Abstract
Hindawi Journal of Robotics Volume 2022, Article ID 4231117, 10 pages https://doi.org/10.1155/2022/4231117 Research Article Parameter-Tunable RBF Neural Network Control Facing Dual-Joint Manipulators Weiying Xu HBUT Detroit Green Technology Institute, Hubei University of Technology, Wuhan 430000, China Correspondence should be addressed to Weiying Xu; 2011611126@hbut.edu.cn Received 8 July 2022; Revised 27 July 2022; Accepted 2 August 2022; Published 8 September 2022 Academic Editor: Shahid Hussain Copyright © 2022 Weiying Xu. �is is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. In order to improve the parameter control eƒect of the double-joint manipulator, this paper combines the RBF neural network to control the parameters of the double-joint manipulator and the command Šltering backstep impedance control method based on the RBF neural network is eƒectively applied to the multijoint manipulator. Moreover, this paper introduces the Šlter error compensation mechanism into the controller design to eliminate the inŒuence caused by the Šlter error. Finally, the eƒectiveness and superiority of the command Šltering backstep impedance control scheme of the multijoint manipulator adaptive neural network designed in this paper is veriŠed by simulation experiments. �e experimental research results verify that the parameter- tunable RBF neural network control method facing the dual-joint manipulator has a certain eƒect on the parameter control process of the dual-joint manipulator and can eƒectively improve the motion accuracy of the dual-joint manipulator. [4]. �e hypothetical mode method usually uses Šnite-di- 1. Introduction mensional continuous modes of several orders to represent inŠnite-dimensional beams. Moreover, each order mode is �e theoretical modeling methods of robotic arms are generally based on the principle of energy conservation. For usually obtained by the product of the modal coordinates a rigid body, its energy usually includes two parts: kinetic and the modal function, and Šnally, the consecutive several energy due to motion inertia and potential energy due to order modes are added to obtain the vibration amount at any gravitational Šeld. However, for a Œexible arm, its potential point on the entire Œexible arm. Literature [5] has dem- energy and kinetic energy are diƒerent from those of a rigid onstrated that the use of the second-order mode can ac- manipulator. Moreover, the Œexible arm part generally in- curately reŒect the vibration oƒset of the Œexible body. �is cludes three deformations, namely, rotational deformation, research method has also been well veriŠed experimentally lateral deformation, and axial deformation [1]. In order to and theoretically; however, this research method has certain consider the bending deformation of beams, there are deŠciencies and cannot be applied to Œexible manipulators currently four basic modeling theories of beams [2]. Since with variable cross-sectional areas [6]. In response to this the axial direction of the beam is generally longer than its situation, some scholars have proposed the Šnite element diameter, the Euler–Bernoulli beam theory is generally used method [7], and it has been shown that FE can reŒect the in many models. At the same time, since the Œexible ma- actual situation well. For the lumped mass method, it is the nipulator has inŠnite degrees of freedom, this is di”cult to simplest approximation to consider. It simply treats the achieve in modeling. �e general solution is to replace the Œexible manipulator as a spring and mass system, but this inŠnite-dimensional degrees of freedom with Šnite-di- research method is rarely used. mensional degrees of freedom and use Hamilton’s principle Traditional modeling of Œexible arms generally chooses or Lagrange dynamic equation to discretize a set of diƒer- classical deformations to derive the Lagrangian dynamic ence equations and Šnally solve them [3]. �ere are generally equations of Œexible arms. However, in some cases, this classical deformation does not conform to the actual three current modeling methods for Œexible manipulators 2 Journal of Robotics basis for the study of the manipulator system, the dynamic situation and cannot accurately describe the dynamic be- havior of the flexible arm. For example, when the flexible modeling of the manipulator provides a dynamic model for the subsequent high-precision motion control of the ma- arm rotates at a high speed, the comprehensive