Adaptive Learning with Large Variability of Teaching Signals for Neural Networks and Its Application to Motion Control of an Industrial Robot

Size: px
Start display at page:

Download "Adaptive Learning with Large Variability of Teaching Signals for Neural Networks and Its Application to Motion Control of an Industrial Robot"

Transcription

1 International Journal of Automation and Computing 8(1), February 2011, DOI: /s Adaptive Learning with Large Variability of Teaching Signals for Neural Networks and Its Application to Motion Control of an Industrial Robot Fusaomi Nagata 1 Keigo Watanabe 2 1 Department of Mechanical Engineering, Faculty of Engineering, Tokyo University of Science, Sanyo-Onoda , Japan 2 Department of Intelligent Mechanical Systems, Graduate School of Natural Science and Technology, Okayama University, Okayama , Japan Abstract: Recently, various control methods represented by proportional-integral-derivative (PID) control are used for robotic control. To cope with the requirements for high response and precision, advanced feedforward controllers such as gravity compensator, Coriolis/centrifugal force compensator and friction compensators have been built in the controller. Generally, it causes heavy computational load when calculating the compensating value within a short sampling period. In this paper, integrated recurrent neural networks are applied as a feedforward controller for PUMA560 manipulator. The feedforward controller works instead of gravity and Coriolis/centrifugal force compensators. In the learning process of the neural network by using back propagation algorithm, the learning coefficient and gain of sigmoid function are tuned intuitively and empirically according to teaching signals. The tuning is complicated because it is being conducted by trial and error. Especially, when the scale of teaching signal is large, the problem becomes crucial. To cope with the problem which concerns the learning performance, a simple and adaptive learning technique for large scale teaching signals is proposed. The learning techniques and control effectiveness are evaluated through simulations using the dynamic model of PUMA560 manipulator. Keywords: Neural networks, large-scale teaching signal, sigmoid function, adaptive learning, servo system, PUMA560 manipulator, trajectory following control, nonlinear control. 1 Introduction In this decade, open architecture industrial robots have been produced from several industrial robot makers such as KAWASAKI Heavy Industries, Ltd., MITSUBISHI Heavy Industries, Ltd. and YASKAWA Electric Corp., and so on. Open architecture described in this article means that the servo system and kinematics of the robot are technically opened, so that various applications required in industrial fields are allowed to be planned and developed at the user side. For example, non-taught operation by using a CAD/CAM system can be considered due to the opened accurate kinematics. Also, force control strategy that uses a force sensor can be easily implemented on the technically open discrete-time servo system [1]. It is now possible to model and simulate many types of robots. For example, Chen et al. [2] presented a new design of an environment for simulation, animation, and visualization of sensor-driven robots. Although conventional computer-graphics-based robot simulation and animation software packages lack the capabilities for robot sensing simulation, the system was designed to overcome the deficiency. Also, Benimeli et al. [3] addressed the implementation and comparison of an indirect identification procedure and a direct identification procedure on an industrial robot provided with an open control architecture. The estimation of dynamic parameters in mechanical systems constituted an issue of crucial importance for dynamic simulations where high accuracy was required. Manuscript received March 2, 2010; revised June 17, 2010 This work was supported by Grant-in-Aid for Scientific Research (C) (No ) of Japan. We have introduced a simulation technique of velocitybased discrete-time control system for open architectural industrial robots [4]. In order to develop a novel velocitybased control system, which is represented in discrete-time domain for an open architecture industrial robot, it is required from the viewpoint of safety, cost and easiness to preliminarily examine and evaluate the characteristics and performance. In such a case, the proposed simulation techniques are useful. Neural network, which is one of the representative intelligent control approaches, was paid attention to improve the control performance of robotic systems. Han and Moraga [5] reported a fundamental and important result in which a sigmoid function was employed in each unit. A variant sigmoid function with three parameters that denoted the dynamic range, symmetry and slope of the function was discussed. How these parameters influence the speed of back propagation learning was illustrated, and a hybrid sigmoidal network with different parameter configuration in different layers was introduced. The error signal problem, oscillation problem and asymmetrical input problem could be handled by regulating and modifying the sigmoid function parameter configuration in different layers. As for the application of neural networks to robotic control, many control systems and their learning techniques have been developed. For example, Torras [6] described that to carry out their tasks autonomously in unknown environments, it is essential for robots to adapt to their environments. The most effective method to endow robots with this capability is to the use of neural networks.

2 F. Nagata and K. Watanabe / Adaptive Learning with Large Variability of Teaching Signals for 55 The related techniques were explained with mobile robots and manipulator arms. Hasan et al. [7] proposed an adaptive learning strategy using an artificial neural network to control the motion of a manipulator robot with six degreeof-freedom and to overcome the inverse kinematics problem, which are mainly singularities and uncertainties in arm configurations. Horng [8] showed that Levenberg-Marquardt back-propagation (LMBP) has faster convergence if compared with other three back-propagation modified algorithms, and it is suitable for system identification and controller design. Matlab/Simulink and LabVIEW with neural networks were successfully integrated to develop a supervisory control and data acquisition (SCADA) system of AC servo motor. Also, Huang et al. [9] developed an adaptive neural network algorithm for a class of interconnected nonlinear systems. Neural networks were applied to approximate the unknown nonlinear functions and interconnections in the subsystems. A systematic approach was established to synthesize the adaptive learning control scheme. The effectiveness was demonstrated by simulations. Furthermore, [10, 11] Wu et al. conducted the stability analysis of neural networks with time-varying delay. Up to now, however, it seems that the consideration about the scale of teaching signal in the output layer has not been sufficiently taken yet. Recently, various control methods represented by proportional-integral-derivative (PID) control are used for robotic control. To cope with the requirements for high response and precision, advanced feedforward controllers such as gravity compensator, Coriolis/centrifugal force compensator and friction compensators have been built in the controller. However, it generally requires a serious computational effort to calculate the compensating value within a short sampling period. In this paper, a recurrent neural network is applied for a feedforward controller of PUMA560 manipulator, which is a representative of industrial robots with six degree-offreedom. Hereafter, the recurrent neural network is called RNN. The feedforward controller includes gravity compensator and Coriolis/centrifugal force compensator. Also, a simple and adaptive learning technique is proposed for a robotic servo system to deal with large-scale joint driving torques. The learning technique and control effectiveness are evaluated through simulations using the dynamic model of PUMA560 manipulator as shown in Fig. 1. The remainder of this paper is organized as follows: Section 2 explains the computed torque control method used as a model-based servo system. Then, teaching signals for RNNs are calculated. In Section 3, integrated RNNs are designed to compensate nonlinear terms such as gravity term and Coriolis/centrifugal force term. Furthermore, a simple and adaptive learning technique by using a scaler is proposed for large-scale teaching signals processed in back propagation algorithm, then the effectiveness is discussed. In Section 4, a servo system using the integrated RNNs is applied to a trajectory following control problem. The result demonstrates that the integrated RNNs can work well instead of gravity and Coriolis/centrifugal force terms. Finally, the conclusions are presented. 2 Model-based robotic servo system In order to simulate the motion of an industrial robot, first of all, a servo system is considered and designed. Here, the computed torque control method [12, 13] is applied in a servo system. The computed torque control method is used for nonlinear control of an industrial manipulator, which is composed of a model based portion and a servo portion. The servo portion is a closed loop with respect to the position and velocity. On the other hand, the model based portion has the inertia term, gravity term and Coriolis/centrifugal force term, which work to cancel the nonlinearity of the manipulator. In order to realize high control stability, the position and velocity feedback gains used in the servo portion should be tuned suitably. In this section, a simple desired trajectory as shown in Fig. 1 is calculated for the computed torque control method. The computed torque control method generates joint driving torque to follow the trajectory. The desired trajectory and joint driving torque are used as teaching signals for neural networks. 2.1 Computed torque control method The dynamic model of a manipulator with friction term is generally given by M(θ) θ + H(θ, θ) + G(θ) + F r(θ, θ) = τ (1) where M(θ) R 6 6 is the inertia term in joint space. H(θ, θ) R 6 1 and G(θ) R 6 1 are the Coriolis/centrifugal force term and gravity term, respectively. F r(θ, θ) R 6 1 is the friction term consisting of viscous friction and Coulomb friction. θ R 6 1, θ R 6 1 and θ R 6 1 are the angle, angle velocity, and acceleration vectors in joint coordinate system, respectively. τ R 6 1 is the joint driving torque vector. In the case that the computed torque control law is employed in the servo system of a manipulator, the desired angle, angle velocity and acceleration vectors in joint coordinate system are respectively given to the references of the servo system, so that the joint driving torque can be calculated from τ = ˆM(θ) [ θr + K v{ θ r θ} ] + K p{θ r θ} + Fig. 1 Image of a desired trajectory composed of a sine curve Ĥ(θ, θ) + Ĝ(θ) (2) where, ˆM, Ĥ(θ, θ), and Ĝ(θ) denote the modeled terms. θ r R 6 1, θr R 6 1, and θ r R 6 1 are the desired

