Space manipulator impedance control method based on joint hysteresis dynamic position compensation

Abstract

The invention discloses a space manipulator impedance control method based on joint hysteresis dynamic position compensation, and belongs to the field of space robot nonlinear dynamics control. The control method comprises the following steps: constructing a flexible joint space manipulator model based on a standard dynamic model of the space manipulator; according to the flexible joint space manipulator model, a virtual torque sensing strategy is designed based on motor side sensor information, and a joint torsion torque estimated value is obtained; building a joint hysteresis model based on a Bi-LSTM mixed hysteresis modeling method fusing multi-scale memory and an attention mechanism, and training the joint hysteresis model; inputting a joint torsion moment estimation value into the trained joint hysteresis model, and estimating a joint torsion angle; based on the estimated joint torsion angle, the joint angle measured by the motor side encoder is subjected to superposition correction, and track error compensation caused by joint hysteresis is achieved.

Description

Technical Field

[0001] The present invention belongs to the field of nonlinear dynamic control of space robots, and in particular relates to an impedance control method for a space robot arm based on joint hysteresis dynamic position compensation. Background Art

[0002] When performing tasks, space manipulators inevitably come into contact with the external environment. While ensuring precise position control, contact force control is essential to prevent damage to the manipulator and external objects caused by excessive contact force, and to prevent failure of delicate operations such as capture and rotation due to insufficient contact force. Therefore, the manipulator needs to exhibit a certain degree of compliance when in contact with the environment. Impedance control is one of the important compliance control methods. Impedance control considers the relationship between the end displacement and the environmental contact force as a "spring-mass-damper" model. Through a reasonable control strategy, it adjusts the inertia, damping, and stiffness parameters to achieve dynamic adjustment of the relationship between the end position and the contact force, making the space manipulator compliant. This method is simple to implement, highly robust, and has a wide range of applications.

[0003] In recent years, with the development of new servo drive technologies, flexible servo-driven joints with harmonic reducers have become an important drive device for position and force control, as well as vibration suppression, in space robots. Harmonic drives offer advantages such as a large transmission ratio and smooth transmission. However, due to space and weight constraints, the flexibility of servo-driven joints in lightweight and compact space precision manipulation robots often cannot be ignored. Due to the unique structure of the harmonic reducer, combined with dynamic friction between the various components and nonlinear stiffness variations during servo joint operation, servo-driven joints exhibit complex, highly nonlinear hysteresis. From a mechanical perspective, joint nonlinear hysteresis is primarily due to the nonlinear constitutive relationship of the servo-driven joint material and the nonlinear mechanical behavior of the transmission structure. Hysteresis causes severe dynamic nonlinearity between the actuation torque and torsion angle of the driven joint. Unlike nonlinearities such as dead zone and saturation in the driven joint, nonlinear hysteresis generally exhibits multi-value mapping, memory, and dynamic rate dependence, resulting in reduced joint angle tracking accuracy and impedance control performance in space robots. Summary of the Invention

[0004] In view of the deficiencies in the prior art, the purpose of the present invention is to provide a spatial manipulator impedance control method based on joint hysteresis dynamic position compensation, which solves the problems in the prior art.

[0005] The purpose of the present invention can be achieved through the following technical solutions:

[0006] The impedance control method of a spatial manipulator based on joint hysteresis dynamic position compensation includes the following steps:

[0007] Based on the standard dynamic model of the space manipulator, a flexible joint space manipulator model is constructed;

[0008] According to the flexible joint space manipulator model, a virtual torque perception strategy is designed based on the motor-side sensor information to obtain the estimated value of the joint torsional torque;

[0009] A joint hysteresis model is constructed based on the Bi-LSTM hybrid hysteresis modeling method that integrates multi-scale memory and attention mechanism, and the joint hysteresis model is trained;

[0010] inputting the estimated value of the joint torsion torque into the trained joint hysteresis model to estimate the joint torsion angle;

[0011] Based on the estimated joint torsion angle, the joint angle measured by the motor side encoder is corrected by superposition to compensate for the trajectory error caused by joint hysteresis.

