the other end is connected to multiple servo motors through multiple flexible couplings.

[0065] Multiple cable reels are slidably connected to multiple rotating shafts, and multiple cable reels are connected to the inside of the cable reel fixing component in the component fixing module, so that multiple cable reels and cable reel fixing component move synchronously.

[0066] Several gaskets are respectively fixed to the bottom of several soft substrates;

[0067] One end of each of the multiple drive ropes is wound and fixed to the outer ring of multiple reels, and the other end is attached to the outer surface of several soft substrates or extends from the top of several soft substrates into the interior of the soft substrates and is finally fixed to the corresponding pads.

[0068] Furthermore, the platform control module includes a power supply, a main drive integrated module, a data feedback sensor, and a host computer;

[0069] The power supply is fixed on the platform frame and is electrically connected to the rope drive module, the main drive module, and the data feedback sensor, respectively.

[0070] Several main drive integrated modules are installed on the platform frame and electrically connected to the cable drive module;

[0071] Data feedback sensors are installed at the bottom of the platform frame to acquire and feedback the position information of several software substrates;

[0072] The host computer is placed on the periphery of the platform frame and is electrically connected to the main drive module and the data feedback sensor.

[0073] Furthermore, the main drive integrated module includes a PCB board, an embedded microcontroller, a power supply module, and a motor drive module;

[0074] The embedded microcontroller, power supply module, and motor drive module are all embedded on the PCB board. The power supply module is electrically connected to the embedded microcontroller and the motor drive module, and the motor drive module is electrically connected to multiple servo motors and ball screw slide modules.

[0075] The beneficial effects of this invention are:

[0076] 1. The control method provided by this invention combines physical information with data-driven approaches to achieve modeling of soft substrates. This method overcomes the limitations of traditional theoretical models that rely on idealized assumptions (e.g., pseudo-rigid body models require the assumption that deformable bodies can be represented as continuous circular arcs), and also avoids the shortcomings of low interpretability and weak generalization ability of "black box" models under pure data-driven approaches. By introducing a first loss function to guide the learning of the physical information neural network, the reliability of the physical information neural network in real-world environments is enhanced, providing a clear and reliable model foundation for the subsequent design of the model predictive controller and system optimization.

[0077] 2. The modeling steps provided by the control method in this invention are universal and applicable to soft substrates with different geometries, topologies, and material properties. The established physical information neural network model undergoes two transformations, improving the control efficiency of the rope-driven soft actuator platform. The complex nonlinear physical information neural network is transformed into a simpler, faster-responding surrogate model, maximizing accuracy while reducing the computational burden on the network. The surrogate model is then transformed into a linear state-space model suitable for efficient solution using the Koopman operator, improving the online computation speed and real-time control response of the model predictive controller (MPC).

[0078] 3. This invention also discloses a rope-driven soft actuator platform for driving the movement of a soft substrate. It uses a lead screw slide module to convert the rotational motion of the stepper motor into axial motion, driving the soft substrate to move along the axis of rotation. Compared with traditional rope-driven soft actuator platforms, this invention can provide the soft substrate with a certain speed and acceleration, optimize the motion and control effect, and enable the soft substrate to simulate the motion of actual application scenarios, providing a motion carrier for the research of soft substrates.

[0079] 4. In the rope-driven soft actuator platform, this invention integrates the power supply module, embedded microcontroller, and motor drive module of the platform control module into a modular PCB board. The main drive module can be combined in parallel with multiple modules. If multiple software substrates are set, each software substrate corresponds to one main drive module, enhancing the maintainability and upgradeability of the rope-driven soft actuator platform. The integrated modular design simplifies the testing and debugging process of the rope-driven soft actuator platform, further improving its ease of use. Attached Figure Description

[0080] Figure 1 This is a flowchart of the control method in this invention;

[0081] Figure 2This is a three-dimensional structural schematic diagram of the rope-driven soft actuator platform in this invention;

[0082] Figure 3 for Figure 2 View from direction A;

[0083] Figure 4 for Figure 2 A view along the B direction;

[0084] Figure 5 This is a front view of the rope-driven software actuator platform in this invention;

[0085] Figure 6 This is a schematic diagram of the modular PCB of the main drive integrated circuit in this invention.

