A soft robot posture perception method and device based on multi-source data fusion

Abstract

This invention relates to the field of soft robot posture perception technology, and in particular to a method and apparatus for soft robot posture perception based on multi-source data fusion. The method includes: acquiring data using an inertial measurement unit (IMU) and flex sensors to obtain direct measurement values ​​and angle calibration values; processing the direct measurement values ​​using a drift function to obtain processed measurement values; obtaining a coefficient matrix using gradient descent based on the processed measurement values ​​and angle calibration values; and achieving posture perception of the soft robot end effector based on the coefficient matrix and multiple flex sensors. This invention provides an efficient and accurate method for soft robot posture perception.

Description

Technical Field

[0001] This invention relates to the field of soft robot posture perception technology, and in particular to a method and apparatus for soft robot posture perception based on multi-source data fusion. Background Technology

[0002] Soft robots are a rising hot topic and a frontier of future development in robotics technology. Traditional robots are mainly based on rigid structures, but these rigid structures prevent them from adapting to complex unstructured environments. This also results in drawbacks such as large size and low interaction safety. Compared with traditional rigid robots, soft robots, with their high adaptability, sensitivity, and agility, are constantly expanding the application fields of robotics and are one of the main trends in the future development of robotics. Due to their flexible materials, soft robots have large deformation and nonlinear characteristics, making it difficult to install or embed traditional rigid sensors, such as gyroscopes, accelerometers, and electronic compasses, into the body of soft robots. Therefore, there are currently few solutions that can directly and accurately perceive the deformation of soft robots themselves.

[0003] The rapid development of new materials has led to the development of sensors for sensing shape changes in soft robots, such as bending, elongation, and compression. These sensors primarily utilize conductive yarns, conductive rubber, carbon nanotubes, screen-printed liquid metal, and multi-ring optical tensile sensors. However, expanding the applications of these materials remains a major challenge. Optical fibers, waveguides, and fiber Bragg gratings offer another effective solution for detecting the posture of soft robots, but achieving precise sensing requires additional, expensive modems. Electromagnetic systems based on the Hall effect can also provide curvature information for soft robots, ensuring high-precision motion under a wide range of dynamic conditions. However, this approach requires protection from external electromagnetic sources, significantly limiting its application scenarios. While commercially available flex resistive sensors can avoid these issues and have received considerable attention in the field of humanoid robots, their drift characteristics and size limitations result in insufficient overall system sensing accuracy, severely restricting their widespread adoption in soft robotics.

[0004] In the existing technology, there is a lack of an efficient and accurate method for soft robot posture perception. Summary of the Invention

[0005] This invention provides a method and apparatus for posture perception of a soft robot based on multi-source data fusion. The technical solution is as follows:

[0006] On the one hand, a method for posture perception of soft robots based on multi-source data fusion is provided. This method is implemented by an electronic device and includes:

[0007] Based on the inertial measurement unit, data is acquired through the flex sensor to obtain direct measurement values ​​and angle calibration values;

[0008] Based on the direct measurement value, data processing is performed using a drift function to obtain the processed measurement value;

[0009] Based on the processed measured values ​​and the angle calibration values, a coefficient matrix is ​​obtained using the gradient descent method;

[0010] Based on the coefficient matrix, attitude perception is performed using the flex sensor.

[0011] The inertial measurement unit is used to provide end-effector posture samples of the soft robot. After the end-effector posture samples are obtained, the inertial measurement unit is removed. The inertial measurement unit is fixed to the end of the soft robot.

[0012] The flex sensor is used to sense the degree of bending of the end-effector posture sample of the soft robot; the flex sensor is attached to the outside of the soft robot.

[0013] Optionally, the step of processing the data using a drift function based on the direct measurement value to obtain the processed measurement value includes:

[0014] The direct measurement value is denoised to obtain the first processed value;

[0015] The first processed value is subjected to outlier removal to obtain the second processed value;

[0016] The signal-to-noise ratio of the second processed value is increased to obtain the third processed value;

[0017] Based on the third processed value, a drift function is used to process it to obtain the processed measurement value.

[0018] The drift function is a mathematical expression for the drift error of the flex sensor as a function of time.