[0012] Furthermore, the flexible joint space manipulator model is:

[0013]

[0014] in, is the joint angle vector, Joint torsional torque, Determined by the inertia matrix of the robotic arm; is the Coriolis matrix; is the gravity matrix; n is the number of degrees of freedom of the manipulator, is the joint angular acceleration, is the joint angular velocity; is the diagonal matrix of motor drive inertia, is the joint angular acceleration on the motor side; The torque vector directly provided to the motor.

[0015] Furthermore, the joint torsional moment estimation value The calculation process is:

[0016]

[0017] r=L[∫(ur)dt-p]

[0018] Among them, r is the residual state vector, L=diag(L i ) is the error gain diagonal matrix, L i >0 is the design constant; u represents the motor drive torque vector; p=[p1,…p n ] is the generalized momentum vector, which represents the dynamic characteristics of the flexible joint. p i is the generalized momentum of the i-th motor, J mi is the moment of inertia of the i-th motor, is the angular velocity of the i-th motor.

[0019] Furthermore, in the joint hysteresis model: first, the joint torsion torque sequence is decomposed into short-term, medium-term and long-term subsequences through a multi-scale memory mechanism, and the medium-term and long-term subsequences are downsampled; then the short-term, medium-term and long-term subsequences are input into Bi-LSTM, and the bidirectional fusion output of the short-term, medium-term and long-term subsequences is output; then the attention mechanism is introduced to fuse multiple time series features, and the short-term, medium-term and long-term attention scores are calculated respectively. Then, the softmax function is used to calculate the bidirectional fusion output weights of the short-term, medium-term and long-term subsequences, and weighted calculation is performed. Finally, the output result of the joint hysteresis model is obtained through the output layer.

[0020] Furthermore, the calculation formulas for the short-term, medium-term and long-term attention scores are:

[0021]

[0022]

[0023]

[0024] Among them, v T 、W e 、b e is the learnable parameter of the attention mechanism; They are the bidirectional fusion outputs of short-term, medium-term, and long-term subsequences respectively; They are short-term, medium-term and long-term attention scores respectively;

[0025] The formula for calculating the bidirectional fusion output weight of short-term, medium-term, and long-term subsequences is:

[0026]

[0027]

[0028]

[0029] Where exp(·) is the exponential function, are the bidirectional fusion output weights of short-term, medium-term, and long-term subsequences respectively;

[0030] The output of the joint hysteresis model is:

[0031] Δq t =W0h t +b

[0032]

[0033] Among them, W0 is the weight of the output layer; ht is the weighted calculation result, b is the deviation; Δq t is the output of the joint hysteresis model at time t.

[0034] Furthermore, during the training of the joint hysteresis model, joint torsion torque and torsion angle data are collected and normalized; and the normalized data are then used to train the joint hysteresis model.

[0035] The loss function during the joint hysteresis model training process is:

[0036]

[0037] Among them, E t ' is the prediction error at time step t; Δθ t is the output angle of the robot's flexible joint at time t.

[0038] The impedance control system of a spatial manipulator based on joint hysteresis dynamic position compensation includes:

[0039] Robotic arm model construction module: Based on the standard dynamic model of the space robot arm, a flexible joint space robot arm model is constructed;

[0040] Torque estimation module: Based on the flexible joint space manipulator model, a virtual torque perception strategy is designed based on the motor-side sensor information to obtain the estimated value of the joint torsional torque;

[0041] Joint hysteresis model construction module: This module builds and trains a joint hysteresis model based on the Bi-LSTM hybrid hysteresis modeling method that integrates multi-scale memory and attention mechanisms.

[0042] Torsion angle estimation module: inputs the estimated value of joint torsion torque into the trained joint hysteresis model to estimate the joint torsion angle;

[0043] Compensation control module: Based on the estimated joint torsion angle, the joint angle measured by the motor side encoder is superimposed and corrected to achieve trajectory error compensation caused by joint hysteresis.

[0044] A computer storage medium stores a readable program. When the program is run, the program can instruct a computing device to execute the above-mentioned spatial manipulator impedance control method based on joint hysteresis dynamic position compensation.

[0045] An electronic device comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other via the communication bus;

[0046] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the above-mentioned spatial manipulator impedance control method based on joint hysteresis dynamic position compensation.