3 56 International Journal of Automation and Computing 8(1), February 2011 angle, angle velocity and acceleration vectors, respectively. K v = diag{k v1,, K v6} and K p = diag{k p1,, K p6} are the feedback gains of velocity and position, respectively. Note that θ, θ in (2) are actual values, i.e., controlled variables. The nonlinear compensation terms ˆM(θ), Ĥ(θ, θ) and Ĝ(θ) are calculated by using the recursive Newton- Euler formulation [14] to cancel the nonlinearity and are effective to achieve stable trajectory control. The block diagram of the computed torque control method is shown in Fig. 2. Although the gravity term and Coriolis/centrifugal force term are ordinarily obtained by computing the inverse dynamics through recursive Newton-Euler formulation, the computational load within a sampling period is considerably large. Fig. 3 Desired joint angle vector θ r R 6 1 realizing the trajectory shown in Fig. 1 Fig. 2 Block diagram of the computed torque control method, in which θ r, θ r, and θ r are the desired angle, velocity and acceleration vectors in joint coordinate system 2.2 Teaching signal for RNN As an example, the desired trajectory in Cartesian space has been designed as shown in Fig. 1. The tip of the robot arm draws a sine curve in x-y plane. The trajectory in Cartesian space is resolved into the trajectories in joint coordinate system by using the inverse kinematics of PUMA560, which are shown in Fig. 3. Also, Fig. 4 shows the joint driving torques needed for realizing the trajectory shown in Fig. 1, which are obtained through a trajectory following simulation by using (2). The simulation was carried out by using the dynamic model of PUMA560 manipulator on Matlab system [14 16]. The robotic dynamic model with the friction torque term given by (1) was applied. The friction torque term F r(θ, θ) consists of the viscous friction torque and Coulomb friction torque, which is represented by F r(θ, θ) = BG 2 r θ + G rτ c{sign( θ)} (3) where B is the coefficient matrix of viscous friction at each motor, G r is the reduction gear ratio matrix which represents the motor speed to joint speed, and τ c{sign( θ)} is the Coulomb friction torque, which appeared at each motor. If sign( θ i) > 0, then τ ci = τ + ci ; if sign( θ i) < 0, then τ ci = τ ci. Also, τ ci becomes 0 in case of sign( θ i) = 0. In simulations, B and G r are set to diag{0.0015, , , , , } and diag{ 62.6, 107.8, 53.7, 76.0, 71.9, 76.7}, respectively; τ + c and τ c are set to [ ] T and [ ] T, respectively [14]. In the next section, six RNNs are respectively tried to acquire the teaching signals composed of desired joint angle and joint driving torque. Fig. 4 Desired joint driving torque vector τ R 6 1 performing the trajectory shown in Fig. 1 3 Independent recurrent neural networks for an industrial robot with six joints 3.1 Adaptive learning of RNNs In order to learn the input/output relation shown in Figs. 3 and 4, an RNN is designed as shown in Fig. 5, in which the output signal with one sampling time delay is used as the feedback signal. Each unit has a standard sigmoid function as shown in Fig. 6. The RNN consists of an input layer, two hidden layers, and an output layer. Each layer has seven units, thirty units, and one unit, respectively. Note that the output layer does not have an activation function, i.e., sigmoid function. Therefore, the RNN directly yields a joint driving torque from the calculation of weighted sum. The RNN acquires the relation between the desired trajectory in joint space and joint driving torque. Although the improvement of learning performance in employing such a sigmoid function is significant, it seems that the adaptability for large-scale teaching signals has not been discussed clearly in the earlier studies. In this subsection, the learning process of RNN is explained in detail. The input to the hidden layers and output

4 F. Nagata and K. Watanabe / Adaptive Learning with Large Variability of Teaching Signals for 57 layer is a weighted sum of the previous layer. The sum is squashed into ±0.5 by the sigmoid function, which is given by f(x) = (4) 1 + exp( X) The input/output relation of each unit is given by X i,l (k) = m j=1 w i,l j,l 1 (k) o j,l 1(k) (5) where X i,l (k) is the state of i-th unit in l-th layer when k-th teaching signal is applied, w i,l j,l 1 (k) is the interconnection weight between the i-th unit in l-th layer and the j-th unit in (l 1)-th layer, o j,l 1 (k) is the output of the j-th unit in (l 1)-th layer, m is the number of units in (l 1)-th layer. Back propagation algorithm is employed in the learning process. The learning was iterated until error function E became small sufficiently. For example, in the case of first joint, the error function E is calculated by E = { S1τ1 (k) τ } 2 1(k) k=1 where τ 1 (k) and τ 1(k) are the teaching signal and output of RNN at the discrete time k (1 k 1000) with respect to the first joint, respectively. That means the pattern number of teaching signals is Here, S 1 is an adaptive element called the down scaler, which is automatically extracted from the teaching signals in output layer as S 1 = (6) 1, if max( τ 1 (k) ) λ λ max( τ 1 (k) ), otherwise (7) where max( τ 1 (k) ) is the maximum absolute value in teaching signals. λ is determined by considering the activation area in which f(x) is steeply varying from about 0.5 to 0.5 as shown in Fig. 6. For example, when the gain of sigmoid function is 1 as given by (4), λ can be set to about 5. Fig. 5 Recurrent type neural network to learn the input/output relation between six joint angles θ r1,, θ r6 and driving torque τ i (i = 1,, 6) Fig. 6 Sigmoid function used in each unit allocated in input and hidden layers w i,l j,l 1 (k) is updated by the well-known steepest descent method given by w i,l j,l 1 (k + 1) = wi,l j,l 1 (k) η E w i,l j,l 1 (k) (8) where η is the learning coefficient. If l-th layer is the output one, i.e., l = 4, the update quantity for a weight is calculated by In the above equation, E w 1,4 = δ1,4 oj,3(k). (9) j,3 (k) δ 1,4 = { S 1τ 1 (k) τ 1(k) } f { X 1,4(k) }. (10) Also, in case of the third layer, the update quantity is computed by In the above equation, E = w i,3 δi,3 oj,2(k). (11) j,2 (k) δ i,3 = δ 1,4 w 1,4 i,3 (k) f { X i,3(k) }. (12) In the same manner, w i,2 j,1 (k) is updated through where δ i,2 = E = w i,2 δi,2 oj,1(k) (13) j,1 (k) 30 n=1 δ n,3 w n,3 i,2 (k) f { X i,2(k) }. (14) After sufficient learning, it is expected that the RNN will give an output similar to the torque curve of the first joint as shown in Fig. 4. Other five RNNs designed for the joints from the second to the sixth can be also learned in the same manner. If an unknown trajectory is given and the dynamic torque for teaching signal is out of the nonlinear range, then the torque is adaptively squashed into the range of about ±5 by using (6) and (7) in the learning process.