deforma- tion theory can more accurately describe the deformation nipulator end effector by making reasonable simplified form of the flexible arm [8]. So far, in describing the elastic assumptions for the actual manipulator system [16]. (e deformation of flexible arms, there are mainly three de- kinematics analysis of the manipulator system is to es- formation descriptions: classical deformation theory, tablish the relationship between the manipulator joint quadratic deformation theory, and comprehensive defor- variables and the end effector pose matrix by analyzing the mation theory. Classical deformation theory is currently two parts, including forward kinematics and inverse ki- nematics [17]. (e positive kinematics issue is to obtain the the most widely used method [9]. Literature [10] uses the classical deformation theory to model the single-joint pose matrix of the end effector by using the transformation matrix of the joint variables of the manipulator and in- flexible manipulator and considers the influence of non- linear centrifugal force on the accuracy of the model. versely solving the value of each joint variable of the manipulator through various transformations. (e inverse Literature [11] uses the classical deformation theory to model a single-joint flexible manipulator and designs a kinematics problem of the manipulator is the basis of the controller based on this model, and the control effect is trajectory planning and trajectory tracking control of the ideal. Literature [12] uses the classical deformation theory joint end. However, in the process of inverse solution of the for modeling and then conducts positive feedback exper- manipulator, multiple sets of solutions are often obtained. iments and negative feedback experiments, respectively, (erefore, it is necessary to select a set of suitable solutions and finally compares and analyzes the results of theoretical according to the working conditions of the specific ma- nipulator. Trajectory planning is another important di- solutions and experiments. Analysis shows that positive feedback experiments are more robust than negative rection of manipulator research. It is based on kinematics and dynamics analysis, according to the requirements of feedback in suppressing vibration. In addition, the end of the flexible manipulator contains a load to reduce the specific tasks, after a set of path points passed by the end effector of the manipulator is known, through the me- natural vibration frequency of the flexible manipulator. Literature [13] established the dynamic model of the chanical arm. (e arm inverse solution is used to obtain the Euler–Bernoulli beam using classical deformation theory corresponding joint angle of each joint. Among them, the and gave the numerical simulation results and concluded commonly used trajectory planning methods include cubic that the established dynamic model can be used as the polynomial interpolation function, quintic polynomial control equation for the design and simulation of the interpolation function, and spline interpolation function control system. Literature [14] established a Lagrangian [18]. In practical engineering applications, in order to ensure the smooth operation of each joint of the manip- dynamic model of the flexible arm based on the assumed modal method and Lagrange equation through classical ulator and to prevent the system from vibrating, the tra- jectory generated by the trajectory planning of the deformation theory and analyzed the physical and struc- tural parameters and external driving torque of the flexible manipulator must be smooth and continuous and should manipulator. It is concluded that these parameters will have have no sudden changes. (e application of robotic arms in an impact on the dynamic characteristics of the flexible various industries is more extensive. It is not just a simple manipulator. Literature [15] studied the theoretical mod- replacement for human work. In practical engineering, it eling problem of two flexible manipulators in the plane and puts forward higher requirements for its work execution designed a controller based on the classical theoretical efficiency and the control accuracy of end effectors. modeling problem. Literature [13] used the classical de- However, because the manipulator is a nonlinear, multi- formation method to study the kinematics of rigid-flexible input, multioutput system, when the system model is established, there are uncertain