[0086] Explanation of reference numerals in the attached figures:

[0087] 1. Soft substrate;

[0088] 2. Platform fixing module; 21. Platform frame; 22. Component fixing module; 221. Base; 222. Power supply fixing component; 223. Cable reel fixing component; 224. Motor support frame; 225. Rotary bearing connecting plate;

[0089] 3. Rope drive module; 31. Servo motor; 32. Rope transmission mechanism; 321. Flexible coupling; 322. Rotating shaft; 323. Winding device; 324. Bearing; 325. Drive rope; 326. Gasket; 33. Screw slide module; 331. Stepper motor; 332. Screw slide assembly;

[0090] 4. Platform control module; 41. Power supply; 42. Main drive integrated module; 421. Embedded microcontroller; 422. Power supply module; 423. Motor drive module; 43. Data feedback sensor; 44. Host computer. Detailed Implementation

[0091] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Preferred embodiments of the invention are shown in the drawings. However, the invention can be implemented in many other different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a thorough and complete understanding of the disclosure of the invention.

[0092] It should be noted that when a component is referred to as "fixed" or "set" on another component, it can be directly on or indirectly on the other component. When a component is referred to as "connected" to another component, it can be directly connected to or indirectly connected to the other component.

[0093] It should be understood that the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the present invention.

[0094] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0095] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0096] It should also be noted that in the embodiments of this application, the same reference numerals are used to represent the same component or part. For the same part in the embodiments of this application, the reference numerals may only be used to mark one part or component as an example in the figure. It should be understood that the reference numerals are also applicable to other identical parts or components.

[0097] Reference Figure 1 This application provides a control method for a rope-driven software actuator platform, comprising the following steps:

[0098] S1. Collect input and deformation state data pairs of soft substrate 1, train a nonlinear physical information neural network using the input and deformation state data pairs, input the input and deformation state data pairs into the nonlinear physical information neural network, generate optimized input and deformation state data pairs, and train the surrogate model using the optimized input and deformation state data pairs.

[0099] S2. Utilize a nonlinear physical information neural network to generate a pair of predicted input and deformation state data for the next step. Combine this with the Koopman operator theory, and based on the pair of input and deformation state data for the next step, transform the trained surrogate model into a linear state-space model using the extended dynamic mode decomposition method. Then, extract the parameters and the upscaling function of the linear state-space model.

[0100] S3. A model predictive controller is designed based on a linear state-space model. The optimal control quantity is solved online by using the increased-dimensional vector generated by the increased-dimensional function as the prediction basis. The control command corresponding to the optimal control quantity is transmitted to the embedded microcontroller 421 to control the motor drive module 423 to complete the driving of the servo motor 31.

[0101] S4. The system status is fed back in real time through the data feedback sensor 43. The system status is the deformation state of the soft substrate 1. The system status is estimated by the Kalman filter and fed back to the model predictive controller. The model predictive controller outputs the optimal control command to drive the servo motor 31 to move. This constitutes closed-loop optimization control, realizing the tracking of the desired trajectory of the soft substrate 1 and the suppression of disturbances when the soft substrate 1 moves along the rotating shaft 322.

[0102] In some embodiments, S1 specifically includes the following steps:

[0103] S11. Collect the input and deformation state data pairs of the software substrate 1;

[0104] S12. Select a physical information neural network based on a long short-term memory neural network framework, and design a first loss function based on the selected physical information neural network.

[0105] S13. The physical information neural network is trained using the input and deformation state data pairs and the first loss function to obtain a nonlinear physical information neural network.

[0106] S14. Then, the collected input and deformation state data pairs are input into a nonlinear physical information neural network to generate optimized input and deformation state data pairs.

[0107] S15. Select a surrogate model and design a second loss function. Train the surrogate model based on the optimized input and deformation state data and the second loss function to obtain the trained surrogate model.

[0108] In some embodiments, the first loss function in S12 is the sum of data loss and physical loss, wherein the physical loss includes energy loss and geometric compatibility loss; the specific formula for calculating the first loss function is as follows:

[0109] ;

[0110] in, This represents data loss, specifically the error between the expected data of the physical neural network and the actual measured data. This represents the energy loss, i.e., the residual that minimizes energy. This refers to the geometric compatibility loss, which is the calculation error of the geometric compatibility between the drive rope 325 and the soft substrate 1. , and These are the weighted averages of the errors for each item;

[0111] The formulas for calculating each error are as follows:

[0112] ;

[0113] ;

[0114] ;

[0115] in, N This represents the total number of steps in the future. M The number of drive ropes 325; L This is the total length of the soft base 1; and The physical neural network at the 1st n +1 step of predicted output versus actual data value, Arc length of the centerline of soft substrate 1 predicted by a physical neural network The function, its second derivative Let be the centerline curvature vector of the soft substrate 1; and The first i Root drive rope 325 in the n The predicted cumulative length and the actual cumulative length of the step. Represents the norm.