5 58 International Journal of Automation and Computing 8(1), February 2011 Fig. 7 Learning results of joint driving torques for six joints 3.2 Learning results of RNNs In this subsection, the learning process of RNN is described. If orders of six torques in Fig. 4 were almost the same, then the output layer could have six units. However, for example, the torque within the range from 30 N m to 5 N m is required for the second joint control. The range is much wider compared with the other five joints. In such case, it is difficult to construct a sufficiently learned RNN with six units in the output layer. When the scales of the teaching signals for each unit in the output layer are quite different, the learning does not go well as a whole. That is the reason why six RNNs are independently designed. Six RNNs are learned for the trajectory following control, respectively. At first, the down scaler vector S = diag{s 1,, S 6} is set to diag{1,, 1}. The learnings of six joints were respectively repeated times in order that the values of the error functions converge to almost zero. However, the learning of the second joint could not be finished satisfactorily, i.e., the error did not decrease from the initial value. To overcome the problem concerning the learning, the second joint s teaching signals in the output layer were adaptively scaled down with S 2 = λ max( τ2 (k) ) = (15) 30 i.e., by giving S = diag{1, 0.17, 1, 1, 1, 1}, so that the error of the second joint could be reduced to almost zero as shown in Fig. 7 (b). The errors of other joints are also shown in Fig. 7. Note that when the learned RNN for the second joint is used for a feedforward controller which can compensate the nonlinear terms such as gravity term Ĝ(θ) and Coriolis/centrifugal term Ĥ(θ, θ) in (2), the output of the second joint must be scaled up with S 1 2 by multiplying S 1 = diag{1, 6, 1, 1, 1, 1}. As it can be seen, the learning is well performed as shown in Fig. 7. Strictly speaking, however, the dynamic BP algorithm is more suitable for the training algorithm of the RNN. 3.3 Discussion It is known that the form and position of sigmoid function can be changed by using a gain α and threshold β, respectively. The function and its differential are respectively represented by f(x) = (16) 1 + exp{ α(x β)} f (X) = αf(x){1 f(x)}. (17) If the gain α is set to 10, then the slope becomes steeper and the nonlinear range on horizontal axis becomes narrow as shown in Fig. 8. Because the learning algorithm written by (8) includes the differential f (X), and then learning effectiveness can be expected only in the narrow range, i.e., activation area. Accordingly, when the absolute value of input to the sigmoid function is large, i.e., out of range, the updates of weights do not progress well. On the other hand, if α is set to 0.1, the slope becomes linear and gentle. The sigmoid function loses the flexible characteristics of nonlinearity. Fig. 9 illustrates the differentials with several α. The most important point is the adjustment of the input value to a sigmoid function by using the down scaler S. For example, if α is set to 1, it is desirable that the input values move within about ±5 from the center as related in the painted area of Fig. 9. In the learning process of the output layer of the second joint, the absolute value of the error between the teaching signal and output of the corresponding RNN 2 became much larger than 5, so that the updates of weights and thresholds could not progress satisfactorily. In this section, we have proposed a simple and adaptive learning method for the second joint, in which the teaching signals are scaled down and to be mapped within about ±5, i.e., into activation area, on the horizontal axis shown in Fig. 6. The effectiveness has been confirmed from the learning result of the second joint. Of course, a similar effectiveness test will be performed by setting the gain α smaller and the learning coefficient η larger. However, the syner-

6 F. Nagata and K. Watanabe / Adaptive Learning with Large Variability of Teaching Signals for 59 gistic tuning according to the error signals in the output layer is not so easy, and time-consuming as well. It is common that antecedent membership functions are manually or automatically placed on support set according to actual fuzzy input signals when a fuzzy approach is incorporated in a control system. In some way, the proposed adaptive scaling down technique for error signals in output layer is similar to the design of such membership functions. In this section, a servo control with the learned RNNs is introduced. The six RNNs independently learned are integrated for a feedforward controller as shown in Fig. 10. The RNNs act instead of the model-base portion, i.e., Coriolis/centrifugal force term and gravity term, in the computed torque control law given by (2). The PUMA560 manipulator has six degree-of-freedoms, i.e., and six joints. The integrated RNNs are composed of six networks to independently produce the driving torque for each joint. Therefore, the array of the networks represents a dynamic model to compensate the nonlinear terms consisting of Coriolis/centrifugal force term and gravity term, which can be seen by comparing Fig. 2 with Fig. 11. Fig. 11 shows the block diagram of the control system including the integrated RNNs, in which the torque vector τ NN from the integrated RNNs is multiplied by S 1 to call back the original scales of teaching signals. Fig. 8 β Sigmoid functions with different gains α and a threshold Fig. 10 Integrated RNNs for a feedforward controller Fig. 11 Block diagram of servo system comprising the integrated RNNs illustrated in Fig. 10 Fig. 9 Differentials f (X) of sigmoid functions with different gains α 4 Advanced servo system using integrated RNNs The control system is similarly applied to the trajectory following control problem as shown in Fig. 1. Fig. 12 (a) (f) show the control results of six joints, respectively. It is observed that each joint can be successfully controlled along the desired trajectory although there exist small delays between the desired joint angles and controlled ones. In this case, the elements of the output τ F B from the servo system without gravity term and Coriolis/centrifugal force term are shown in Fig. 13 (a) (f), respectively. Note that τ F B is the output only from the feedback control with K p and K v, which does not include the compensation τ NN using the RNN. As it can be seen from the small values of six joints driving torques, the integrated RNNs work well as a feedforward controller which can compensate both the gravity term and Coriolis/centrifugal force term.

7 60 International Journal of Automation and Computing 8(1), February 2011 Fig. 12 Control results of joint angles θ 1,, θ 6 Fig. 13 Joint driving torques τ F B generated from the servo system without gravity term and Coriolis/centrifugal force term as shown in Fig Conclusions and future work In this paper, RNNs have been proposed for PUMA560 manipulator to feedforwardly control a trajectory. A basic sigmoid function is used in each unit. The feedforward controller using the integrated RNNs approximately includes gravity compensator and Coriolis/centrifugal force compensator. In order to adaptively learn such large-scale teaching signals that are out of the activation area of sigmoid function used in back propagation algorithm, the proposed RNNs scale down the teaching signals into the activation area. When the integrated RNNs are used, the output from the RNNs have only to be scaled up inversely. The adaptive learning technique by using a scale down factor and the control effectiveness of integrated RNNs has been evaluated through simulations using the dynamic model of PUMA560 manipulator. Consequently, a promising result has been confirmed compared with the conventional intuitive learning technique, in which the gain of sigmoid function and learning coefficient is synergistically selected by trial and error. As for future work, it is being planned to apply the proposed adaptive learning technique to the feedback error learning [17, 18], which is expected to be effective so as to realize online learning of the integrated RNNs. References [1] F. Nagata, K. Watanabe, T. Hase, Z. Haga, M. Omoto, K. Tsuda, O. Tsukamoto, M. Komino, Y. Kusumoto. Application of open architectural industrial robots and its simulation technique. International Journal of Computer Research, vol. 17, no. 1 2, pp. 1 40, [2] C. Chen, M. M. Trivedi, C. R. Bidlack. Simulation and animation of sensor-driven robots. IEEE Transactions on Robotics and Automation, vol. 10, no. 5, pp , 1994.