factors such as friction coupled double rods in the plane and concluded that if the natural frequency of the first-order system of the flexible between joints, coupling, measurement error, and external disturbance. (erefore, reducing the influence of model beam is lower than that of the natural frequency of the rigid beam, the frequency difference will increase with the in- uncertainty in the robotic arm system on the control ac- crease in frequency. It increases with the increase in the curacy of the end effector has always been a hot research natural frequency of the system; under the premise of topic in the field of control at home and abroad [19]. In certain initial conditions, the second-order natural fre- recent years, with the development of the intelligent control quency of the system will also increase with the increase in theory, some advanced control theories have also been the natural frequency of the whole system, and it shows that proposed. Although these advanced control theories can the motion and deformation process of the manipulator improve the tracking accuracy of the system to a certain will coordinate with each other when doing a large range of extent, there are still some problems to be solved. (is paper combines the RBF neural network to control motion. (ere are mutual coupling characteristics. When studying the control of the manipulator system, the parameters of the double-joint manipulator, improve the control effect of the two-joint manipulator, and promote the it is first necessary to study the dynamics modeling and kinematics analysis of the manipulator. As a theoretical work efficiency of the manipulator. Journal of Robotics 3 − 1 2. Command Filtering Backstep Impedance x _ � D x u − τ − C x , x x + G x . (2) 2 1 e 1 2 2 1 Control of the Adaptive Neural Network for (e command filter used in this chapter can be seen in the Dual-Joint Manipulator formula (2). In order to improve the control accuracy of the dual-joint manipulator system, the compensation signal will 2.1. Command Filtering Backstep Impedance Control of be constructed to compensate for the filtering error gen- Adaptive Neural Network. In recent years, to enable robots erated by the command filter. (e compensation signal to handle increasingly complex tasks, many robots have design is as follows: installed dual-joint robotic arms with higher dexterity. When the robot provides services, the double-joint robotic ⎧ ⎪ ξ � −K ξ + ξ + x − α , ⎨ 1 1 1 2 1,c arm will physically interact with the unknown external (3) environment, so the compliance and safety of the double- ξ � 0. joint robotic arm should be fully ensured to ensure the safety of physical interaction. To adjust the physical interaction Among them, the control gain matrix is K � K > 0 and 1 1 force between the two-joint manipulator and the external x is the output signal vector of the filter. When t � 0, ξ and environment, the impedance controller of the two-joint ξ are zero vectors. manipulator has been widely used in the interactive control (e error variable is defined, where x is the tracking of the robot. trajectory of the dual-joint manipulator through the im- At present, there have been many research achievements pedance relationship. (e design steps of the controller are in the backstepping impedance controller of the double- as follows: joint manipulator. However, the “computational com- Step 1: the algorithm constructs the Lyapunov function plexity” and “singularity” problems that may occur in the as V � (1/2)v v and derives V to get 1 1 1 design process of the traditional backstepping method and T T T _ _ the uncertain nonlinear terms in the dual-joint manipulator _ _ _ _ (4) V � v v � v z − ξ � v x − x − ξ . 1 1 1 1 1 1 1 2 r 1 system due to the difficulty of obtaining the model pa- rameters accurately deserve attention. In order to better (e virtual control law α is designed as solve these problems and make the dual-joint manipulator α � −Kz + x _ . (5) achieve better impedance control performance, this chapter 1 r introduces the adaptive neural network control technology Among them, the control gain matrix is K � K > 0. into the design of the command filtering backstepping 1 1 impedance controller. Moreover, this paper constructs an By substituting formulas (3) and (5) into formula (4), adaptive neural network command filtering backstep im- we get pedance controller for the dual-joint manipulator so that T T T V � v z + x + α − α − x _ − ξ � −v K v + v v . the uncertain dual-joint manipulator system can better 1 1 2 1,c r 1 1 1 1 1 2 realize the physical interaction control with the unknown (6) external environment. Compared with the existing dual- joint manipulator control scheme, this control scheme has Step 2: next, the algorithm selects the Lyapunov the following advantages: function as follows: (e dynamic equation of the double-joint manipulator system can be described as follows: (7) V � V + v D x v . 