[0019] Optionally, obtaining the coefficient matrix using gradient descent based on the processed measured values ​​and the angle calibration values ​​includes:

[0020] Based on the processed measured values ​​and the angle calibration values, data fitting is performed to obtain a fusion equation;

[0021] Based on the fusion equation, the loss function is obtained;

[0022] Based on the loss function, the step size and number of iterations are adjusted using gradient descent. When the loss function reaches its minimum, the coefficient matrix is ​​obtained.

[0023] The coefficient matrix includes a bending coefficient matrix and a scaling coefficient matrix;

[0024] The bending coefficient matrix is ​​used to fit the bending motion of the soft robot's end effector.

[0025] The scaling coefficient matrix is ​​used to fit the scaling motion of the soft robot's end effector.

[0026] On the other hand, a soft robot posture perception device based on multi-source data fusion is provided. This device is applied to a soft robot posture perception method based on multi-source data fusion. The device includes:

[0027] The data acquisition module is used to acquire data based on the inertial measurement unit and through the flex sensor to obtain direct measurement values ​​and angle calibration values;

[0028] The data processing module is used to process the data according to the direct measurement value through a drift function to obtain the processed measurement value;

[0029] The coefficient matrix acquisition module is used to obtain the coefficient matrix by gradient descent based on the processed measurement values ​​and the angle calibration values.

[0030] An attitude perception module is used to perform attitude perception based on the coefficient matrix and a flex sensor.

[0031] The inertial measurement unit is used to provide end-effector posture samples of the soft robot. After the end-effector posture samples are obtained, the inertial measurement unit is removed. The inertial measurement unit is fixed to the end of the soft robot.

[0032] The flex sensor is used to sense the degree of bending of the end-effector posture sample of the soft robot; the flex sensor is attached to the outside of the soft robot.

[0033] Optionally, the data processing module is further configured to:

[0034] The direct measurement value is denoised to obtain the first processed value;

[0035] The first processed value is subjected to outlier removal to obtain the second processed value;

[0036] The signal-to-noise ratio of the second processed value is increased to obtain the third processed value;

[0037] Based on the third processed value, a drift function is used to process it to obtain the processed measurement value.

[0038] The drift function is a mathematical expression for the drift error of the flex sensor as a function of time.

[0039] Optionally, the coefficient matrix acquisition module is further configured to:

[0040] Based on the processed measured values ​​and the angle calibration values, data fitting is performed to obtain a fusion equation;

[0041] Based on the fusion equation, the loss function is obtained;

[0042] Based on the loss function, the step size and number of iterations are adjusted using gradient descent. When the loss function reaches its minimum, the coefficient matrix is ​​obtained.

[0043] The coefficient matrix includes a bending coefficient matrix and a scaling coefficient matrix;

[0044] The bending coefficient matrix is ​​used to fit the bending motion of the soft robot's end effector.

[0045] The scaling coefficient matrix is ​​used to fit the scaling motion of the soft robot's end effector.

[0046] On the other hand, an electronic device is provided, comprising a processor and a memory, wherein the memory stores at least one instruction, which is loaded and executed by the processor to implement the aforementioned method for soft robot posture perception based on multi-source data fusion.

[0047] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction is stored in the storage medium, the at least one instruction being loaded and executed by a processor to implement the above-described method for soft robot posture perception based on multi-source data fusion.

[0048] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0049] This invention proposes a soft robot posture perception method based on multi-source data fusion. Targeting the drift characteristics of commercial flex sensors, it explores the mathematical relationship between drift characteristics and bending angles, improving the perception accuracy of the soft robot. Considering the bending and scaling states of the soft robot, and employing gradient descent to fit a fusion equation for the robot's end-effector posture, it further enhances the robot's self-perception capabilities and eliminates the influence of external electromagnetic interference. This invention is a highly efficient and accurate soft robot posture perception method. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 This is a flowchart of a soft robot posture perception method based on multi-source data fusion provided in an embodiment of the present invention;

[0052] Figure 2 This is a schematic diagram of a flex sensor distribution provided in an embodiment of the present invention;

[0053] Figure 3 This is a schematic diagram of the iterative relationship of a gradient descent process provided in an embodiment of the present invention;