[0047] A computer program product comprising computer instructions for instructing a computing device to perform operations corresponding to the above-mentioned method for controlling the impedance of a spatial manipulator based on dynamic position compensation of joint hysteresis.

[0048] Beneficial effects of the present invention:

[0049] 1. This paper proposes a hybrid hysteresis modeling method that integrates multi-scale memory and an attention mechanism, Bi-LSTM, to establish a joint hysteresis model. The multi-scale memory mechanism decomposes joint torque into short-term, medium-term, and long-term subsequences, decoupling the dynamic time scale of hysteresis. The attention mechanism dynamically weights multi-scale features through learning parameters to obtain key information about the hysteresis process. The Bi-LSTM bidirectional grid structure captures temporal dependencies between the past and the future, improving the accuracy of modeling complex hysteresis behaviors. This solution overcomes the limitations of traditional modeling methods, which rely on single-scale, fixed-weight, and unidirectional modeling, and achieves a synergistic combination of time scale decoupling, feature weight adaptation, and bidirectional temporal modeling.

[0050] 2. The present invention designs a virtual torque sensing (VTS) scheme for joint torque based on the sensor information of the motor-side encoder to obtain joint torque information in real time; then, based on the obtained joint torque, the established joint hysteresis model is introduced to estimate the joint torsion angle in real time. Finally, the joint torsion angle is superimposed on the joint angle measured by the motor-side encoder, and the actual trajectory calculated by the joint motor-side sensor information is corrected to achieve trajectory error compensation caused by joint hysteresis. Therefore, the trajectory error caused by joint hysteresis can be compensated, and the impedance control performance of the flexible joint space manipulator can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0052] Figure 1 Schematic diagram of the LSTM unit structure of the present invention;

[0053] Figure 2 This is a flow chart of the impedance control of a spatial manipulator based on joint hysteresis dynamic position compensation according to the present invention. DETAILED DESCRIPTION

[0054] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0055] Example 1

[0056] like Figure 2 As shown, the impedance control method of a spatial manipulator based on joint hysteresis dynamic position compensation includes the following steps:

[0057] S1, based on the standard dynamic model of the space manipulator, to build a flexible joint space manipulator model;

[0058] The standard dynamic model of the space manipulator is:

[0059]

[0060] in, is the joint angle vector, The joint torsional torque is the torque vector of the flexible joint driving the load. The rigid body dynamics is given by the inertia matrix of the manipulator. Decide;

[0061] is the Coriolis matrix; is the gravity matrix; n is the number of degrees of freedom of the manipulator; is the joint angular acceleration; is the joint angular velocity.

[0062] Considering joint flexibility, each joint generates an additional degree of freedom, so the vector Represents the rotation angle of the motor side. The flexible joint space robot model is:

[0063]

[0064] in, is the diagonal matrix of motor drive inertia, is the joint angular acceleration on the motor side; is the torque vector directly provided by the motor. Due to the joint flexibility, the joint torsional torque τ will cause the joint torsional angle, and the two show a nonlinear hysteresis relationship, which is manifested as multi-value mapping, memory, and dynamic rate correlation. This leads to a decrease in the joint angle tracking accuracy of the space robot, which can be expressed as:

[0065] τ=f(Δq)

[0066]

[0067] Where Δq is the joint torsion angle vector; f(·) is the nonlinear hysteresis relationship between the joint torsion torque τ and Δq; is the joint transmission ratio matrix.

[0068] S2, based on the flexible joint space manipulator model, designs a virtual torque sensing (VTS) strategy based on the motor-side sensor information to obtain the estimated value of the joint torsional torque;

[0069] During the modeling and identification process, the joint torsional torque τ changes dynamically during operation due to nonlinear factors such as nonlinear friction, clearance, and dead zone, and is difficult to measure. In this embodiment, a strategy for estimating the joint torsional torque τ is proposed based on the dynamic model of motor drive and generalized momentum. The specifics are as follows:

[0070] Define the generalized momentum vector p=[p1,…p n ] represents the dynamic characteristics of the flexible joint, and its expression is Among them, p i is the generalized momentum of the i-th motor, J mi is the moment of inertia of the i-th motor, is the angular velocity of the i-th motor.