2 1 2 1 2 € _ _ D(x)x + C(x, x)x + G(x) � u − τ . (1) (e time derivative of the abovementioned formula is We define x � x and x � x _, and the model of the 1 2 obtained as follows: double-joint robotic arm system can be rewritten in the following form: T T _ _ V � V + v D x v + v D x v , 2 1 2 1 2 2 1 2 (8) u − τ − C x , x x + v e 1 2 2 1 T⎜ ⎟ T T ⎛ ⎜ ⎞ ⎟ ⎜ ⎟ _ ⎝ ⎠ � v − v K v + v D x v . 2 1 1 1 2 1 2 −G x − D x x + ξ 1 1 1,c 2 4 Journal of Robotics Substituting formulas (8) to (10) and (14) into the de- Since there are uncertain nonlinear terms in D(x ), rivative of formula (15), we get C(x , x ) and G(x ), adaptive neural network techniques T T 1 2 1 T T T T _ V � −v K v − v K v + v W S Z x _ + v W S Z x _ 1 1 1 2 2 2 2 D D D 1,c 2 C C C 1,c are used to approximate the matrix terms D(x ), C(x , x ), 1 1 2 T T T and G(x ). +v W S Z + v E − K sign v G G r r 2 1 2 G 2 (e real control law u for designing a dual-joint robotic n n T T T − σ W W − W S Z x _ v arm system is DK Dk Dk Dk DK D 1,c 2k k�1 k�1 T T n n n u � −K z + τ + W S Z x + W S Z x T T T T T 2 2 e D D 1,c C C 1,c D C − W S Z x _ v − σ W W − σ W W Ck CK C 1,c 2k CK Ck Ck GK Gk Gk (9) k�1 k�1 k�1 + W S Z − v − K sign v . G G 1 r 2 − W S Z v . Gk GK G 2k (e neural network weight adaptation law is designed as k�1 (18) (10) W � −Γ σ W + S Z x _ v , Dk Dk Dk Dk Dk D 1,c 2k (e following relation can be obtained as follows: T T (11) T W � −Γ σ W + S Z x _ v , Ck Ck Ck Ck Ck C 1,c 2k _ _ v W S Z x � W S Z x v , (19) 2 D D D 1,c Dk Dk D 1,c 2k k�1 (12) W � −Γ σ W + ϕ Z v . Gk Gk Gk Gk Gk G 2k T T _ _ v W S Z x � W S Z x v , (20) 2 C C C 1,c Ck Ck C 1,c 2k Among them, the matrix is Γ > 0,Γ > 0,Γ > 0. Dk Ck Gk k�1 σ , σ , σ are all positive numbers to improve robustness. Dk Ck GFk ∗ T T T (13) T W S Z � D + ε , D D D D v W S Z � W S Z v . (21) G G Gk G 2k 2 G Gk k�1 ∗ T W S Z � C + ε , (14) C C C C Based on Young’s inequality, the following formula can be obtained as ∗ T W S Z � G + ε . (15) G G G G 1 1 T T T T ∗T ∗ (22) −W W ≤ − W W + W W , Dk Dk Dk Dk Dk Dk Among them, ε , ε , and ε are minimal approximation 2 2 D C G error matrices. 1 1 T T T T ∗T ∗ T T T T T T (23) −W W ≤ − W W + W W , _ _ _ V � −v K v − v K v + v W S Z x + v W S Z x Ck Ck Ck Ck Ck Ck 2 1 1 1 2 2 2 2 D D D 1,c 2 C C C 1,c 2 2 T T + v W S Z + v E − K sign v . G G r r 2 2 G 2 1 1 T T T T ∗T ∗ (24) −W W ≤ − W W + W W . (16) Gk Gk Gk Gk Gk Gk 2 2 Substituting formulas (19) to (24) into formulas (18) and Note 1: within the designed control law, the control K ≥ ‖E ‖, we get rii ri parameter K should be selected as K ≥ ‖E ‖. In order r rii ri n n 1 T 1 T T T T T to ensure the stability of the method proposed in this _ V≤ − v K v − v K v − σ W W − σ W W 1 1 2 2 DK CK 1 2 Dk Dk Ck Ck 2 2 chapter when the double-joint manipulator is working, k�1 k�1 a larger value of K should be selected, but this is also r n n n 1 T 1 ∗ 1 ∗ T ∗T ∗T − σ W W + σ W W + σ W W not the best way for the system to induce chattering. GK Gk Gk DK Dk Dk CK Dk Ck 2 2 2 k�1 k�1 k�1 (erefore, the control parameter K should be changed to K � k x _ + k x + k . Among them, k ≥ ‖ε ‖, r D 1,c c 1,c G D D 1 ∗T ∗ + σ W W GK Gk Gk k ≥ ‖ε ‖, and k ≥ ‖ε ‖. C C G G k�1 ≤ a V + c . 0 0 2.2. System Stability Analysis. In this section, the Lyapunov (25) theorem will be used to determine the stability of the closed- loop system of the double-joint manipulator under the From inequality (25), it can be deduced that all control proposed control scheme. (e Lyapunov function is selected signals of the closed-loop system of the double-joint ma- as follows: nipulator are semiglobally uniformly asymptotically boun- n n ded. (e design flow chart of the control method in this 1 1 T T T T −1 −1 V � V + W Γ W + W Γ W chapter can be seen in Figure 1: 2 Dk DK Dk Ck CK Ck 2 2 k�1 k�1 Note 2: the size of the control parameter determines the (17) radius of the tracking error domain. (at is, the larger T T −1 + W Γ W . Gk GK Gk the control parameter λ (K ), the smaller the radius max i k�1 