[0054] Figure 4 This is a schematic diagram of a dual-modal soft robot provided in an embodiment of the present invention;

[0055] Figure 5 This is a block diagram of a soft robot posture perception device based on multi-source data fusion provided in an embodiment of the present invention;

[0056] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0057] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0058] This invention provides a method for posture perception of soft robots based on multi-source data fusion. This method can be implemented by an electronic device, which can be a terminal or a server. Figure 1 The flowchart shown is a method for posture perception of a soft robot based on multi-source data fusion. The processing flow of this method may include the following steps:

[0059] S1. Based on the inertial measurement unit, data is acquired through the flex sensor to obtain direct measurement values ​​and angle calibration values.

[0060] In one feasible implementation, the soft robot body of the present invention is jointly modulated from the compounds Hei-Cast8400 and 8400N. The four channels inside the arm are made by a coating process. The four channels are parallel and symmetrical in pairs and are used to arrange rope actuators. One side of the four cables is fixed to the end of the soft robot, and the other side is fixed to the synchronous pulley of the stepper motor.

[0061] The soft robot is driven by four stepper motors, controlled by a Raspberry Pi 4b, and its actuators are high-powered fishing lines that pass through four channels inside the soft robot and are fixed to its end.

[0062] The tension exerted by the stepper motor on the rope determines the degree of bending of the soft robot, which can be achieved by controlling the stepper motor's movement with pulses sent from the Raspberry Pi. When two adjacent actuators operate simultaneously, additional degrees of freedom can be created due to the difference in tension. Based on the inherent properties of the materials, the extension problem of the soft robot is negligible, and this extension problem can be avoided by the controller based on the additional degrees of freedom.

[0063] The inertial measurement unit (IMU) is used to provide end-effector posture samples of the soft robot. After the end-effector posture samples are obtained, the IMU is removed. The IMU is fixed to the end of the soft robot.

[0064] In one feasible implementation, the present invention employs an inertial measurement unit (IMU) from microelectromechanical systems (MEMS) technology to provide empirical data for multi-source fusion processing. The IMU is fixed to the end effector of the soft robot. Note that after acquiring a certain number of end effector posture samples, the IMU can be removed to simplify the robot's movement.

[0065] After selecting the sensor calibration values ​​and fusion values, the fusion equation can be obtained using the gradient descent method. This entire process greatly enhances the perception capabilities of the soft robot. On the one hand, it inherits the measurement accuracy of electromagnetic systems; on the other hand, it is freed from the constraints of electromagnetic disturbance environments.

[0066] The flex sensor is used to sense the degree of bending of the end-effector posture sample of the soft robot; the flex sensor is attached to the outside of the soft robot.

[0067] In one feasible implementation, the flex sensor, the sensing unit in this invention, is attached to the outside of the soft robot and communicates with the control unit via I2C. When the soft robot carries a large number of sensors, an ADC expansion board can be used to provide more interfaces. The flex sensors are distributed as follows: Figure 2 As shown.

[0068] S2. Based on the direct measurement value, the data is processed using a drift function to obtain the processed measurement value.

[0069] Optionally, based on the direct measurements, data processing is performed using a drift function to obtain processed measurements, including:

[0070] The direct measurement value is denoised to obtain the first processed value;

[0071] The first processed value is subjected to outlier removal to obtain the second processed value;

[0072] The signal-to-noise ratio of the second processed value is increased to obtain the third processed value;

[0073] Based on the third processed value, a drift function is used to process it, and the processed measurement value is obtained.

[0074] In one feasible implementation, the main function of data processing in this invention is to reduce noise and remove outliers from the state signal, thereby improving the signal-to-noise ratio. Direct readings from flex sensors typically contain a significant amount of noise and outliers. The noise is caused by interference from the sensor's internal manufacturing process and the external environment, while the outliers are due to quantization during analog-to-digital conversion. To improve the perception accuracy of the soft robot, the data processing section needs to correct the drift characteristics of the sensing unit.

[0075] The drift function is a mathematical expression for how the drift error of the flex sensor changes over time.