[0071] The dynamic equation of the flexible joint can be rewritten as:

[0072]

[0073] Where, is the rate of change of generalized momentum.

[0074] Generalized rate of change of momentum Further estimation is performed through the observer:

[0075]

[0076]

[0077] Where r is the momentum residual vector, which is proportional to the generalized momentum error vector; L = diag(L i ) is the error gain diagonal matrix; L i >0 is the design constant; is the estimated value of generalized momentum; is the estimated value of the rate of change of generalized momentum;

[0078] The momentum residual vector r can be expressed as:

[0079]

[0080] Among them, u=[u1…un ] represents the motor drive torque vector, which can be directly calculated through the motor current and torque constant; the motor speed is measured by the motor encoder.

[0081] Taking the derivative of the above formula, we can get the residual state dynamics equation:

[0082]

[0083] It can be seen from the above formula that the residual state vector r follows the joint torsional torque τ and has a linear exponentially stable dynamic characteristic. Therefore, the estimated value of the joint torsional torque is It can be expressed as:

[0084]

[0085] S3, constructs a joint hysteresis model based on the Bi-LSTM hybrid hysteresis modeling method that integrates multi-scale memory and attention mechanism, and trains the joint hysteresis model;

[0086] The joint torsional torque τ causes the joint torsional angle, and the two exhibit a nonlinear hysteresis relationship, exhibiting multi-value mapping, memory, and dynamic rate correlation. This complex characteristic with memory effects makes it difficult to accurately describe traditional feedforward neural networks and static models.

[0087] This embodiment proposes a bidirectional long short-term memory neural network (Bi-LSTM) hysteresis modeling method that integrates multi-scale memory and attention mechanism. The multi-scale memory mechanism is used to distinguish multiple physical time scales in the hysteresis phenomenon. The attention mechanism integrates the characteristics of different time scales. Multiple modules work together to achieve accurate characterization of the time dependence of complex hysteresis systems. The hysteresis effect of flexible joints has direction dependence (such as different loading / unloading paths) and historical memory (the current state is affected by the previous input sequence). The bidirectional long short-term memory network (Bi-LSTM) designed based on the LSTM unit structure solves this problem through the following characteristics: forward LSTM layer, capturing the temporal dependency from the past to the present moment; reverse LSTM layer: tracing back from the future to the present moment to enhance the modeling of the hysteresis loop closure characteristics.

[0088] 1. LSTM unit

[0089] LSTM (Long Short-Term Memory) is an ideal choice for solving the problem of time-dependent hysteresis characteristics of flexible joints due to its unique loop structure and gating mechanism. The core of LSTM is the memory unit (C t ) and three gating mechanisms: input gate i t 、Forget Gate t , output gate o t , its structure is as follows Figure 1 shown.

[0090] Let x t represents the input at time t (flexible joint torque), h t represents the hidden layer output at time t, C t Represents the state of the memory unit at time t.

[0091] Forget gate (f t ): Determine the memory unit c at the previous moment t-1 How much information is retained? The output vector of the forget gate at time t is:

[0092] f t =σ(W f,h h t-1 +W f,x x t +b f )

[0093] Among them, W f,h , W f,x represents the weight of the forget gate function matrix; b f represents the bias term of the forget gate; h t-1 Represents the output of the hidden layer at time t-1.

[0094] Input gate (i t ): Controls the ratio of the current candidate memory to the memory unit. The input gate has two parts:

[0095] i t =σ(W i,h h t-1 +W i,x x t +b i )

[0096]

[0097] Where: is a candidate memory unit; W i,h , W c,h , W i,x , W c,x Represents different weight parameters of the input gate function matrix; b i and b c Represents different bias terms for the input gate.

[0098] The expression of the memory unit at time t is:

[0099]

[0100] Where ⊙ is the element-wise multiplication (Hadamard product).

[0101] Output gate (o t ): Determine memory unit Ct How much information is output to h t , the output expression of the output gate at time t is:

[0102] o t =σ(W o,h h t-1 +W o,x x t +b o )

[0103] h t =o t ⊙tanh(C t )

[0104] Among them, W o,h , W o,x Represents the weight of the output gate function matrix; b o Represents the bias term of the forget gate.