of the tracking error domain. Journal of Robotics 5 x x x v 1 r 2 1 + + x z x z r 1 1,c 2 ∑ CF ∑ u - - -+ + + ∑ ∑ v 2 - - 1 2 ∑ 0 sI+K · W Dk Dk Dk sI+σ Ґ Dk Dk Ck 1 Ck Ck sI+σ Ґ Ck Ck MUX 1 Gk ^ sI+σ Ґ Gk Gk Gk Gk Dk 1,c x S Ck 1,c RBF NNs x S Gk Figure 1: . Flowchart of the controller design. ��������������� � � Furthermore, through the integral inequality (25), we n� � � � Ω � v (t) ∈ R v (t) ≤ 2(V(0) + c /a , � � v 1 1 0 0 can get c c ��������������� − a t 0 − a t 0 0 0 V(t)≤ V(0)e + 1 − e ≤ V(0) + . (26) ⎧ ⎨ � � ⎫ ⎬ 2(V(0) + c /a � � a a n 0 0 � � 0 0 Ω � v (t) ∈ R v (t) ≤ , � � v 2 2 2 ⎩ ⎭ λ D x min 1 From the abovementioned inequality, we derive ��������������� � � ⎧ ⎪ ⎫ ⎪ 1 c � �2 ⎪ ⎪ 0 ⎨ � � ⎬ � � 2(V(0) + c /a � � �v (t)� ≤ V(0) + . h 0 0 (27) 1 � � Ω � W (t) ∈ R �v (t)�≤ , 2 a DK 2 W ⎪ −1 ⎪ 0 Dk ⎪ ⎪ λ Γ ⎩ ⎭ min Dk From the abovementioned inequality, it can be known ��������������� that, under the condition that all state quantities of the ⎧ ⎪ ⎫ ⎪ ⎪ ⎪ ⎨ � � ⎬ 2(V(0) + c /a � � dual-joint manipulator system can be measured, for the h 0 0 � � Ω � W (t) ∈ R �v (t)�≤ , CK 2 W ⎪ −1 ⎪ Ck ⎪ ⎪ initial compact set, λ Γ ⎩ ⎭ min Ck (v (0), v (0), W (0), W (0), W (0)) ∈ Ω , v (t) 1 2 DK CK GK 0 1 converges into the compact set Ω . By the same ��������������� ⎧ ⎪ ⎫ ⎪ ⎪ ⎪ method, v (t) can be converged into the compact set 2 ⎨ � � ⎬ 2(V(0) + c /a � � h 0 0 � � Ω , and W (0), W (0), W (0) can be converged Ω � W (t) ∈ R �v (t)�≤ GK 2 v DK CK GK W ⎪ −1 ⎪ 2 Gk ⎪ ⎪ λ Γ ⎩ ⎭ min Gk into Ω , Ω , and Ω , respectively. W W W Dk Ck Gk (28) 6 Journal of Robotics (e model of the planar two-joint manipulator system used in this chapter is as follows: − 1 ⎧ x _ � D x u − τ − C x , x x − G x , 2 1 e 1 2 2 1 (29) y � x . m T T Here, x � [x , x ] and x � [x _ , x _ ] . Among 1 11 12 2 11 12 them, the connecting rod used in this simulation is a uni- form and regular cuboid rod; then, there is l � (l /2). ci i m According to the physical model of the two-joint ma- Obstacle nipulator, its kinetic energy equation can be calculated as 1 1 1 1 2 2 2 2 2 K(q, q _) � m l q _ + I q _ + m l q _ + m l l q _ q _ + q _ cos q _ 1 c1 1 1 1 2 1 1 2 1 c2 1 1 2 2 2 2 2 1 1 Figure 2: Plane two-joint robotic arm model. + m l q _ + q _ + I q _ + q _ . 2 c2 1 2 2 1 2 2 2 (30) ������� From inequality (27), lim ‖v (t)‖≤ 2(c /a ) is easy 1 0 0 t⟶∞ (e potential energy of the two-joint manipulator can be to obtain. (at is, for any given constant u, ������� expressed as 2(c /a )≤ μ can be obtained by adjusting the pa- 0 0 rameters a and c , and thus, lim ‖v (t)‖≤ μ can be 0 0 1 P(q) � m gl sin q + m gl sin q + l sin q + q . 1 C2 1 2 1 1 c2 1 2 t⟶∞ obtained. (31) Note 3: increasing α or decreasing c can effectively ������� reduce 2(c /a ). (erefore, selecting a larger pa- According to the Euler–Lagrange equation 0 0 d/dtz(K − P)/zq _ − z(K − P)/zq � 0, the inertia matrix rameter a can make the system obtain a sufficiently small tracking error of 0. However, because the pa- D(q) of the two-joint manipulator in the joint coordinate system, the centrifugal force and Coriolis force matrix rameters α and c both contain the parameter σ , σ , σ (k � 1, 2, . . . , n), the best control perfor- C(q, q), the gravity term matrix G(q), and the Jacobian Dk Ck Gk matrix of the two-joint manipulator in the joint coordinate mance of the system requires the designer to accu- mulate experience in simulation research. system are expressed as 3. Analysis of Simulation Results As shown in Figure 2, it is used for simulation experiments. 2 2 2 m l + m l + 2l l cos q + I + I , m l + l l cos q + I , 1 c1 2 1 1 c2 2 1 2 2 c2 1 c2 2 2 ⎢ ⎥ ⎡ ⎢ ⎤ ⎥ ⎢ ⎥ ⎢ ⎥ D(q) � ⎣ , (32) 2 2 m l + l l cos q + I , m l + I . 2 c2 1 c2 2 2 2 c2 2 −m l l q _ sin q , −m l l q _ + q _ sin q , 2 1 c2 2 2 2 1 c2 1 2 2 22 C(q, q _) � , (33) m l l q _ sin q , 0. 2 1 c2 1 2 m l + m l g cos q + m l g cos q + q 1 c2 2 1 1 2 c2 1 2 G(q) � , (34) m l g cos q + q . 2 c2 1 2 − l sin q + l sin q + q , −l sin q + q , 1 1 2 1 2 2 1 2 G(q) � . (35) l cos q + l cos q + q , l cos q + q . 