[0076] In one feasible implementation, the rational equation for the drift characteristics in this invention is shown in equation (1):

[0077]

[0078] The values ​​of the fitting parameters are: p1 = -1.657, p2 = -2.76, p3 = 11.2, p4 = 19.02, q1 = -2.256, q2 = -7.372, q3 = 19.13, q4 = 35.25t, where t is the system power-on time. Therefore, the mathematical expression for the relationship between the drift error of the sensing unit and time in this invention is shown in equation (2) below:

[0079]

[0080] S3. Based on the processed measured values ​​and angle calibration values, obtain the coefficient matrix using the gradient descent method.

[0081] In one feasible implementation, the present invention requires data from different sensors to be fused to obtain measurements that are closer to the true values. The small sample gradient descent method under distributed data fusion is selected to fit the end effector angular strain of the soft robot.

[0082] Optionally, based on the processed measured values ​​and angle calibration values, a coefficient matrix is ​​obtained using the gradient descent method, including:

[0083] The data is fitted based on the processed measured values ​​and angle calibration values ​​to obtain the fusion equation;

[0084] Based on the fusion equation, the loss function is obtained;

[0085] Based on the loss function, the step size and number of iterations are adjusted using gradient descent. When the loss function reaches its minimum, the coefficient matrix is ​​obtained.

[0086] In one feasible implementation, the gradient descent method of the present invention provides the following equation (3) for the fusion between distributed measurements and the true strain angle:

[0087] ξ f =k0+k1ξ m1 +k2ξ m2 +k3ξ m3 (3)

[0088] Wherein, the coefficient matrix K = [k0, k1, k2, k3] represents the weighting coefficients of each sensor; ξ m1 , ζ m2 ξ m3 These are the values ​​processed by the three sensors. The calibration value was measured by the IMU system, and the other acquired values ​​were obtained by the Flex sensor.

[0089] Furthermore, to obtain the values ​​of the coefficient matrix in the above equation, the loss function is defined as shown in equation (4) below:

[0090]

[0091] To obtain the magnitude of the weighting coefficients, partial derivatives are taken with respect to each coefficient in the equation as shown in equation (5):

[0092]

[0093] In this equation, the symbol ":=" represents the assignment operation, and κ represents the span value. During the solution process, the minimum value of the loss function can be approximated by adjusting the step size and the number of iterations, thus obtaining the optimal coefficient matrix K. The step size is set to α = 1.0 × 10⁻⁶. -4 The sample size was 500 points.

[0094] Figure 3 This diagram illustrates the fitting behavior of the gradient descent method when the number of iterations is only i = 100, where '*' represents the f value at each iteration. Loss The solid line represents the magnitude of the original data collected by the electromagnetic system when the end-effector bending angle of the soft robot ranges from 0° to 120°, and the dashed line represents the fitted value. For this case, the Pearson correlation coefficient r = 0.9963, and the energy value decreases to 6.9321. By comparing the relationships between different iteration numbers and the magnitudes of different Pearson correlation coefficients, this invention ultimately selects an iteration number of i = 10. 6 .

[0095] The coefficient matrix includes the curvature coefficient matrix and the scaling coefficient matrix.

[0096] The bending coefficient matrix is ​​used to fit the bending motion of the end effector of a soft robot;

[0097] The scaling coefficient matrix is ​​used to fit the scaling motion of the soft robot's end effector.

[0098] In one feasible implementation method, Figure 4 This is a schematic diagram of the dual-mode motion of the soft robot in this invention. The soft robot exhibits two motion modes during bending motion because the required tension value of the rope is completely different in these two states. The dual-mode motion of the soft robot includes two processes: bending and contraction. The bending process of the actuator is described as A→D, while the contraction process of the actuator is described as D→A.

[0099] At iteration number 10 6 The total time consumed was 1619.2336s and 1717.888s respectively. At this time, the loss function value f... Loss The values ​​are 1.2296 and 2.5685 respectively. The values ​​of the curvature coefficient matrix and scaling coefficient matrix are described as follows:

[0100] K1=[-6.7069 2.4788 -3.5063 6.6329], K2=[-3.8708 -9.8276 0.49994.4460]

[0101] S4. Based on the coefficient matrix, perform attitude perception using the flex sensor.