[0105] In the above gate operation, σ is the Sigmoid activation function (output 0 to 1, controlling the information passing ratio), and the calculation formula is:

[0106]

[0107] Tanh is the hyperbolic tangent function (normalized to -1 to 1), and the calculation formula is:

[0108]

[0109] 2. The construction process of the hysteresis model is as follows:

[0110] While traditional LSTM networks have a certain degree of memory capacity, they still suffer from insufficient precision and inadequate capture of long-term dependencies when dealing with the complex hysteresis characteristics of flexible joints in industrial robots. To better describe the hysteresis characteristics of flexible joints, this paper proposes a bidirectional LSTM hysteresis model based on multi-temporal memory and attention mechanisms.

[0111] (1) Data collection and preprocessing

[0112] Collect the torque τ of the flexible joint at different speeds t and angle Δθ t The data is normalized and the input and output are mapped to the interval [0, 1] to avoid the influence of numerical scale differences on training.

[0113] The multi-scale memory mechanism is introduced to decompose the torsional torque sequence of the robot's flexible joint into short-term (time window T S =50), mid-term (time window T M =100), long-term subsequence (time window TL =200), frequency separation is achieved by downsampling.

[0114] Short-term subsequence: No downsampling is performed.

[0115] Mid-term subsequence: Downsampling rate K M =2, uniform sampling.

[0116] Long-term subsequence: Downsampling rate K L =4, uniform sampling.

[0117] (2) Network structure design

[0118] Input layer: subsequences of torsional moments of flexible joints at different scales τ S , τ M , τ L .

[0119] Hidden layer: Bidirectional LSTM layer (core structure), used to learn the dynamic memory characteristics of hysteresis.

[0120] Short-term subsequence forward hidden layer output:

[0121] Backward hidden layer output:

[0122] Bidirectional fusion output:

[0123] in, and These are the calculation processes of the forward and reverse LSTM units respectively; and Represent the hidden layer states of the forward and reverse LSTM at the previous time step respectively; Concat is a vector concatenation operation.

[0124] The mid-term subsequence and long-term subsequence can be obtained similarly, and the bidirectional fusion outputs are: and

[0125] The attention mechanism is introduced to integrate multiple temporal features. The calculation formulas for short-term, medium-term and long-term attention scores are as follows:

[0126]

[0127]

[0128]

[0129] Among them, v T 、We 、b e is the learnable parameter of the attention mechanism, which is updated simultaneously with the LSTM weight parameters through back propagation; They are the bidirectional fusion outputs of short-term, medium-term, and long-term subsequences respectively; They are short-term, medium-term and long-term attention scores respectively.

[0130] The weights are calculated using the softmax function:

[0131]

[0132]

[0133]

[0134] Where exp(·) is an exponential function. are the bidirectional fusion output weights of short-term, medium-term, and long-term subsequences respectively;

[0135] The weighted result h is output by multiple temporal features based on the attention mechanism t :

[0136]

[0137] Final output layer:

[0138] Δq=(Δq t-T ,Δq t-T+1 ,···Δq t )

[0139] Δq t =W0h t +b

[0140] Where T is the length of the time series; W0 is the weight of the output layer; b is the bias; Δq t is the output of the joint hysteresis model at time t.

[0141] 3. Training strategy design

[0142] Define the loss function (mean square error):

[0143]

[0144] Among them, E t ' is the prediction error at time step t; Δθ t is the output angle of the robot's flexible joint at time t. Using the gradient descent method, update the output layer weight W o :

[0145] ΔW o =γ(Δθ t-Δq t )h t

[0146] Among them, γ is the learning rate.

[0147] Attention core parameter v T , W e , b e The parameter gradient calculation formula is as follows:

[0148]

[0149]

[0150]

[0151] Among them, Q∈{S,M,L} are different time scales.

[0152] Joint hysteresis model input gate weight update:

[0153]

[0154]

[0155]

[0156] Among them, W i (k+1) and W i (k) are the input gate weight matrices for the k+1th and kth iterations respectively; is the loss function L versus W in the kth iteration i gradient; and is the bias term of the input gate at the k+1th and kth iterations; is the loss function L for the bias term b in the kth iteration i gradient; is the error term of the input gate at time step t.