[0116] In some embodiments, S2 specifically includes the following steps:

[0117] S21. In offline training, first, the input quantity of the servo motor 31 in the current step is... The deformation state of the current step software base 1 The input is fed into a nonlinear physical information neural network to generate the predicted input for the next step of the servo motor 31. And the predicted deformation state of the next step of the software substrate 1. Among them, deformation state , , , , , , and These are the deflection angle, deflection angular velocity, bending angle, bending angular velocity, curvature, axial movement velocity, and axial movement acceleration for the current step (n steps); T Represents the transpose of a matrix;

[0118] S22, Input of the next servo motor 31 based on prediction And the predicted deformation state of the next step of the software substrate 1. Using Koopman operator theory, the trained nonlinear surrogate model is transformed into an equivalent linear state-space model through an extended dynamic mode decomposition method, and the parameters of the linear state-space model are extracted. A , B , C With increasing dimensionality functions.

[0119] In some embodiments, S3 specifically includes the following steps:

[0120] S31. Design a model predictive controller based on a linear state-space model, and construct an optimization strategy for the model predictive controller;

[0121] S32. The model predictive controller uses the increased-dimensional vector output by the increased-dimensional function as the basis for prediction and solves the optimal control quantity online in a rolling manner.

[0122] S33. Convert the optimal control quantity into a control command and transmit the control command to the embedded microcontroller 421 to control the motor drive module 423 to drive the servo motor 31.

[0123] In some embodiments, S4 specifically includes the following steps:

[0124] S41, Data feedback sensor 43 collects the deformation state of the current step software substrate 1 in real time. The data is then transmitted to a Kalman filter for optimal estimation to obtain the filtered deformation state. ;

[0125] S42. Filter the deformation state The input is fed into the dimension-upgrading function to calculate the corresponding dimension-upgrading vector. ;

[0126] S43, Model Predictive Controller with Up-Dimensional Vector As initial conditions, and based on the parameters of the state-space model A , B , C Based on the desired trajectory of software substrate 1, the optimal control problem is solved online to obtain the optimal control command for the current control cycle. The desired trajectory of the soft substrate 1 includes the target bending angle, the target deflection angle, and the target curvature.

[0127] S44, Optimal control command The data is transmitted from the host computer 44 to the embedded microcontroller 421, which then processes the optimal control commands. Converted into a drive signal for the motor drive module 423;

[0128] S45, the motor drive module 423 controls the servo motor 31 to move according to the drive signal, so that the length of the drive rope 325 decreases by the amount of length change. The soft substrate 1 moves to its actual trajectory; the actual trajectory of the soft substrate 1 includes the actual bending angle. Actual deflection angle and actual curvature ;

[0129] S46. Repeat steps S41 to S45, continuously optimize the model predictive controller using the optimization strategy of the model predictive controller until the error between the expected trajectory and the actual trajectory of the software substrate 1 is within the set range, thereby achieving tracking of the expected trajectory of the software substrate 1 and suppression of disturbances when the software substrate 1 moves along the rotation axis 322.

[0130] In some embodiments, the input to the model prediction controller comprises three parts: the initial state, the reference trajectory, and the state-space model parameters. A , B , C :

[0131] The initial state is an up-dimensional vector. The reference trajectory is the future of the soft matrix 1. N The expected trajectory of the step;

[0132] The optimization strategy of the model predictive controller is to seek the optimal control scheme by minimizing a multi-step prediction cost function while satisfying the constraints.

[0133] The cost function consists of two parts. The first part measures the tracking error between the system output and the desired trajectory, which is the weighted sum of squares of the difference between the predicted value of the model predictor and the desired predicted value at each step. The second part evaluates the cost of the control action, which is the weighted sum of squares of the input of the servo motor 31 at each control step.

[0134] The cost function is calculated as follows: ;

[0135] in, Indicates the current cost; The model predicts the controller's first... n+i +1 step prediction value, Indicates the first n + i +1 step expected prediction value, Indicates the first n+i Input quantity of step servo motor 31; Q, R These represent the defined weighting matrices;

[0136] Constraints include state evolution constraints, state space mapping constraints, and physical implementation constraints;

[0137] The state evolution constraint states that the system state at any step is jointly determined by the system state at the previous step and the input of the servo motor 31. This constraint ensures that all predictions follow the inherent dynamic characteristics of the system; the expression is as follows:

[0138] ;

[0139] in, Indicates the first n + i +1 step: Deformation state of soft substrate 1; Represents a function that increases dimensionality;

[0140] The state-space mapping constraint is that the system state at each step is obtained by mapping the actual deformation state of the software matrix 1 through an up-dimensional function. The mapping relationship links the system state inside the model-predicted controller with the actual state of the physical system, ensuring the effectiveness of the optimization. The expression is as follows:

[0141] ;

[0142] in, The model predicts the controller's first... n+i The predicted value of the step; Show the first n+i Deformation state of the software substrate 1;

[0143] The physical implementation constraint requires that the input values ​​of all servo motors 31 must be within the feasible operating range of the servo motors 31, i.e., not lower than the minimum input value and not exceeding the maximum input value. This constraint ensures the physical implementability of the control commands, as expressed below:

[0144] ;

[0145] in, , These represent the minimum and maximum input values ​​of the servo motor 31, respectively.