[0102] In one feasible implementation, the bimodal fusion equation of the soft robot can be obtained based on the bending coefficient matrix and the scaling coefficient matrix in the coefficient matrix. The bimodal fusion equation is shown in equation (6) below:

[0103]

[0104] Based on the processed data from the real-time flex sensor, the end-effector posture of the soft robot can be accurately and efficiently obtained through a dual-modal fusion equation.

[0105] This invention proposes a soft robot posture perception method based on multi-source data fusion. Targeting the drift characteristics of commercial flex sensors, it explores the mathematical relationship between drift characteristics and bending angles, improving the perception accuracy of the soft robot. Considering the bending and scaling states of the soft robot, and employing gradient descent to fit a fusion equation for the robot's end-effector posture, it further enhances the robot's self-perception capabilities and eliminates the influence of external electromagnetic interference. This invention is a highly efficient and accurate soft robot posture perception method.

[0106] Figure 5 This is a block diagram illustrating a soft robot posture perception device based on multi-source data fusion, according to an exemplary embodiment. (Refer to...) Figure 5 The device includes:

[0107] The data acquisition module 510 is used to acquire data based on the inertial measurement unit and through the flex sensor to obtain direct measurement values ​​and angle calibration values;

[0108] Data processing module 520 is used to process the data according to the direct measurement value through a drift function to obtain the processed measurement value;

[0109] The coefficient matrix acquisition module 530 is used to obtain the coefficient matrix by gradient descent based on the processed measurement value and the angle calibration value.

[0110] The attitude perception module 540 is used to perform attitude perception based on the coefficient matrix and the flex sensor.

[0111] The inertial measurement unit is used to provide end-effector posture samples of the soft robot. After the end-effector posture samples are obtained, the inertial measurement unit is removed. The inertial measurement unit is fixed to the end of the soft robot.

[0112] The flex sensor is used to sense the degree of bending of the end-effector posture sample of the soft robot; the flex sensor is attached to the outside of the soft robot.

[0113] Optionally, the data processing module 520 is further configured to:

[0114] The direct measurement value is denoised to obtain the first processed value;

[0115] The first processed value is subjected to outlier removal to obtain the second processed value;

[0116] The signal-to-noise ratio of the second processed value is increased to obtain the third processed value;

[0117] Based on the third processed value, a drift function is used to process it to obtain the processed measurement value.

[0118] The drift function is a mathematical expression for the drift error of the flex sensor as a function of time.

[0119] Optionally, the coefficient matrix acquisition module 530 is further configured to:

[0120] Based on the processed measured values ​​and the angle calibration values, data fitting is performed to obtain a fusion equation;

[0121] Based on the fusion equation, the loss function is obtained;

[0122] Based on the loss function, the step size and number of iterations are adjusted using gradient descent. When the loss function reaches its minimum, the coefficient matrix is ​​obtained.

[0123] The coefficient matrix includes a bending coefficient matrix and a scaling coefficient matrix;

[0124] The bending coefficient matrix is ​​used to fit the bending motion of the soft robot's end effector.

[0125] The scaling coefficient matrix is ​​used to fit the scaling motion of the soft robot's end effector.

[0126] This invention proposes a soft robot posture perception method based on multi-source data fusion. Targeting the drift characteristics of commercial flex sensors, it explores the mathematical relationship between drift characteristics and bending angles, improving the perception accuracy of the soft robot. Considering the bending and scaling states of the soft robot, and employing gradient descent to fit a fusion equation for the robot's end-effector posture, it further enhances the robot's self-perception capabilities and eliminates the influence of external electromagnetic interference. This invention is a highly efficient and accurate soft robot posture perception method.

[0127] Figure 6 This is a schematic diagram of the structure of an electronic device 600 provided in an embodiment of the present invention. The electronic device 600 may vary considerably due to different configurations or performance. It may include one or more central processing units (CPUs) 601 and one or more memories 602. The memory 602 stores at least one instruction, which is loaded and executed by the processor 601 to implement the steps of the above-mentioned soft robot posture perception method based on multi-source data fusion.