1 1 2 1 2 2 1 2 In order to more convincingly verify the effectiveness of We assume that the initial position of the two-joint the proposed command filter backstep impedance control manipulator in Cartesian coordinates is selected as method of the adaptive neural network, the model-based x (0) � 0.5 m and x (0) � 0.8 m. In order to verify the command filter impedance control method and the adaptive 11 12 effectiveness of the designed robot arm impedance method, neural network dynamic surface backstepping impedance an obstacle is set at the position of x0 � 0.8 m to obtain two- control are also used in the two-joint manipulator simu- joint mechanical desired impedance parameters. lation experiment for comparative analysis. (e gain Journal of Robotics 7 0.8 0.8 0.51 0.51 0.6 0.6 0.495 0.495 0 0.15 0.3 0 0.15 0.3 0.4 0.4 0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10 Time t/s Time t/s 0.8 0.8 0.8 0.8 0.6 0.6 0.7 0.7 0 0.15 0.3 0 0.15 0.3 0.4 0.4 0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10 Time t/s Time t/s Obstacle Obstacle Desired Desired MBCFIC MBCFIC Figure 5: Position tracking curve (the dynamic surface method). Figure 3: Position tracking curve (the model-based control method). 0.06 -3 ×10 0.2 0.01 0.04 1 0 0.02 -2 -0.01 4.2 4.4 4.6 0 0.2 0.4 0.8 0.51 -0.02 0 1 2 3 4 5 6 7 8 9 10 0.6 Time t/s 0.495 0 0.15 0.3 0.1 -3 0.4 0.1 ×10 0 1 2 3 4 5 6 7 8 9 10 0.05 0.05 0 Time t/s -3 1 0 0.2 0.4 4.2 4.4 4.6 0.8 -0.05 0 1 2 3 4 5 6 7 8 9 10 0.8 Time t/s 0.6 0.7 MBCFIC 0 0.15 0.3 0.4 ANNCFIC 0 1 2 3 4 5 6 7 8 9 10 ANNDSIC Time t/s Figure 6: Comparison of position tracking errors of three control Obstacle schemes. Desired MBCFIC Figure 4: Position tracking curve (the command filtering method). ×[−1, 1] × [−1, 1]. (e initial weight of the RBF neural network is set to 0, and its adaptive law parameters are selected as Γ � Γ � Γ � diag[50, 50] and DK CK GK parameters of the three manipulator impedance control σ � σ � σ � 0.01. (e command filter param- DK CK GK methods are selected as follows: eters are selected as w = 300 and g = 0.5, and the n n Method 1: for the model-based command filter im- control gain parameters are selected as pedance control (MBCFIC) method, the control law of K � diag[50, 50] and K � diag[50, 50]. 1 2 the control method is designed as u � −K z + τ 2 2 e Method 3: for the adaptive neural network dynamic +D(x )x _ + C(x , x )x + G(x ) − v . (e control 1 1,c 1 2 1,c 1 1 surface backstepping impedance control (ANNDSIC) gain parameters are chosen as K � diag[50, 50] and strategy, the dynamic surface used in this strategy is K � diag[50, 50]. 2 selected as Tα _ + α � α (T is a positive number). d d Method 2: for the adaptive neural network command (erefore, the real control law of the control strategy is filtering backstep impedance control (ANNCFIC) set to u � −K z + τ + W S (Z ) (α − α /T) 2 2 e D D D d T T method, the center point of the RBF neural network in x _ + W S (Z )α + W S (Z ) − Ksign(v ), and the 1,c C C C G G G 2 this paper is selected as [−1, 1]< [−1, 1]× neural network structure and parameter selection are [−1, 1]∗ [−1, 1] × [−1, 1] × [−11]∗ [−1, 1]∗ [−11] consistent with method 1. (e control gain parameters Y/m X/m Y/m X/m Position error/m Position error/m Y/m X/m of Y axis of X axis 8 Journal of Robotics 0.5 -70 -140 0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 456 789 10 Time t/s Time t (s) u (MBCFIC) u (MBCFIC) 1.5 0 1 2 3 456 789 10 -60 Time t (s) -120 -180 0 1 2 3 4 5 6 7 8 9 10 Time t/s u (MBCFIC) 1 0 0 1 2 3 456 789 10 u (MBCFIC) Time t (s) Figure 9: Neural network weight estimation curve for the proposed -60 control scheme. -120 -180 0 1 2 3 4 5 6 7 8 9 10 Time t/s u (MBCFIC) Table 1: Statistical table of the motion error of robotic arm. u (MBCFIC) Num Error ratio Figure 7: Control input curves of three control schemes. 1 0.0431 2 0.0458 3 0.0305 4 0.0479 5 0.0030 6 0.0387 7 0.0218 8 0.0105 9 0.0262 -1 10 0.0397 -2 11 0.0361 0 1 23456789 10 12 0.0197 Time t (s) 13 0.0053 x , 1 c1 14 0.0161 15 0.0330 16 0.0455 0 17 0.0378 18 0.0223 -2 19 0.0286 -4 20 0.0317 -6 21 0.0270 0 1 2 3 4 56789 10 22 0.0268 Time t (s) 23 0.0196 x , 24 0.0420 1 c1 1 25 0.0195 26 0.0316 Figure 8: Curves of α and x . 1,c 27 0.0415 28 0.0267 29 0.0407 are selected as K � diag[50, 50] and K � diag[50, 50], 1 2 30 0.0093 and the dynamic surface control parameters are se- 31 0.0196 lected as T = 0.01. 32 0.0367 Figures 3–6 show the position tracking diagrams ob- 33 0.0406 34 0.0428 tained from the simulation experiments of the three 35 0.0393 schemes. 3 to 5 are the position tracking curves of the model- 36 0.0386 based command filter impedance control method, the adaptive neural network command filter backstep u (N) α &x , α &x , u (N) u (N) 1 1 c1 1 1 c1 ^ ^ ^ | | W | | | | W | | | | W | | G C D Journal of Robotics 9 improve the motion accuracy of the double-joint impedance control method, and the adaptive neural network dynamic surface backstep impedance control method, manipulator. respectively. Figure 6 is a comparison diagram of the position tracking errors of these three control strategies. 