[0157] Joint hysteresis model forget gate weight update:

[0158]

[0159]

[0160]

[0161] Among them, W f (k+1) and are the forget gate weight matrices for the k+1th and kth iterations respectively; is the loss function L versus W in the kth iteration f gradient; and is the bias term of the forget gate of the k+1th and kth iterations; is the loss function L for the bias term b in the kth iteration f gradient; is the error term of the forget gate at time step t.

[0162] Joint hysteresis model output gate weight update:

[0163]

[0164]

[0165]

[0166] Among them, W o (k+1) and are the output gate weight matrices for the k+1th and kth iterations respectively;

[0167] is the loss function L versus W in the kth iteration o gradient; and is the bias term for the output gate of the k+1th and kth iterations; is the loss function L for the bias term b in the kth iteration o gradient; is the error term of the output gate at time step t.

[0168] Joint hysteresis model candidate memory unit weight update:

[0169]

[0170]

[0171]

[0172] Among them, W c (k+1) and W c (k) are the candidate memory unit weight matrices for the k+1th and kth iterations respectively; is the loss function L versus W in the kth iteration c gradient; and is the candidate memory unit bias term for the k+1th and kth iterations; is the loss function L on b in the kth iterationc The gradient of δ c (t) is the error term of the candidate memory unit at time step t.

[0173] Gradient error calculation:

[0174]

[0175] in, is the gradient contribution of the attention mechanism to the LSTM, and is calculated as follows:

[0176]

[0177] Weight update of joint hysteresis model:

[0178]

[0179] Where W is the weight matrix to be updated; k is the number of iterations; is the gradient of the loss function with respect to the weight matrix.

[0180] S4, inputting the estimated value of the joint torsion torque into the trained joint hysteresis model to estimate the joint torsion angle;

[0181] S5, based on the estimated joint torsion angle, performs superimposed correction on the joint angle measured by the motor side encoder to compensate for the trajectory error caused by joint hysteresis.

[0182] Specifically, the specific steps for trajectory error compensation based on the joint torsion angle include:

[0183] In step S51, due to the hysteresis effect of the robot arm joint, the robot arm end position, velocity, and acceleration calculated from the robot arm joint angle are not the actual values. The estimated joint torsion angle obtained in step S4 is superimposed and corrected on the joint angle measured by the motor side encoder to achieve trajectory error compensation caused by joint hysteresis:

[0184] q a =q+Δq

[0185] Among them, q a is the corrected joint angle vector, q is the joint angle vector measured by the encoder on the joint motor side, and Δq is the torsion angle vector predicted by the joint hysteresis model.

[0186] S52, through the robot arm kinematics model and Jacobian matrix, the corrected joint angle vector Calculate the end position X of the robot arm:

[0187] X=K(q a )q a

[0188] where K(·) represents the forward kinematics of the robot; the end-effector velocity which can be calculated by joint angular velocity and Jacobian matrix:

[0189]

[0190] where J is the Jacobian matrix of the robot; is the corrected joint angular velocity vector.

[0191] Taking the derivative of the velocity, the acceleration can be obtained

[0192]

[0193] S6, the impedance controller calculates the reference force of the end-effector of the robot according to the designed motion trajectory, the actual motion trajectory and the impedance model parameters, to overcome the environmental contact force. The designed contact force is tracked by the actual environmental force through the internal torque controller. Generally, the mathematical model of the impedance relationship between the end-effector of the spatial robot and the environment can be expressed in the form of a second-order differential equation:

[0194]

[0195] where X, and are the actual displacement, velocity and acceleration of the end-effector, respectively; X r , and are the reference displacement, velocity and acceleration of the end-effector, respectively; M p , B p and K p are the inertia matrix, damping matrix and stiffness matrix of the impedance model, respectively; F p is the output torque of the end-effector.