[0128] In an exemplary embodiment, a computer-readable storage medium is also provided, such as a memory including instructions that can be executed by a processor in a terminal to complete the aforementioned method for soft robot posture perception based on multi-source data fusion. For example, the computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, or optical data storage device.

[0129] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0130] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for posture perception of a soft robot based on multi-source data fusion, characterized in that, The method includes: Based on the inertial measurement unit, data is acquired through the flex sensor to obtain direct measurement values ​​and angle calibration values; Based on the direct measurement value, data processing is performed using a drift function to obtain the processed measurement value; Based on the processed measured values ​​and the angle calibration values, a coefficient matrix is ​​obtained using the gradient descent method; The coefficient matrix includes a bending coefficient matrix and a scaling coefficient matrix; The bending coefficient matrix is ​​used to fit the bending motion of the soft robot's end effector. The scaling coefficient matrix is ​​used to fit the scaling motion of the soft robot's end effector. Based on the coefficient matrix, attitude perception is performed using the flex sensor.

2. The method for soft robot posture perception based on multi-source data fusion according to claim 1, characterized in that, The inertial measurement unit is used to provide end-effector posture samples of the soft robot. After the end-effector posture samples are obtained, the inertial measurement unit is removed. The inertial measurement unit is fixed to the end of the soft robot.

3. The method for soft robot posture perception based on multi-source data fusion according to claim 1, characterized in that, The flex sensor is used to sense the degree of bending of the end-effector posture sample of the soft robot; the flex sensor is attached to the outside of the soft robot.

4. The method for soft robot posture perception based on multi-source data fusion according to claim 1, characterized in that, The step of processing the data using a drift function based on the direct measurement value to obtain the processed measurement value includes: The direct measurement value is denoised to obtain the first processed value; The first processed value is subjected to outlier removal to obtain the second processed value; The signal-to-noise ratio of the second processed value is increased to obtain the third processed value; Based on the third processed value, a drift function is used to process it to obtain the processed measurement value.

5. The method for soft robot posture perception based on multi-source data fusion according to claim 1, characterized in that, The drift function is a mathematical expression for how the drift error of the flex sensor changes over time.

6. The method for soft robot posture perception based on multi-source data fusion according to claim 1, characterized in that, The step of obtaining the coefficient matrix using gradient descent based on the processed measured values ​​and the angle calibration values ​​includes: Based on the processed measured values ​​and the angle calibration values, data fitting is performed to obtain a fusion equation; Based on the fusion equation, the loss function is obtained; Based on the loss function, the step size and number of iterations are adjusted using gradient descent. When the loss function reaches its minimum, the coefficient matrix is ​​obtained.

7. A soft robot posture perception device based on multi-source data fusion, characterized in that, The device includes: The data acquisition module is used to acquire data based on the inertial measurement unit and through the flex sensor to obtain direct measurement values ​​and angle calibration values; The data processing module is used to process the data according to the direct measurement value through a drift function to obtain the processed measurement value; The coefficient matrix acquisition module is used to obtain the coefficient matrix by gradient descent based on the processed measurement values ​​and the angle calibration values. The coefficient matrix includes a bending coefficient matrix and a scaling coefficient matrix; The bending coefficient matrix is ​​used to fit the bending motion of the soft robot's end effector. The scaling coefficient matrix is ​​used to fit the scaling motion of the soft robot's end effector. An attitude perception module is used to perform attitude perception based on the coefficient matrix and a flex sensor.

8. The soft robot posture perception device based on multi-source data fusion according to claim 7, characterized in that, The data processing module is further used for: The direct measurement value is denoised to obtain the first processed value; The first processed value is subjected to outlier removal to obtain the second processed value; The signal-to-noise ratio of the second processed value is increased to obtain the third processed value; Based on the third processed value, a drift function is used to process it to obtain the processed measurement value.

9. A soft robot posture perception device based on multi-source data fusion according to claim 7, characterized in that, The coefficient matrix acquisition module is further used for: Based on the processed measured values ​​and the angle calibration values, data fitting is performed to obtain a fusion equation; Based on the fusion equation, the loss function is obtained; Based on the loss function, the step size and number of iterations are adjusted using gradient descent. When the loss function reaches its minimum, the coefficient matrix is ​​obtained.

Images