8 F. Nagata and K. Watanabe / Adaptive Learning with Large Variability of Teaching Signals for 61 [3] F. Benimeli, V. Mata, F. Valero. A comparison between direct and indirect dynamic parameter identification methods in industrial robots. Robotica, vol. 24, no. 5, pp , [4] F. Nagata. Simulation technique of velocity-based discretetime control system with intelligent control concepts for open architectural industrial robots. The Open Automation and Control Systems Journal, vol. 1, pp , [5] J. Han, C. Moraga. The Influence of the sigmoid function parameters on the speed of backpropagation learning. Lecture Notes in Computer Science, Springer, vol. 930, pp , [6] C. Torras. Robot adaptivity. Robotics and Autonomous Systems, vol. 15, no. 1 2, pp , [7] A. T. Hasan, A. M. S. Hamouda, N. Ismail, H. M. A. A. Al- Assadi. An adaptive-learning algorithm to solve the inverse kinematics problem of a 6 D.O.F serial robot manipulator. Advances in Engineering Software, vol. 37, no. 7, pp , [8] J. H. Horng. Hybrid MATLAB and LabVIEW with neural network to implement a SCADA system of AC servo motor. Advances in Engineering Software, vol. 39, no. 3, pp , [9] S. N. Huang, K. K. Tana, T. H. Lee. Neural network learning algorithm for a class of interconnected nonlinear systems. Neurocomputing, vol. 72, no 4 6, pp , [10] Y. Y. Wu, Y. Q. Wu. Stability analysis for recurrent neural networks with time-varying delay. International Journal of Automation and Computing, vol. 6, no. 3, pp , [11] Y. Y. Wu, T. Li, Y. Q. Wu. Improved exponential stability criteria for recurrent neural networks with time-varying discrete and distributed delays. International Journal of Automation and Computing, vol. 7, no. 2, pp , [12] R. P. Paul. Modeling, Trajectory Calculation and Servoing of a Computer Controlled Arm, Technical Report AIM- 177, Artificial Intelligence Laboratory, Stanford University, USA, [13] J. J. Craig. Introduction to Robitics Mechanics and Control, 2nd ed., Reading MA, USA: Addison Wesley Publishing Company, [14] P. Corke. A robotics toolbox for MATLAB. IEEE Robotics & Automation Magazine, vol. 3, no. 1, pp , [15] P. Corke. MATLAB toolboxes: Robotics and vision for students and teachers. IEEE Robotics & Automation Magazine, vol. 14, no. 4, pp , [16] F. Nagata, K. Kuribayashi, K. Kiguchi, K. Watanabe. Simulation of fine gain tuning using genetic algorithms for model-based robotic servo controllers. In Proceedings of the IEEE International Symposium on Computational Intelligence in Robotics and Automation, IEEE, Jacksonville, USA, pp , [17] M. Kawato. The feedback-error-learning neural network for supervised motor learning. Advanced Neural Computers, R. Eckmiller Ed., North-Holland, Holland: Elsevier, pp , [18] J. Nakanishi, S. Schaal. Feedback error learning and nonlinear adaptive control. Neural Networks, vol. 17, no. 10, pp , Fusaomi Nagata received the B. Eng. degree from the Department of Electronic Engineering at Kyushu Institute of Technology, Japan in 1985, and the D. Eng. degree from the Faculty of Engineering Systems and Technology at Saga University, Japan in He was a research engineer with Kyushu Matsushita Electric Co., Japan from 1985 to 1988, and a special researcher with Fukuoka Industrial Technology Center, Japan from 1988 to He is currently an associate professor at the Department of Mechanical Engineering, Faculty of Engineering, Tokyo University of Science, Yamaguchi, Japan. His research interests include intelligent control of industrial robot and its applications. nagata@ed.yama.tus.ac.jp (Corresponding author) Keigo Watanabe received the B. Eng. and M. Eng. degrees in mechanical engineering from the University of Tokushima, Japan in 1976 and 1978, respectively, and the D. Eng. degree in aeronautical engineering from Kyushu University, Japan in From 1980 to 1985, he was a research associate at Kyushu University. From 1985 to 1990, he was an associate professor at the College of Engineering, Shizuoka University, Japan. From April 1990 to March 1993, he was an associate professor, and from April 1993 to March 1998, he was a full professor in the Department of Mechanical Engineering at Saga University, Japan. From April 1998, he was with the Department of Advanced Systems Control Engineering, Graduate School of Science and Engineering, Saga University. Currently, he is with the Department of Intelligent Mechanical Systems, Graduate School of Natural Science and Technology, Okayama University, Japan. His research interests include stochastic adaptive estimation and control, robust control, neural network control, fuzzy control, genetic algorithms and their applications to the robotic control. watanabe@sys.okayama-u.ac.jp

On-line Learning of Robot Arm Impedance Using Neural Networks

On-line Learning of Robot Arm Impedance Using Neural Networks On-line Learning of Robot Arm Impedance Using Neural Networks Yoshiyuki Tanaka Graduate School of Engineering, Hiroshima University, Higashi-hiroshima, 739-857, JAPAN Email: ytanaka@bsys.hiroshima-u.ac.jp

More information

Adaptive Robust Tracking Control of Robot Manipulators in the Task-space under Uncertainties

Adaptive Robust Tracking Control of Robot Manipulators in the Task-space under Uncertainties Australian Journal of Basic and Applied Sciences, 3(1): 308-322, 2009 ISSN 1991-8178 Adaptive Robust Tracking Control of Robot Manipulators in the Task-space under Uncertainties M.R.Soltanpour, M.M.Fateh

More information

Introduction to Robotics

Introduction to Robotics J. Zhang, L. Einig 277 / 307 MIN Faculty Department of Informatics Lecture 8 Jianwei Zhang, Lasse Einig [zhang, einig]@informatik.uni-hamburg.de University of Hamburg Faculty of Mathematics, Informatics

More information

Nonlinear PD Controllers with Gravity Compensation for Robot Manipulators

Nonlinear PD Controllers with Gravity Compensation for Robot Manipulators BULGARIAN ACADEMY OF SCIENCES CYBERNETICS AND INFORMATION TECHNOLOGIES Volume 4, No Sofia 04 Print ISSN: 3-970; Online ISSN: 34-408 DOI: 0.478/cait-04-00 Nonlinear PD Controllers with Gravity Compensation

More information

Effect of number of hidden neurons on learning in large-scale layered neural networks

Effect of number of hidden neurons on learning in large-scale layered neural networks ICROS-SICE International Joint Conference 009 August 18-1, 009, Fukuoka International Congress Center, Japan Effect of on learning in large-scale layered neural networks Katsunari Shibata (Oita Univ.;

More information

GAIN SCHEDULING CONTROL WITH MULTI-LOOP PID FOR 2- DOF ARM ROBOT TRAJECTORY CONTROL

GAIN SCHEDULING CONTROL WITH MULTI-LOOP PID FOR 2- DOF ARM ROBOT TRAJECTORY CONTROL GAIN SCHEDULING CONTROL WITH MULTI-LOOP PID FOR 2- DOF ARM ROBOT TRAJECTORY CONTROL 1 KHALED M. HELAL, 2 MOSTAFA R.A. ATIA, 3 MOHAMED I. ABU EL-SEBAH 1, 2 Mechanical Engineering Department ARAB ACADEMY

More information

Robot Manipulator Control. Hesheng Wang Dept. of Automation

Robot Manipulator Control. Hesheng Wang Dept. of Automation Robot Manipulator Control Hesheng Wang Dept. of Automation Introduction Industrial robots work based on the teaching/playback scheme Operators teach the task procedure to a robot he robot plays back eecute

More information

Application of Neural Networks for Control of Inverted Pendulum

Application of Neural Networks for Control of Inverted Pendulum Application of Neural Networks for Control of Inverted Pendulum VALERI MLADENOV Department of Theoretical Electrical Engineering Technical University of Sofia Sofia, Kliment Ohridski blvd. 8; BULARIA valerim@tu-sofia.bg

More information

CONTROL OF ROBOT CAMERA SYSTEM WITH ACTUATOR S DYNAMICS TO TRACK MOVING OBJECT

CONTROL OF ROBOT CAMERA SYSTEM WITH ACTUATOR S DYNAMICS TO TRACK MOVING OBJECT Journal of Computer Science and Cybernetics, V.31, N.3 (2015), 255 265 DOI: 10.15625/1813-9663/31/3/6127 CONTROL OF ROBOT CAMERA SYSTEM WITH ACTUATOR S DYNAMICS TO TRACK MOVING OBJECT NGUYEN TIEN KIEM