[0196] S7, the control torque of the joint of the robot can be calculated through the impedance model. Assuming that the external environmental contact force is F e , the difference between the calculated control force F p and F e is mapped to the joint space through the Jacobian matrix:

[0197] τ=J T (F p -F e )

[0198] Based on the similar inventive concept, the embodiment of the present application also provides a computer storage medium, which stores a readable program, when the program is run by a processor, the program can execute the above-mentioned impedance control method for the spatial robot based on joint hysteresis dynamic position compensation.

[0199] Based on similar inventive concepts, an embodiment of the present invention provides an electronic device, comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other via the communication bus;

[0200] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the above-mentioned spatial manipulator impedance control method based on joint hysteresis dynamic position compensation.

[0201] Based on similar inventive concepts, an embodiment of the present invention further provides a computer program product, including computer instructions, which instruct a computing device to execute operations corresponding to the above-mentioned spatial manipulator impedance control method based on joint hysteresis dynamic position compensation.

[0202] Example 2

[0203] Based on the spatial manipulator impedance control method based on joint hysteresis dynamic position compensation proposed in Example 1, in this embodiment, a spatial manipulator impedance control system based on joint hysteresis dynamic position compensation is proposed, specifically including:

[0204] Robotic arm model construction module: Based on the standard dynamic model of the space robot arm, a flexible joint space robot arm model is constructed;

[0205] Torque estimation module: Based on the flexible joint space manipulator model, a virtual torque perception strategy is designed based on the motor-side sensor information to obtain the estimated value of the joint torsional torque;

[0206] Joint hysteresis model construction module: This module builds and trains a joint hysteresis model based on the Bi-LSTM hybrid hysteresis modeling method that integrates multi-scale memory and attention mechanisms.

[0207] Torsion angle estimation module: inputs the estimated value of joint torsion torque into the trained joint hysteresis model to estimate the joint torsion angle;

[0208] Compensation control module: Based on the estimated joint torsion angle, the joint angle measured by the motor side encoder is superimposed and corrected to achieve trajectory error compensation caused by joint hysteresis.

[0209] The method of the present invention can be implemented in hardware, firmware, or as software or computer code that can be stored in a recording medium (such as a CDROM, RAM, floppy disk, hard disk or magneto-optical disk), or as computer code that is originally stored in a remote recording medium or a non-temporary machine-readable medium downloaded over a network and will be stored in a local recording medium, so that the method described herein can be stored in such software processing on a recording medium using a general-purpose computer, a special-purpose processor or programmable or special-purpose hardware (such as an ASIC or FPGA). It will be understood that a computer, a processor, a microprocessor controller or programmable hardware includes a storage component (e.g., RAM, ROM, flash memory, etc.) that can store or receive software or computer code, and when the software or computer code is accessed and executed by a computer, a processor or hardware, the method described herein is implemented. In addition, when a general-purpose computer accesses the code for implementing the method shown here, the execution of the code converts the general-purpose computer into a special-purpose computer for executing the method shown here.

[0210] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention, and such changes and modifications fall within the scope of the invention as claimed.

Claims

1. A spatial manipulator impedance control method based on joint hysteresis dynamic position compensation is characterized in that: The following steps are involved: Based on the standard dynamic model of the space manipulator, a flexible joint space manipulator model is constructed; According to the flexible joint space manipulator model, a virtual torque perception strategy is designed based on the motor-side sensor information to obtain the estimated value of the joint torsional torque; A joint hysteresis model is constructed based on the Bi-LSTM hybrid hysteresis modeling method that integrates multi-scale memory and attention mechanism, and the joint hysteresis model is trained; inputting the estimated value of the joint torsion torque into the trained joint hysteresis model to estimate the joint torsion angle; Based on the estimated joint torsion angle, the joint angle measured by the motor side encoder is corrected by superposition to compensate for the trajectory error caused by joint hysteresis.

2. The spatial manipulator impedance control method based on joint hysteresis dynamic position compensation according to claim 1 is characterized in that: The flexible joint space manipulator model is: in, is the joint angle vector, Joint torsional torque, Determined by the inertia matrix of the robotic arm; is the Coriolis matrix; is the gravity matrix; n is the number of degrees of freedom of the manipulator, is the joint angular acceleration, is the joint angular velocity; is the diagonal matrix of motor drive inertia, is the joint angular acceleration on the motor side; The torque vector directly provided to the motor.