[0146] Indicates the first n+i +1 step model predictive controller weighted sum of squares of the difference between the predicted value and the expected predicted value; Indicates the first n+i The weighted sum of squares of the input quantities of the step servo motor 31.

[0147] This invention addresses the factors influencing control effectiveness (including the arrangement path of the drive rope 325, the actual displacement of the drive rope 325, and the preset working space of the soft substrate 1) and designs a control method for a rope-driven soft actuator platform. By limiting the driving mode of the rope drive module 3, the preset arrangement path of the drive rope 325, and the movement mode of the winding device 323, the control accuracy of the method is improved. Furthermore, the control commands and data feedback in this invention can be visualized on the actuator platform control software in the host computer 44, facilitating convenient top-down control.

[0148] Furthermore, the control method provided by this invention combines physical information with data-driven approaches, enabling the modeling of soft substrates. This avoids the limitations of existing theoretical models based on assumptions (existing theoretical models can only be realized under certain conditions, such as the pseudo-rigid body model based on the assumption that its shape can be represented by continuous arcs), and also overcomes the problem of low interpretability of "black box" models under pure data-driven approaches, facilitating subsequent model optimization.

[0149] The modeling steps provided by the control method in this invention are universal and applicable to soft substrates of various shapes, structures, and materials. The established model undergoes two transformations, improving the control efficiency of the rope-driven soft actuator platform. The complex nonlinear model is transformed into a simpler, faster-responding surrogate model, maximizing accuracy while reducing the computational burden on the network. The surrogate model is then transformed into a linear state-space model suitable for efficient solution using the Koopman operator, improving the online computation speed and real-time control response of the model predictive controller (MPC).

[0150] Reference Figures 2 to 5 A second aspect of the present invention also provides a rope-driven soft actuator platform for driving the movement of a soft substrate, controlled by the above-described control method, including:

[0151] Platform fixing module 2 includes platform frame 21 and component fixing module 22 mounted on platform frame 21;

[0152] The rope drive module 3 includes multiple servo motors 31 mounted on the component fixing module 22, a rope transmission mechanism 32 that is connected to the output end of the multiple servo motors 31, and a screw slide module 33 mounted on the component fixing module 22. Several soft substrates 1 are mounted on the movable part of the screw slide module 33, and the screw slide module 33 is used to drive the movement of several soft substrates 1.

[0153] Platform control module 4 is used to control rope drive module 3.

[0154] In some embodiments, the number of software bases 1 is one or more.

[0155] In some embodiments, the material of the soft substrate 1 is an elastic material including but not limited to silicone, and the structure includes but is not limited to cylindrical, conical, corrugated, etc.

[0156] In some embodiments, the platform frame 21 is composed of multiple modular steel L-shaped strips with multiple fixing holes, which are interconnected to form the platform frame 21.

[0157] In some embodiments, refer to Figures 3 to 5 The lead screw slide module 33 includes a stepper motor 331 and a lead screw slide assembly 332;

[0158] The lead screw slide assembly 332 includes a slide rail, a slider slidably connected to the slide rail, and a lead screw drivenly connected to the slider; the slider is configured as the movable part of the lead screw slide module 33; and a stepper motor 331 is drivenly connected to the lead screw on the lead screw slide assembly 332. The slide rail is arranged along the axial direction of the rotating shaft 322, meaning the sliding direction of the slider is the axial direction of the rotating shaft 322.

[0159] In some embodiments, refer to Figures 3 to 5 The component fixing module 22 includes a base 221, a cable reel fixing component 223, a motor support frame 224, and a swivel bearing plate 225;

[0160] The base 221 is fixedly connected to the slider of the lead screw slide module 33. The winding device fixing part 223 is mounted on the top surface of the base 221. The motor support frame 224 is mounted on the platform frame 21 and has multiple layers of partitions that accommodate the servo motor 31 at intervals. The swivel bearing plate 225 is vertically fixed on the platform frame 21.

[0161] In some embodiments, the motor support frame 224 is composed of multiple parallel and spaced-apart motor support plates and multiple copper columns spliced ​​together.

[0162] In some embodiments, refer to Figures 3 to 5 The rope transmission mechanism 32 includes multiple flexible couplings 321, multiple horizontally spaced rotating shafts 322, multiple windings 323, multiple drive ropes 325, and several gaskets 326.