More information

Decoupling Identification for Serial Two-link Robot Arm with Elastic Joints

Decoupling Identification for Serial Two-link Robot Arm with Elastic Joints Preprints of the 1th IFAC Symposium on System Identification Saint-Malo, France, July 6-8, 9 Decoupling Identification for Serial Two-link Robot Arm with Elastic Joints Junji Oaki, Shuichi Adachi Corporate

More information

Online Identification And Control of A PV-Supplied DC Motor Using Universal Learning Networks

Online Identification And Control of A PV-Supplied DC Motor Using Universal Learning Networks Online Identification And Control of A PV-Supplied DC Motor Using Universal Learning Networks Ahmed Hussein * Kotaro Hirasawa ** Jinglu Hu ** * Graduate School of Information Science & Electrical Eng.,

More information

Seul Jung, T. C. Hsia and R. G. Bonitz y. Robotics Research Laboratory. University of California, Davis. Davis, CA 95616

Seul Jung, T. C. Hsia and R. G. Bonitz y. Robotics Research Laboratory. University of California, Davis. Davis, CA 95616 On Robust Impedance Force Control of Robot Manipulators Seul Jung, T C Hsia and R G Bonitz y Robotics Research Laboratory Department of Electrical and Computer Engineering University of California, Davis

More information

ADAPTIVE NEURAL NETWORK CONTROL OF MECHATRONICS OBJECTS

ADAPTIVE NEURAL NETWORK CONTROL OF MECHATRONICS OBJECTS acta mechanica et automatica, vol.2 no.4 (28) ADAPIE NEURAL NEWORK CONROL OF MECHARONICS OBJECS Egor NEMSE *, Yuri ZHUKO * * Baltic State echnical University oenmeh, 985, St. Petersburg, Krasnoarmeyskaya,

More information

Exponential Controller for Robot Manipulators

Exponential Controller for Robot Manipulators Exponential Controller for Robot Manipulators Fernando Reyes Benemérita Universidad Autónoma de Puebla Grupo de Robótica de la Facultad de Ciencias de la Electrónica Apartado Postal 542, Puebla 7200, México

More information

Neural Network Control of Robot Manipulators and Nonlinear Systems

Neural Network Control of Robot Manipulators and Nonlinear Systems Neural Network Control of Robot Manipulators and Nonlinear Systems F.L. LEWIS Automation and Robotics Research Institute The University of Texas at Arlington S. JAG ANNATHAN Systems and Controls Research

More information

Robotics & Automation. Lecture 25. Dynamics of Constrained Systems, Dynamic Control. John T. Wen. April 26, 2007

Robotics & Automation. Lecture 25. Dynamics of Constrained Systems, Dynamic Control. John T. Wen. April 26, 2007 Robotics & Automation Lecture 25 Dynamics of Constrained Systems, Dynamic Control John T. Wen April 26, 2007 Last Time Order N Forward Dynamics (3-sweep algorithm) Factorization perspective: causal-anticausal

More information

Design Artificial Nonlinear Controller Based on Computed Torque like Controller with Tunable Gain

Design Artificial Nonlinear Controller Based on Computed Torque like Controller with Tunable Gain World Applied Sciences Journal 14 (9): 1306-1312, 2011 ISSN 1818-4952 IDOSI Publications, 2011 Design Artificial Nonlinear Controller Based on Computed Torque like Controller with Tunable Gain Samira Soltani

More information

Gain Scheduling Control with Multi-loop PID for 2-DOF Arm Robot Trajectory Control

Gain Scheduling Control with Multi-loop PID for 2-DOF Arm Robot Trajectory Control Gain Scheduling Control with Multi-loop PID for 2-DOF Arm Robot Trajectory Control Khaled M. Helal, 2 Mostafa R.A. Atia, 3 Mohamed I. Abu El-Sebah, 2 Mechanical Engineering Department ARAB ACADEMY FOR

More information

Automatic Structure and Parameter Training Methods for Modeling of Mechanical System by Recurrent Neural Networks

Automatic Structure and Parameter Training Methods for Modeling of Mechanical System by Recurrent Neural Networks Automatic Structure and Parameter Training Methods for Modeling of Mechanical System by Recurrent Neural Networks C. James Li and Tung-Yung Huang Department of Mechanical Engineering, Aeronautical Engineering

More information

Keywords- Source coding, Huffman encoding, Artificial neural network, Multilayer perceptron, Backpropagation algorithm

Keywords- Source coding, Huffman encoding, Artificial neural network, Multilayer perceptron, Backpropagation algorithm Volume 4, Issue 5, May 2014 ISSN: 2277 128X International Journal of Advanced Research in Computer Science and Software Engineering Research Paper Available online at: www.ijarcsse.com Huffman Encoding

More information

for Articulated Robot Arms and Its Applications

for Articulated Robot Arms and Its Applications 141 Proceedings of the International Conference on Information and Automation, December 15-18, 25, Colombo, Sri Lanka. 1 Forcefree Control with Independent Compensation for Articulated Robot Arms and Its

More information

Fuzzy Based Robust Controller Design for Robotic Two-Link Manipulator

Fuzzy Based Robust Controller Design for Robotic Two-Link Manipulator Abstract Fuzzy Based Robust Controller Design for Robotic Two-Link Manipulator N. Selvaganesan 1 Prabhu Jude Rajendran 2 S.Renganathan 3 1 Department of Instrumentation Engineering, Madras Institute of

More information

An experimental robot load identification method for industrial application

An experimental robot load identification method for industrial application An experimental robot load identification method for industrial application Jan Swevers 1, Birgit Naumer 2, Stefan Pieters 2, Erika Biber 2, Walter Verdonck 1, and Joris De Schutter 1 1 Katholieke Universiteit

More information

Decoupling Identification with Closed-loop-controlled Elements for Two-link Arm with Elastic Joints

Decoupling Identification with Closed-loop-controlled Elements for Two-link Arm with Elastic Joints Preprints of the 9th International Symposium on Robot Control (SYROCO'9) The International Federation of Automatic Control Nagaragawa Convention Center, Gifu, Japan, September 9-2, 29 Decoupling Identification

More information

Intelligent Systems and Control Prof. Laxmidhar Behera Indian Institute of Technology, Kanpur

Intelligent Systems and Control Prof. Laxmidhar Behera Indian Institute of Technology, Kanpur Intelligent Systems and Control Prof. Laxmidhar Behera Indian Institute of Technology, Kanpur Module - 2 Lecture - 4 Introduction to Fuzzy Logic Control In this lecture today, we will be discussing fuzzy

More information

An Adaptive LQG Combined With the MRAS Based LFFC for Motion Control Systems

An Adaptive LQG Combined With the MRAS Based LFFC for Motion Control Systems Journal of Automation Control Engineering Vol 3 No 2 April 2015 An Adaptive LQG Combined With the MRAS Based LFFC for Motion Control Systems Nguyen Duy Cuong Nguyen Van Lanh Gia Thi Dinh Electronics Faculty

More information

Lecture Schedule Week Date Lecture (M: 2:05p-3:50, 50-N202)

Lecture Schedule Week Date Lecture (M: 2:05p-3:50, 50-N202) J = x θ τ = J T F 2018 School of Information Technology and Electrical Engineering at the University of Queensland Lecture Schedule Week Date Lecture (M: 2:05p-3:50, 50-N202) 1 23-Jul Introduction + Representing

More information

Application of Artificial Neural Networks in Evaluation and Identification of Electrical Loss in Transformers According to the Energy Consumption

Application of Artificial Neural Networks in Evaluation and Identification of Electrical Loss in Transformers According to the Energy Consumption Application of Artificial Neural Networks in Evaluation and Identification of Electrical Loss in Transformers According to the Energy Consumption ANDRÉ NUNES DE SOUZA, JOSÉ ALFREDO C. ULSON, IVAN NUNES

More information

3- DOF Scara type Robot Manipulator using Mamdani Based Fuzzy Controller

3- DOF Scara type Robot Manipulator using Mamdani Based Fuzzy Controller 659 3- DOF Scara type Robot Manipulator using Mamdani Based Fuzzy Controller Nitesh Kumar Jaiswal *, Vijay Kumar ** *(Department of Electronics and Communication Engineering, Indian Institute of Technology,