3. The spatial manipulator impedance control method based on joint hysteresis dynamic position compensation according to claim 2, characterized in that: The estimated value of the joint torsional torque The calculation process is: r=L[∫(ur)dt-p] Among them, r is the residual state vector, L=diag(L i ) is the error gain diagonal matrix, L i >0 is the design constant; u represents the motor drive torque vector; p=[p1,…p n ] is the generalized momentum vector, which represents the dynamic characteristics of the flexible joint. p i is the generalized momentum of the i-th motor, J mi is the moment of inertia of the i-th motor, is the angular velocity of the i-th motor.

4. The spatial manipulator impedance control method based on joint hysteresis dynamic position compensation according to claim 1, characterized in that: In the joint hysteresis model, the joint torsional torque sequence is first decomposed into short-term, medium-term, and long-term subsequences through a multi-scale memory mechanism, and the medium-term and long-term subsequences are downsampled. The short-term, medium-term, and long-term subsequences are then input into a Bi-LSTM, which outputs a bidirectional fusion output of the short-term, medium-term, and long-term subsequences. Then, the attention mechanism is introduced to fuse multiple time series features, and the short-term, medium-term and long-term attention scores are calculated respectively. The softmax function is then used to calculate the bidirectional fusion output weights of the short-term, medium-term and long-term subsequences, and weighted calculation is performed. Finally, the output result of the joint hysteresis model is obtained through the output layer.

5. The spatial manipulator impedance control method based on joint hysteresis dynamic position compensation according to claim 4 is characterized in that: The calculation formulas for the short-term, medium-term and long-term attention scores are: Among them, v T 、W e 、b e is the learnable parameter of the attention mechanism; They are the bidirectional fusion outputs of short-term, medium-term and long-term subsequences respectively; They are short-term, medium-term and long-term attention scores respectively; The formula for calculating the bidirectional fusion output weight of short-term, medium-term, and long-term subsequences is: Where exp(·) is the exponential function, are the bidirectional fusion output weights of short-term, medium-term, and long-term subsequences respectively; The output of the joint hysteresis model is: Δq t =W0h t +b Among them, W0 is the weight of the output layer; h t is the weighted calculation result, b is the deviation; Δq t is the output of the joint hysteresis model at time t.

6. The spatial manipulator impedance control method based on joint hysteresis dynamic position compensation according to claim 5, characterized in that: During the training of the joint hysteresis model, joint torsion torque and torsion angle data are collected and normalized; and the joint hysteresis model is trained using the normalized data. The loss function during the joint hysteresis model training process is: Among them, E t ' is the prediction error at time step t; Δθ t is the output angle of the robot's flexible joint at time t.

7. Spatial manipulator impedance control system based on joint hysteresis dynamic position compensation, characterized in that: include: Robotic arm model construction module: Based on the standard dynamic model of the space robot arm, a flexible joint space robot arm model is constructed; Torque estimation module: Based on the flexible joint space manipulator model, a virtual torque perception strategy is designed based on the motor-side sensor information to obtain the estimated value of the joint torsional torque; Joint hysteresis model construction module: This module builds and trains a joint hysteresis model based on the Bi-LSTM hybrid hysteresis modeling method that integrates multi-scale memory and attention mechanisms. Torsion angle estimation module: inputs the estimated value of joint torsion torque into the trained joint hysteresis model to estimate the joint torsion angle; Compensation control module: Based on the estimated joint torsion angle, the joint angle measured by the motor side encoder is superimposed and corrected to achieve trajectory error compensation caused by joint hysteresis.

8. A computer storage medium storing a readable program, characterized in that: When the program is run, the program can instruct the computing device to execute the spatial manipulator impedance control method based on joint hysteresis dynamic position compensation as described in any one of claims 1 to 6.

9. An electronic device, characterized in that: include: A processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the spatial manipulator impedance control method based on joint hysteresis dynamic position compensation as described in any one of claims 1 to 6.

10. A computer program product comprising computer instructions, characterized in that The computer instructions instruct the computing device to execute operations corresponding to the spatial manipulator impedance control method based on joint hysteresis dynamic position compensation as described in any one of claims 1-6.

Images