[0163] Multiple rotating shafts 322 are laterally rotatably connected at one end to the rotating bearing plate 225 of the component fixing module 22, and at the other end are respectively connected to multiple servo motors 31 through multiple flexible couplings 321.

[0164] Multiple cable reels 323 are slidably connected to multiple rotating shafts 322, and the multiple cable reels 323 are connected to the inside of the cable reel fixing member 223 in the component fixing module 22, so that the multiple cable reels 323 and the cable reel fixing member 223 move synchronously.

[0165] Several gaskets 326 are respectively fixed to the bottom of several soft substrates 1;

[0166] One end of each of the multiple drive ropes 325 is wound and fixed to the outer ring of multiple reels 323, and the other end is attached to the outer surface of several soft substrates 1 or extends into the interior of several soft substrates 1 from the top and is finally fixed to the corresponding pads 326. In this embodiment and the following embodiments, the bottom of the multiple drive ropes 325 preferably passes through the interior of several soft substrates 1. This structure is simpler, more secure, and less expensive.

[0167] In this embodiment, the ratio of the number of drive ropes 325 to the number of soft substrates 1 can be 3:1 (i.e., the soft substrates 1 are controlled by three drive ropes 325), or 4:1 (i.e., the soft substrates 1 are controlled by four drive ropes 325), or other ratios.

[0168] The rope drive mechanism 32 in this invention connects the output shaft of the servo motor 31 to the rotating shaft 322 via an elastic coupling 321. A bearing 324 is fixed on the rotating bearing plate 225 on the side opposite to the output end (i.e., the output shaft) of the servo motor 31, and the output end of the servo motor 31 is kept at the same horizontal height as the rotating shaft 322. The winding device 323 is fixed on the rotating shaft 322, and the position of the winding device 323 is such that the winding surface is tangent to the preset channel at the top of the soft substrate 1. This design can reduce the friction between the soft substrate 1 and the drive rope 325 and optimize the control effect of the soft actuator platform. One end of the drive rope 325 is wound and fixed on the winding device 323, and the other end enters the rope arrangement path vertically and is limited by the aluminum sleeve to the gasket 326.

[0169] In some embodiments, the winding surface of the reel 323 in the rope drive module 3 is tangent to the top preset channel of the soft substrate 1, thereby limiting the vertical entry of the drive rope 325 into the arrangement path of the drive rope 325, thereby reducing the friction between the drive rope 325, the soft substrate 1 and the reel 323.

[0170] This invention uses a lead screw slide module 33 to convert the rotational motion of the stepper motor 331 into axial motion, driving the soft substrate 1 to move along the axis of the rotating shaft 322. Compared with the traditional rope-driven soft actuator platform, this invention can provide the soft substrate 1 with a certain speed and acceleration, optimize the motion and control effect, and enable the soft substrate 1 to simulate the motion of actual application scenarios, providing a motion carrier for the research of the soft substrate 1.

[0171] In some embodiments, refer to Figures 3 to 5 The platform control module 4 includes a power supply 41, a main drive integrated module 42, a data feedback sensor 43, and a host computer 44.

[0172] Power supply 41 is fixed on platform frame 21 and electrically connected to rope drive module 3, main drive integrated module 42 and data feedback sensor 43 respectively.

[0173] Several main drive integrated modules 42 are installed on the platform frame 21 and electrically connected to the rope drive module 3.

[0174] The data feedback sensor 43 is installed at the bottom of the platform frame 21 and is used to acquire and feed back the position information of several software substrates 1 to the host computer 44.

[0175] The host computer 44 is placed on the periphery of the platform frame 21 and is electrically connected to the main drive integrated module 42 and the data feedback sensor 43 respectively.

[0176] In some embodiments, the component fixing module 22 further includes a power supply fixing component 222, through which the power supply 41 is fixedly mounted on the platform frame 21.

[0177] In some embodiments, refer to Figure 3 and Figure 6 The main drive integrated module 42 includes a PCB board, an embedded microcontroller 421, a power supply module 422, and a motor drive module 423.

[0178] The embedded microcontroller 421, the power supply module 422, and the motor drive module 423 are all embedded on the PCB board. The power supply module 422 is electrically connected to the embedded microcontroller 421 and the motor drive module 423 respectively. The motor drive module 423 is electrically connected to multiple servo motors 31 and lead screw slide modules 33 respectively.

[0179] In some embodiments, the embedded microcontroller 421 includes a central processing unit (CPU), memory, timer, and an integrated circuit chip with multiple I / O ports. The embedded microcontroller 421 acts as a lower-level machine, enabling communication with the upper-level machine 44, providing motion commands to multiple servo motors 31 and stepper motors 331, and collecting feedback on the operating data of the motors.