More information

Design On-Line Tunable Gain Artificial Nonlinear Controller

Design On-Line Tunable Gain Artificial Nonlinear Controller Journal of Computer Engineering 1 (2009) 3-11 Design On-Line Tunable Gain Artificial Nonlinear Controller Farzin Piltan, Nasri Sulaiman, M. H. Marhaban and R. Ramli Department of Electrical and Electronic

More information

Address for Correspondence

Address for Correspondence Research Article APPLICATION OF ARTIFICIAL NEURAL NETWORK FOR INTERFERENCE STUDIES OF LOW-RISE BUILDINGS 1 Narayan K*, 2 Gairola A Address for Correspondence 1 Associate Professor, Department of Civil

More information

MOBILE ROBOT DYNAMICS WITH FRICTION IN SIMULINK

MOBILE ROBOT DYNAMICS WITH FRICTION IN SIMULINK MOBILE ROBOT DYNAMICS WITH FRICTION IN SIMULINK J. Čerkala, A. Jadlovská Department of Cybernetics and Artificial Intelligence, Faculty of Electrical Engineering and Informatics, Technical University of

More information

Application of Neuro Fuzzy Reduced Order Observer in Magnetic Bearing Systems

Application of Neuro Fuzzy Reduced Order Observer in Magnetic Bearing Systems Application of Neuro Fuzzy Reduced Order Observer in Magnetic Bearing Systems M. A., Eltantawie, Member, IAENG Abstract Adaptive Neuro-Fuzzy Inference System (ANFIS) is used to design fuzzy reduced order

More information

ARTIFICIAL NEURAL NETWORK WITH HYBRID TAGUCHI-GENETIC ALGORITHM FOR NONLINEAR MIMO MODEL OF MACHINING PROCESSES

ARTIFICIAL NEURAL NETWORK WITH HYBRID TAGUCHI-GENETIC ALGORITHM FOR NONLINEAR MIMO MODEL OF MACHINING PROCESSES International Journal of Innovative Computing, Information and Control ICIC International c 2013 ISSN 1349-4198 Volume 9, Number 4, April 2013 pp. 1455 1475 ARTIFICIAL NEURAL NETWORK WITH HYBRID TAGUCHI-GENETIC

More information

CHAPTER 4 FUZZY AND NEURAL NETWORK FOR SR MOTOR

CHAPTER 4 FUZZY AND NEURAL NETWORK FOR SR MOTOR CHAPTER 4 FUZZY AND NEURAL NETWORK FOR SR MOTOR 4.1 Introduction Fuzzy Logic control is based on fuzzy set theory. A fuzzy set is a set having uncertain and imprecise nature of abstract thoughts, concepts

More information

Lecture 10. Neural networks and optimization. Machine Learning and Data Mining November Nando de Freitas UBC. Nonlinear Supervised Learning

Lecture 10. Neural networks and optimization. Machine Learning and Data Mining November Nando de Freitas UBC. Nonlinear Supervised Learning Lecture 0 Neural networks and optimization Machine Learning and Data Mining November 2009 UBC Gradient Searching for a good solution can be interpreted as looking for a minimum of some error (loss) function

More information

Neural Network Control of an Inverted Pendulum on a Cart

Neural Network Control of an Inverted Pendulum on a Cart Neural Network Control of an Inverted Pendulum on a Cart VALERI MLADENOV, GEORGI TSENOV, LAMBROS EKONOMOU, NICHOLAS HARKIOLAKIS, PANAGIOTIS KARAMPELAS Department of Theoretical Electrical Engineering Technical

More information

MODELING WITH CURRENT DYNAMICS AND VIBRATION CONTROL OF TWO PHASE HYBRID STEPPING MOTOR IN INTERMITTENT DRIVE

MODELING WITH CURRENT DYNAMICS AND VIBRATION CONTROL OF TWO PHASE HYBRID STEPPING MOTOR IN INTERMITTENT DRIVE MODELING WITH CURRENT DYNAMICS AND VIBRATION CONTROL OF TWO PHASE HYBRID STEPPING MOTOR IN INTERMITTENT DRIVE Ryota Mori, Yoshiyuki Noda, Takanori Miyoshi, Kazuhiko Terashima Department of Production Systems

More information

Decentralized PD Control for Non-uniform Motion of a Hamiltonian Hybrid System

Decentralized PD Control for Non-uniform Motion of a Hamiltonian Hybrid System International Journal of Automation and Computing 05(2), April 2008, 9-24 DOI: 0.007/s633-008-09-7 Decentralized PD Control for Non-uniform Motion of a Hamiltonian Hybrid System Mingcong Deng, Hongnian

More information

Robust Controller Design for Speed Control of an Indirect Field Oriented Induction Machine Drive

Robust Controller Design for Speed Control of an Indirect Field Oriented Induction Machine Drive Leonardo Electronic Journal of Practices and Technologies ISSN 1583-1078 Issue 6, January-June 2005 p. 1-16 Robust Controller Design for Speed Control of an Indirect Field Oriented Induction Machine Drive

More information

Computational statistics

Computational statistics Computational statistics Lecture 3: Neural networks Thierry Denœux 5 March, 2016 Neural networks A class of learning methods that was developed separately in different fields statistics and artificial

More information

(W: 12:05-1:50, 50-N202)

(W: 12:05-1:50, 50-N202) 2016 School of Information Technology and Electrical Engineering at the University of Queensland Schedule of Events Week Date Lecture (W: 12:05-1:50, 50-N202) 1 27-Jul Introduction 2 Representing Position

More information

FUZZY LOGIC CONTROL Vs. CONVENTIONAL PID CONTROL OF AN INVERTED PENDULUM ROBOT

FUZZY LOGIC CONTROL Vs. CONVENTIONAL PID CONTROL OF AN INVERTED PENDULUM ROBOT http:// FUZZY LOGIC CONTROL Vs. CONVENTIONAL PID CONTROL OF AN INVERTED PENDULUM ROBOT 1 Ms.Mukesh Beniwal, 2 Mr. Davender Kumar 1 M.Tech Student, 2 Asst.Prof, Department of Electronics and Communication

More information

Introduction to System Identification and Adaptive Control

Introduction to System Identification and Adaptive Control Introduction to System Identification and Adaptive Control A. Khaki Sedigh Control Systems Group Faculty of Electrical and Computer Engineering K. N. Toosi University of Technology May 2009 Introduction

More information

General procedure for formulation of robot dynamics STEP 1 STEP 3. Module 9 : Robot Dynamics & controls

General procedure for formulation of robot dynamics STEP 1 STEP 3. Module 9 : Robot Dynamics & controls Module 9 : Robot Dynamics & controls Lecture 32 : General procedure for dynamics equation forming and introduction to control Objectives In this course you will learn the following Lagrangian Formulation

More information

Grinding Experiment by Direct Position / Force Control with On-line Constraint Estimation

Grinding Experiment by Direct Position / Force Control with On-line Constraint Estimation ICROS-SICE International Joint Conference 2009 August 18-21, 2009, Fukuoka International Congress Center, Japan Grinding Experiment by Direct Position / Force Control with On-line Constraint Estimation

More information

A SEASONAL FUZZY TIME SERIES FORECASTING METHOD BASED ON GUSTAFSON-KESSEL FUZZY CLUSTERING *

A SEASONAL FUZZY TIME SERIES FORECASTING METHOD BASED ON GUSTAFSON-KESSEL FUZZY CLUSTERING * No.2, Vol.1, Winter 2012 2012 Published by JSES. A SEASONAL FUZZY TIME SERIES FORECASTING METHOD BASED ON GUSTAFSON-KESSEL * Faruk ALPASLAN a, Ozge CAGCAG b Abstract Fuzzy time series forecasting methods

More information

Neural Network to Control Output of Hidden Node According to Input Patterns

Neural Network to Control Output of Hidden Node According to Input Patterns American Journal of Intelligent Systems 24, 4(5): 96-23 DOI:.5923/j.ajis.2445.2 Neural Network to Control Output of Hidden Node According to Input Patterns Takafumi Sasakawa, Jun Sawamoto 2,*, Hidekazu