4. Conclusion It can be seen from Figures 3 and 4 that the proposed (e force/position control of the double-joint manipulator neural network control method can achieve a good system has become the general trend. (e impedance control position tracking effect when the dynamic model of the strategy of the double-joint manipulator treats the force manipulator is difficult to obtain accurately. (is effec- control and the position control as a unified whole. Com- tively proves that the adaptive neural network control pared with other current multijoint manipulator force/po- can better approximate the uncertain nonlinear terms in sition control methods, this method has very significant the system. It can be seen from Figures 4 and 5 that the advantages. In the process of actual use, the amount of command filtering method used in this section has a calculation is relatively small, and it has good robustness to smaller tracking error than the dynamic surface method. changes in the external unknown environment. To achieve (is proves that the designed error compensation good impedance control of a multijoint manipulator, its end mechanism can effectively eliminate the filtering error, effector needs to have good position tracking performance. which is beneficial for the robotic arm to be better ap- However, the multijoint manipulator is a highly nonlinear plied to tasks with high tracking accuracy. It can be seen system with strong coupling effect which makes it difficult to from Figure 6 that the proposed neural network com- construct the system model accurately, and also there are mand filter impedance control method has a good po- many unknown external disturbances. (is paper combines sition tracking performance whether it is in contact with the RBF neural network to control the parameters of the external obstacles or not. double-joint manipulator to improve the control effect of the Figure 7 shows a graph of the control input of the three two-joint manipulator and promote the work efficiency of control methods. It can be seen from the graph that the the manipulator. (e experimental study verifies that the control input of the manipulator is kept within an ap- parameter-tunable RBF neural network control method propriate range. Furthermore, it can be seen from Figure 8 facing the double-joint manipulator has a certain effect on that when the end effector of the robotic arm comes in the parameter control process of the double-joint manip- contact with an external obstacle, a contact collision force ulator. Moreover, the parameter-tunable RBF neural net- is initially generated, but the impedance relationship can work control method facing the double-joint manipulator be quickly used to keep the contact force in a suitable can effectively improve the motion accuracy of the double- range. (erefore, the proposed filtering can quickly realize joint manipulator. impedance control, so that the end of the manipulator has a relatively safe physical interaction force when it contacts Data Availability the external unknown environment. (is proves that the proposed control method can effectively guarantee the (e labeled dataset used to support the findings of this study safety of the manipulator when it is in contact with an is available from the corresponding author upon request. unknown external environment. It can be seen from Figure 9 that the estimated value of the neural network Conflicts of Interest weight of the control method proposed in this chapter is bounded. (rough the comparative analysis of the above (e author declares that there are no conflicts of interest. three control schemes, it can be proved that the designed adaptive neural network command filtering backstep Acknowledgments impedance control method for the dual-joint manipulator system can effectively solve the problem that the dynamic (is work was supported by Hubei University of model of the manipulator is difficult to obtain accurately. Technology. Moreover, it enables the robot arm to obtain a good position tracking effect and impedance tracking perfor- References mance, ensuring the safety of the robot arm in contact with the unknown external environment. [1] Z. Xin, X. Wenbo, and L. Wenru, “(e control strategy of (e abovementioned research study verifies that