[0180] In some embodiments, the host computer 44 is a computer with built-in rope-driven soft actuator platform control software developed using QT (C++ application development framework). The rope-driven soft actuator platform control software adopts an independent UI design and includes six functional parts: communication connection, motor control, software control, preset actions, parameter adjustment, and data display, providing a good human-computer interaction interface.

[0181] This invention integrates the power supply module 422, embedded microcontroller 421, and motor drive module 423 of the platform control module 4 onto a modular PCB board, allowing for parallel combination of multiple modules. If multiple software bases 1 are provided, each software base 1 corresponds to one main drive integrated module 42, enhancing the maintainability and upgradeability of the rope-driven software actuator platform. The integrated modular design simplifies the testing and debugging process of the rope-driven software actuator platform, further improving the platform's ease of use.

[0182] The working principle of the rope-driven software actuator platform based on this control method in this invention is as follows:

[0183] The host computer 44 acquires control commands and issues control instructions to the soft substrate 1. After signal preprocessing by the embedded microcontroller 421, the signals are converted into drive signals for the servo motor 31. The servo motor 31 is driven to rotate via the motor drive module 423, which in turn drives the rope transmission mechanism 32 to move. The rotating shaft 322, connected via the elastic coupling 321, rotates together. The winding reel 323 is fixed at a position where the winding surface is tangent to the preset channel of the soft substrate 1. As the rotating shaft 322 rotates coaxially, the drive rope 325 axially contracts. Since the relative position between the drive rope 325 and the soft substrate 1 is limited by the shim 326, a non-uniform stress distribution is formed between the shim 326 and the soft substrate 1. Ultimately, the linear displacement of the drive rope 325 is converted into strain of the soft substrate 1, thereby achieving control of the soft substrate 1. Simultaneously, the stepper motor 331 receives a movement signal, and the lead screw of the lead screw slide assembly 332 converts the rotational motion of the stepper motor 331 into axial movement, driving the soft substrate 1 to move along the axial direction of the rotating shaft 322. The data feedback sensor 43 acquires the speed information of each motor (including the servo motor 31 and the stepper motor 331) and the position information of the soft substrate 1, and sends it to the embedded microcontroller 421 via the main drive integrated module 42. After communication with the host computer 44, the data feedback is completed. A Kalman filter is used to process the feedback data, and the data is transmitted to the pre-established model predictive controller to realize the control closed loop.

[0184] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Furthermore, the technical solutions of the various embodiments of the present invention can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A control method for a rope-driven soft actuator platform, characterized in that, Includes the following steps: S1. Collect the input and deformation state data pairs of the soft substrate (1), use the input and deformation state data pairs to train a nonlinear physical information neural network, input the input and deformation state data pairs into the nonlinear physical information neural network, generate optimized input and deformation state data pairs, and use the optimized input and deformation state data pairs to train the surrogate model. S2. Utilize a nonlinear physical information neural network to generate a pair of predicted input and deformation state data for the next step. Combine this with the Koopman operator theory, and based on the pair of input and deformation state data for the next step, transform the trained surrogate model into a linear state-space model using the extended dynamic mode decomposition method. Then, extract the parameters and the upscaling function of the linear state-space model. S3. Based on the linear state space model, a model predictive controller is designed. The optimal control quantity is solved online by using the upgraded vector generated by the upgraded function as the prediction basis. The control command corresponding to the optimal control quantity is transmitted to the embedded microcontroller (421) and the motor drive module (423) is controlled to complete the driving of the servo motor (31). S4. The system status is fed back in real time through the data feedback sensor (43). The system status is the deformation state of the soft substrate (1). The system status is estimated by the Kalman filter and fed back to the model predictive controller. The model predictive controller outputs the optimal control command to drive the servo motor (31) to move. This constitutes closed-loop optimization control, realizing the tracking of the desired trajectory of the soft substrate (1) and the suppression of disturbances when the soft substrate (1) moves along the rotating shaft (322). S1 specifically includes the following steps: S11. Collect the input and deformation state data pairs of the software substrate (1); S12. Select a physical information neural network based on a long short-term memory neural network framework, and design a first loss function based on the selected physical information neural network. S13. The physical information neural network is trained using the input and deformation state data pairs and the first loss function to obtain a nonlinear physical information neural network. S14. Then, the collected input and deformation state data pairs are input into a nonlinear physical information neural network to generate optimized input and deformation state data pairs. S15. Select a proxy model and design a second loss function. Train the proxy model based on the optimized input and deformation state data and the second loss function to obtain the trained proxy model. In S12, the first loss function is the sum of data loss and physical loss, where physical loss includes energy loss and geometric compatibility loss; the specific formula for calculating the first loss function is as follows: ； in, This represents data loss, specifically the error between the expected data of the physical neural network and the actual measured data. This represents the energy loss, i.e., the residual that minimizes energy. The geometric compatibility loss is the calculation error of geometric compatibility between the drive rope (325) and the soft substrate (1); , and These are the weighted averages of the errors for each item; The formulas for calculating each error are as follows: ； ； ； in, N This represents the total number of steps in the future. M The number of drive ropes (325); L The total length of the soft substrate (1); and The physical neural network at the 1st n The predicted output at step +1 and the actual data value Arc length of the centerline of the soft matrix (1) predicted by the physical neural network The function, its second derivative Let be the centerline curvature vector of the soft substrate (1); and The first i root drive rope (325) in the n The predicted cumulative length and the actual cumulative length at the step point, Represents the norm.