More information

A new large projection operator for the redundancy framework

A new large projection operator for the redundancy framework 21 IEEE International Conference on Robotics and Automation Anchorage Convention District May 3-8, 21, Anchorage, Alaska, USA A new large projection operator for the redundancy framework Mohammed Marey

More information

In this section of notes, we look at the calculation of forces and torques for a manipulator in two settings:

In this section of notes, we look at the calculation of forces and torques for a manipulator in two settings: Introduction Up to this point we have considered only the kinematics of a manipulator. That is, only the specification of motion without regard to the forces and torques required to cause motion In this

More information

Rigid Manipulator Control

Rigid Manipulator Control Rigid Manipulator Control The control problem consists in the design of control algorithms for the robot motors, such that the TCP motion follows a specified task in the cartesian space Two types of task

More information

One-Hour-Ahead Load Forecasting Using Neural Network

One-Hour-Ahead Load Forecasting Using Neural Network IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 17, NO. 1, FEBRUARY 2002 113 One-Hour-Ahead Load Forecasting Using Neural Network Tomonobu Senjyu, Member, IEEE, Hitoshi Takara, Katsumi Uezato, and Toshihisa Funabashi,

More information

Extremum Seeking for Dead-Zone Compensation and Its Application to a Two-Wheeled Robot

Extremum Seeking for Dead-Zone Compensation and Its Application to a Two-Wheeled Robot Extremum Seeking for Dead-Zone Compensation and Its Application to a Two-Wheeled Robot Dessy Novita Graduate School of Natural Science and Technology, Kanazawa University, Kakuma, Kanazawa, Ishikawa, Japan

More information

DESIGNING CNN GENES. Received January 23, 2003; Revised April 2, 2003

DESIGNING CNN GENES. Received January 23, 2003; Revised April 2, 2003 Tutorials and Reviews International Journal of Bifurcation and Chaos, Vol. 13, No. 10 (2003 2739 2824 c World Scientific Publishing Company DESIGNING CNN GENES MAKOTO ITOH Department of Information and

More information

ANN Control of Non-Linear and Unstable System and its Implementation on Inverted Pendulum

ANN Control of Non-Linear and Unstable System and its Implementation on Inverted Pendulum Research Article International Journal of Current Engineering and Technology E-ISSN 2277 4106, P-ISSN 2347-5161 2014 INPRESSCO, All Rights Reserved Available at http://inpressco.com/category/ijcet ANN

More information

Appendix A Prototypes Models

Appendix A Prototypes Models Appendix A Prototypes Models This appendix describes the model of the prototypes used in Chap. 3. These mathematical models can also be found in the Student Handout by Quanser. A.1 The QUANSER SRV-02 Setup

More information

( t) Identification and Control of a Nonlinear Bioreactor Plant Using Classical and Dynamical Neural Networks

( t) Identification and Control of a Nonlinear Bioreactor Plant Using Classical and Dynamical Neural Networks Identification and Control of a Nonlinear Bioreactor Plant Using Classical and Dynamical Neural Networks Mehmet Önder Efe Electrical and Electronics Engineering Boðaziçi University, Bebek 80815, Istanbul,

More information

ELEC4631 s Lecture 2: Dynamic Control Systems 7 March Overview of dynamic control systems

ELEC4631 s Lecture 2: Dynamic Control Systems 7 March Overview of dynamic control systems ELEC4631 s Lecture 2: Dynamic Control Systems 7 March 2011 Overview of dynamic control systems Goals of Controller design Autonomous dynamic systems Linear Multi-input multi-output (MIMO) systems Bat flight

More information

(Feed-Forward) Neural Networks Dr. Hajira Jabeen, Prof. Jens Lehmann

(Feed-Forward) Neural Networks Dr. Hajira Jabeen, Prof. Jens Lehmann (Feed-Forward) Neural Networks 2016-12-06 Dr. Hajira Jabeen, Prof. Jens Lehmann Outline In the previous lectures we have learned about tensors and factorization methods. RESCAL is a bilinear model for

More information

On Comparison of Neural Observers for Twin Rotor MIMO System

On Comparison of Neural Observers for Twin Rotor MIMO System nternational Journal of Electronic and Electrical Engineering. SSN 0974-74 Volume 7, Number 9 (04), pp. 987-99 nternational Research Publication House http://www.irphouse.com On Comparison of Neural Observers

More information

MODELLING AND SIMULATION OF AN INVERTED PENDULUM SYSTEM: COMPARISON BETWEEN EXPERIMENT AND CAD PHYSICAL MODEL

MODELLING AND SIMULATION OF AN INVERTED PENDULUM SYSTEM: COMPARISON BETWEEN EXPERIMENT AND CAD PHYSICAL MODEL MODELLING AND SIMULATION OF AN INVERTED PENDULUM SYSTEM: COMPARISON BETWEEN EXPERIMENT AND CAD PHYSICAL MODEL J. S. Sham, M. I. Solihin, F. Heltha and Muzaiyanah H. Faculty of Engineering, Technology &

More information

Chapter 2 Review of Linear and Nonlinear Controller Designs

Chapter 2 Review of Linear and Nonlinear Controller Designs Chapter 2 Review of Linear and Nonlinear Controller Designs This Chapter reviews several flight controller designs for unmanned rotorcraft. 1 Flight control systems have been proposed and tested on a wide

More information

ROBUST FRICTION COMPENSATOR FOR HARMONIC DRIVE TRANSMISSION

ROBUST FRICTION COMPENSATOR FOR HARMONIC DRIVE TRANSMISSION Proceedings of the 1998 IEEE International Conference on Control Applications Trieste, Italy 1-4 September 1998 TAO1 12:lO ROBUST FRICTION COMPENSATOR FOR HARMONIC DRIVE TRANSMISSION H.D. Taghirad K. N.

More information

Introduction to centralized control

Introduction to centralized control ROBOTICS 01PEEQW Basilio Bona DAUIN Politecnico di Torino Control Part 2 Introduction to centralized control Independent joint decentralized control may prove inadequate when the user requires high task

More information

Balancing and Control of a Freely-Swinging Pendulum Using a Model-Free Reinforcement Learning Algorithm

Balancing and Control of a Freely-Swinging Pendulum Using a Model-Free Reinforcement Learning Algorithm Balancing and Control of a Freely-Swinging Pendulum Using a Model-Free Reinforcement Learning Algorithm Michail G. Lagoudakis Department of Computer Science Duke University Durham, NC 2778 mgl@cs.duke.edu

More information

A Sliding Mode Controller Using Neural Networks for Robot Manipulator

A Sliding Mode Controller Using Neural Networks for Robot Manipulator ESANN'4 proceedings - European Symposium on Artificial Neural Networks Bruges (Belgium), 8-3 April 4, d-side publi., ISBN -9337-4-8, pp. 93-98 A Sliding Mode Controller Using Neural Networks for Robot

More information

Eigen Vector Descent and Line Search for Multilayer Perceptron

Eigen Vector Descent and Line Search for Multilayer Perceptron igen Vector Descent and Line Search for Multilayer Perceptron Seiya Satoh and Ryohei Nakano Abstract As learning methods of a multilayer perceptron (MLP), we have the BP algorithm, Newton s method, quasi-

More information

Adaptive fuzzy observer and robust controller for a 2-DOF robot arm

Adaptive fuzzy observer and robust controller for a 2-DOF robot arm Adaptive fuzzy observer and robust controller for a -DOF robot arm S. Bindiganavile Nagesh, Zs. Lendek, A.A. Khalate, R. Babuška Delft University of Technology, Mekelweg, 8 CD Delft, The Netherlands (email:

More information

Introduction to centralized control

Introduction to centralized control Industrial Robots Control Part 2 Introduction to centralized control Independent joint decentralized control may prove inadequate when the user requires high task velocities structured disturbance torques