the manipulator based on fractional-order iterative learning,” Automatic Control and Computer Sciences, vol. 55, no. 4, parameter-tunable RBF neural network control method for pp. 368–376, 2021. the dual-joint manipulator has certain effects on the pa- [2] M. Bugday and M. Karali, “Design optimization of industrial rameter control process of the dual-joint manipulator. On robot arm to minimize redundant weight,” Engineering Sci- this basis, through the simulation of multiple sets of data, the ence and Technology, an International Journal, vol. 22, no. 1, motion error of the manipulator is calculated, and the results pp. 346–352, 2019. shown in Table 1 are obtained. [3] S. Phukan and C. Mahanta, “A position synchronization It can be seen from the abovementioned research study controller for co-ordinated links (COOL) dual robot arm that the parameter-tunable RBF neural network control based on integral sliding mode: design and experimental method facing the double-joint manipulator can effectively validation,” International Journal of Automation and Com- puting, vol. 18, no. 1, pp. 110–123, 2021. 10 Journal of Robotics [4] K. Morales, C. Hoyos, and J. M. Garc´ıa, “Mechanical structure design and optimization of a hu-manoid robot arm for ed- ucation,” Journal of Autonomous Intelligence, vol. 5, no. 1, pp. 95–109, 2022. [5] J. K. Min, D. W. Kim, and J. B. Song, “A wall-mounted robot arm equipped with a 4-DOF yaw-pitch-yaw-pitch counter- balance mechanism,” IEEE Robotics and Automation Letters, vol. 5, no. 3, pp. 3768–3774, 2020. [6] S. Mori, K. Tanaka, S. Nishikawa, R. Niiyama, and Y. Kuniyoshi, “High-speed and lightweight humanoid robot arm for a skillful badminton robot,” IEEE Robotics and Au- tomation Letters, vol. 3, no. 3, pp. 1727–1734, 2018. ´ ´ ´ [7] D. Saenz Zamarron, N. I. Arana de las Casas, E. Garcıa Grajeda, J. F. Alatorre Avila, and J. U. Naciff Arroyo, “Ed- ucational robot arm development,” Computacion Y Sistemas, vol. 24, no. 4, pp. 1387–1401, 2020. [8] H. Fu and C. Ma, “Asynchronous resource-aware control for uncertain semi-Markov jump systems with application to robot arm,” Optimal Control Applications and Methods, vol. 43, no. 3, pp. 925–942, 2022. [9] M. Mukhtar, D. Khudher, and T. Kalganova, “A control structure for ambidextrous robot arm based on Multiple Adaptive Neuro-Fuzzy Inference System,” IET Control >eory & Applications, vol. 15, no. 11, pp. 1518–1532, 2021. [10] A. Hitzmann, H. Masuda, S. Ikemoto, and K. Hosoda, “Anthropomorphic musculoskeletal 10 degrees-of-freedom robot arm driven by pneumatic artificial muscles,” Advanced Robotics, vol. 32, no. 15, pp. 865–878, 2018. [11] H. H. Olewi, “Design and implementation of haptic robot arm,” Association of Arab Universities Journal of Engineering Sciences, vol. 25, no. 5, pp. 256–268, 2018. [12] F. Xu, H. Wang, K. W. S. Au, W. Chen, and Y. Miao, “Underwater dynamic modeling for a cable-driven soft robot arm,” IEEE/ASME transactions on Mechatronics, vol. 23, no. 6, pp. 2726–2738, 2018. [13] N. M. Ghaleb and A. A. Aly, “Modeling and control of 2-DOF robot arm,” International Journal of Emerging Engineering Research and Technology, vol. 6, no. 11, pp. 24–31, 2018. [14] W. S. Pambudi, E. Alfianto, A. Rachman, and D. P. Hapsari, “Simulation design of trajectory planning robot manipulator,” Bulletin of Electrical Engineering and Informatics, vol. 8, no. 1, pp. 196–205, 2019. [15] H. Yang, M. Xu, W. Li, and S. Zhang, “Design and imple- mentation of a soft robotic arm driven by SMA coils,” IEEE Transactions on Industrial Electronics, vol. 66, no. 8, pp. 6108–6116, 2019. [16] K. Kunal, A. Z. Arfianto, J. E. Poetro, F. Waseel, and R. A. Atmoko, “Accelerometer implementation as feedback on 5 degree of freedom arm robot,” Journal of Robotics and Control (JRC), vol. 1, no. 1, pp. 31–34, 2020. [17] M. Giuliani, D. Szcze˛´sniak-Stanczyk, ´ N. Mirnig et al., “User- centred design and evaluation of a tele-operated echocardi- ography robot,” Health Technology, vol. 10, no. 3, pp. 649–665, [18] X. Liang, H. Cheong, Y. Sun, J. Guo, C. K. Chui, and C. H. Yeow, “Design, characterization, and implementation of a two-DOF fabric-based soft robotic arm,” IEEE Robotics and Automation Letters, vol. 3, no. 3, pp. 2702–2709, 2018. [19] V. D. Cong, “Industrial robot arm controller based on pro- grammable System-on-Chip device,” FME Transactions, vol. 49, no. 4, pp. 1025–1034, 2021.
Journal
Journal of Robotics
– Hindawi Publishing Corporation
Published: Sep 8, 2022