2. The control method for the rope-driven soft actuator platform according to claim 1, characterized in that, S2 specifically includes the following steps: S21. In offline training, first, the input quantity of the servo motor (31) of the current step is... The deformation state of the current step software substrate (1) The input is fed into a nonlinear physical information neural network to generate the predicted input for the next servo motor (31). And the predicted deformation state of the next step of the soft matrix (1) ; S22, Input of the next servo motor (31) based on prediction And the predicted deformation state of the next step of the soft matrix (1) Using Koopman operator theory, the trained nonlinear surrogate model is transformed into an equivalent linear state-space model through an extended dynamic mode decomposition method, and the parameters of the linear state-space model are extracted. A , B , C With increasing dimensionality functions.

3. The control method for the rope-driven soft actuator platform according to claim 2, characterized in that, S3 specifically includes the following steps: S31. Design a model predictive controller based on a linear state-space model, and construct an optimization strategy for the model predictive controller; S32. The model predictive controller uses the increased-dimensional vector output by the increased-dimensional function as the basis for prediction and solves the optimal control quantity online in a rolling manner. S33. Convert the optimal control quantity into a control command and transmit the control command to the embedded microcontroller (421) to control the motor drive module (423) to drive the servo motor (31).

4. The control method for the rope-driven soft actuator platform according to claim 3, characterized in that, S4 specifically includes the following steps: S41, Data feedback sensor (43) collects the deformation state of the current step software substrate (1) in real time. The data is then transmitted to a Kalman filter for optimal estimation to obtain the filtered deformation state. ; S42. Filter the deformation state The input is fed into the dimension-upgrading function to calculate the corresponding dimension-upgrading vector. ; S43, Model Predictive Controller with Up-Dimensional Vector Using the initial conditions and based on the parameters A, B, and C of the state-space model and the desired trajectory of the soft matrix (1), the optimal control problem is solved online to obtain the optimal control command for the current control cycle. The desired trajectory of the soft substrate (1) includes the target bending angle, the target deflection angle, and the target curvature. S44, Optimal control command The data is transmitted from the host computer (44) to the embedded microcontroller (421), which then executes the optimal control instructions. Converted into a drive signal for the motor drive module (423); S45, the motor drive module (423) controls the servo motor (31) to move according to the drive signal, so that the length of the drive rope (325) decreases by the amount of length change. The soft substrate (1) moves to its actual trajectory; the actual trajectory of the soft substrate (1) includes the actual bending angle. Actual deflection angle and actual curvature ; S46. Repeat steps S41 to S45, continuously optimize the model predictive controller using the optimization strategy of the model predictive controller until the error between the expected trajectory and the actual trajectory of the soft substrate (1) is within the set range, thereby realizing the tracking of the expected trajectory of the soft substrate (1) and the suppression of disturbances when the soft substrate (1) moves along the rotating axis (322).

5. The control method for the rope-driven soft actuator platform according to claim 4, characterized in that, The input to the model predictive controller consists of three parts: the initial state, the reference trajectory, and the state-space model parameters. A , B , C : The initial state is an up-dimensional vector. ; The reference trajectory is the soft matrix (1) future N The expected trajectory of the step; The optimization strategy of the model predictive controller is to seek the optimal control scheme by minimizing a multi-step prediction cost function while satisfying the constraints. The cost function consists of two parts. The first part measures the tracking error between the system output and the desired trajectory, which is the weighted sum of squares of the difference between the predicted value of the model predictor and the desired predicted value at each step. The other part evaluates the cost of the control action, which is the weighted sum of squares of the input of the servo motor (31) at each step. The cost function is calculated as follows: ; in, Indicates the current cost; The model predicts the controller's first... n+i +1 step prediction value, Indicates the first n + i +1 step expected prediction value, Indicates the first n+i Input quantity of step servo motor (31); Q, R These represent the defined weighting matrices; Constraints include state evolution constraints, state space mapping constraints, and physical implementation constraints; The state evolution constraint is that the system state at any step is determined by the system state at the previous step and the input of the servo motor (31); the expression is as follows: ； in, Indicates the first n + i +1 step soft matrix (1) deformation state; Represents a function that increases dimensionality; The state space mapping constraint is that the system state at each step is obtained by mapping the actual deformation state of the soft matrix (1) through an up-dimensional function, as shown in the following expression: ； in, The model predicts the controller's first... n+i The predicted value of the step; Show the first n+i Deformation state of the soft substrate (1); The physical implementation constraint is that the input quantities of all servo motors (31) must be within the feasible operating range of the servo motors (31), as expressed below: ； in, , These represent the minimum and maximum input values ​​of the servo motor (31), respectively.

6. A rope-driven soft actuator platform for driving the movement of a soft substrate, controlled by the control method described in any one of claims 1 to 5, characterized in that, include: The platform fixing module (2) includes a platform frame (21) and a component fixing module (22) mounted on the platform frame (21). The rope drive module (3) includes multiple servo motors (31) mounted on the component fixing module (22), a rope transmission mechanism (32) connected to the output end of the multiple servo motors (31), and a screw slide module (33) mounted on the component fixing module (22). Several soft substrates (1) are mounted on the movable part of the screw slide module (33), and the screw slide module (33) is used to drive the movement of several soft substrates (1). The platform control module (4) is used to control the rope drive module (3).

7. The rope-driven software actuator platform for driving the movement of the driving software substrate according to claim 6, characterized in that, The lead screw slide module (33) includes a stepper motor (331) and a lead screw slide assembly (332). The lead screw slide assembly (332) includes a slide rail, a slider slidably connected to the slide rail, and a lead screw drivenly connected to the slider; the slider is configured as the moving part of the lead screw slide module (33); and a stepper motor (331) is drivenly connected to the lead screw on the lead screw slide assembly (332). The component fixing module (22) includes a base (221), a reel fixing component (223), a motor support frame (224), and a swivel bearing plate (225). The base (221) is fixedly connected to the slider of the lead screw slide module (33), the winding device fixing part (223) is mounted on the top surface of the base (221), the motor support frame (224) is mounted on the platform frame (21), and multiple layers of partitions for accommodating the servo motor (31) are distributed at intervals; the swivel bearing plate (225) is vertically fixed on the platform frame (21); The rope transmission mechanism (32) includes multiple flexible couplings (321), multiple horizontally spaced rotating shafts (322), multiple winding reels (323), multiple drive ropes (325), and several gaskets (326). One end of multiple rotating shafts (322) is rotatably connected to the rotating bearing plate (225) of the component fixing module (22) at one end, and the other end is connected to multiple servo motors (31) through multiple flexible couplings (321); Multiple reels (323) are slidably connected to multiple rotating shafts (322), and multiple reels (323) are connected to the inside of the reel fixing member (223) in the component fixing module (22) so that multiple reels (323) and reel fixing member (223) move synchronously; Several gaskets (326) are respectively fixed to the bottom of several soft substrates (1); One end of each of the multiple drive ropes (325) is wound and fixed on the outer ring of multiple reels (323), and the other end is attached to the outer surface of several soft substrates (1) or extends from the top of several soft substrates (1) into the interior of the soft substrate (1) and is finally fixed on the corresponding pad (326).

8. The rope-driven software actuator platform for driving the movement of the driving software substrate according to claim 7, characterized in that, The platform control module (4) includes a power supply (41), a main drive integrated module (42), a data feedback sensor (43), and a host computer (44). The power supply (41) is fixed on the platform frame (21) and is electrically connected to the rope drive module (3), the main drive integrated module (42), and the data feedback sensor (43) respectively. Several main drive integrated modules (42) are installed on the platform frame (21) and electrically connected to the rope drive module (3); A data feedback sensor (43) is installed at the bottom of the platform frame (21) to acquire and feedback the position information of several software substrates (1); The host computer (44) is placed on the periphery of the platform frame (21) and is electrically connected to the main drive integrated module (42) and the data feedback sensor (43) respectively; The main drive integrated module (42) includes a PCB board, an embedded microcontroller (421), a power supply module (422), and a motor drive module (423). The embedded microcontroller (421), power supply module (422) and motor drive module (423) are all embedded on the PCB board. The power supply module (422) is electrically connected to the embedded microcontroller (421) and the motor drive module (423) respectively. The motor drive module (423) is electrically connected to multiple servo motors (31) and ball screw slide module (33) respectively.

Images