More information

DISTURBANCE ATTENUATION IN A MAGNETIC LEVITATION SYSTEM WITH ACCELERATION FEEDBACK

DISTURBANCE ATTENUATION IN A MAGNETIC LEVITATION SYSTEM WITH ACCELERATION FEEDBACK DISTURBANCE ATTENUATION IN A MAGNETIC LEVITATION SYSTEM WITH ACCELERATION FEEDBACK Feng Tian Department of Mechanical Engineering Marquette University Milwaukee, WI 53233 USA Email: feng.tian@mu.edu Kevin

More information

The PVTOL Aircraft. 2.1 Introduction

The PVTOL Aircraft. 2.1 Introduction 2 The PVTOL Aircraft 2.1 Introduction We introduce in this chapter the well-known Planar Vertical Take-Off and Landing (PVTOL) aircraft problem. The PVTOL represents a challenging nonlinear systems control

More information

Methods for Supervised Learning of Tasks

Methods for Supervised Learning of Tasks Methods for Supervised Learning of Tasks Alper Denasi DCT 8.43 Traineeship report Coaching: Prof. dr. H. Nijmeijer Technische Universiteit Eindhoven Department Mechanical Engineering Dynamics and Control

More information

Nonlinear System Identification Based on a Novel Adaptive Fuzzy Wavelet Neural Network

Nonlinear System Identification Based on a Novel Adaptive Fuzzy Wavelet Neural Network Nonlinear System Identification Based on a Novel Adaptive Fuzzy Wavelet Neural Network Maryam Salimifard*, Ali Akbar Safavi** *School of Electrical Engineering, Amirkabir University of Technology, Tehran,

More information

Speed Control of PMSM Drives by Using Neural Network Controller

Speed Control of PMSM Drives by Using Neural Network Controller Advance in Electronic and Electric Engineering. ISSN 2231-1297, Volume 4, Number 4 (2014), pp. 353-360 Research India Publications http://www.ripublication.com/aeee.htm Speed Control of PMSM Drives by

More information

Virtual Passive Controller for Robot Systems Using Joint Torque Sensors

Virtual Passive Controller for Robot Systems Using Joint Torque Sensors NASA Technical Memorandum 110316 Virtual Passive Controller for Robot Systems Using Joint Torque Sensors Hal A. Aldridge and Jer-Nan Juang Langley Research Center, Hampton, Virginia January 1997 National

More information

Using NEAT to Stabilize an Inverted Pendulum

Using NEAT to Stabilize an Inverted Pendulum Using NEAT to Stabilize an Inverted Pendulum Awjin Ahn and Caleb Cochrane May 9, 2014 Abstract The inverted pendulum balancing problem is a classic benchmark problem on which many types of control implementations

More information

A Sliding Mode Control based on Nonlinear Disturbance Observer for the Mobile Manipulator

A Sliding Mode Control based on Nonlinear Disturbance Observer for the Mobile Manipulator International Core Journal of Engineering Vol.3 No.6 7 ISSN: 44-895 A Sliding Mode Control based on Nonlinear Disturbance Observer for the Mobile Manipulator Yanna Si Information Engineering College Henan

More information

Study on Dahlin algorithm of Brushless DC motor based on Neural. Network

Study on Dahlin algorithm of Brushless DC motor based on Neural. Network Joint International Mechanical, Electronic and Information Technology Conference (JIMET 205) Study on Dahlin algorithm of Brushless DC motor based on Neural Network ZILONG HUANG DAN WANG2, LELE XI 3, YANKAI

More information

Secondary Frequency Control of Microgrids In Islanded Operation Mode and Its Optimum Regulation Based on the Particle Swarm Optimization Algorithm

Secondary Frequency Control of Microgrids In Islanded Operation Mode and Its Optimum Regulation Based on the Particle Swarm Optimization Algorithm International Academic Institute for Science and Technology International Academic Journal of Science and Engineering Vol. 3, No. 1, 2016, pp. 159-169. ISSN 2454-3896 International Academic Journal of

More information

1. Introduction. 2. Artificial Neural Networks and Fuzzy Time Series

1. Introduction. 2. Artificial Neural Networks and Fuzzy Time Series 382 IJCSNS International Journal of Computer Science and Network Security, VOL.8 No.9, September 2008 A Comparative Study of Neural-Network & Fuzzy Time Series Forecasting Techniques Case Study: Wheat

More information

Acceleration of Levenberg-Marquardt method training of chaotic systems fuzzy modeling

Acceleration of Levenberg-Marquardt method training of chaotic systems fuzzy modeling ISSN 746-7233, England, UK World Journal of Modelling and Simulation Vol. 3 (2007) No. 4, pp. 289-298 Acceleration of Levenberg-Marquardt method training of chaotic systems fuzzy modeling Yuhui Wang, Qingxian

More information

Lecture «Robot Dynamics»: Dynamics 2

Lecture «Robot Dynamics»: Dynamics 2 Lecture «Robot Dynamics»: Dynamics 2 151-0851-00 V lecture: CAB G11 Tuesday 10:15 12:00, every week exercise: HG E1.2 Wednesday 8:15 10:00, according to schedule (about every 2nd week) office hour: LEE

More information

Lecture Note 12: Dynamics of Open Chains: Lagrangian Formulation

Lecture Note 12: Dynamics of Open Chains: Lagrangian Formulation ECE5463: Introduction to Robotics Lecture Note 12: Dynamics of Open Chains: Lagrangian Formulation Prof. Wei Zhang Department of Electrical and Computer Engineering Ohio State University Columbus, Ohio,

More information

DOUBLE ARM JUGGLING SYSTEM Progress Presentation ECSE-4962 Control Systems Design

DOUBLE ARM JUGGLING SYSTEM Progress Presentation ECSE-4962 Control Systems Design DOUBLE ARM JUGGLING SYSTEM Progress Presentation ECSE-4962 Control Systems Design Group Members: John Kua Trinell Ball Linda Rivera Introduction Where are we? Bulk of Design and Build Complete Testing

More information

Adaptive Neuro-Sliding Mode Control of PUMA 560 Robot Manipulator

Adaptive Neuro-Sliding Mode Control of PUMA 560 Robot Manipulator Journal of Automation, Mobile Robotics & Intelligent Systems VOLUME 1, N 4 216 Adaptive Neuro-Sliding Mode Control of PUMA 56 Robot Manipulator Submitted: 28 th June 216; accepted: 7 th October 216 Ali

More information

Minimax Differential Dynamic Programming: An Application to Robust Biped Walking

Minimax Differential Dynamic Programming: An Application to Robust Biped Walking Minimax Differential Dynamic Programming: An Application to Robust Biped Walking Jun Morimoto Human Information Science Labs, Department 3, ATR International Keihanna Science City, Kyoto, JAPAN, 619-0288

More information

ARTIFICIAL NEURAL NETWORK PART I HANIEH BORHANAZAD

ARTIFICIAL NEURAL NETWORK PART I HANIEH BORHANAZAD ARTIFICIAL NEURAL NETWORK PART I HANIEH BORHANAZAD WHAT IS A NEURAL NETWORK? The simplest definition of a neural network, more properly referred to as an 'artificial' neural network (ANN), is provided

More information

Robot Dynamics II: Trajectories & Motion

Robot Dynamics II: Trajectories & Motion Robot Dynamics II: Trajectories & Motion Are We There Yet? METR 4202: Advanced Control & Robotics Dr Surya Singh Lecture # 5 August 23, 2013 metr4202@itee.uq.edu.au http://itee.uq.edu.au/~metr4202/ 2013

More information

Hover Control for Helicopter Using Neural Network-Based Model Reference Adaptive Controller

Hover Control for Helicopter Using Neural Network-Based Model Reference Adaptive Controller Vol.13 No.1, 217 مجلد 13 العدد 217 1 Hover Control for Helicopter Using Neural Network-Based Model Reference Adaptive Controller Abdul-Basset A. Al-Hussein Electrical Engineering Department Basrah University

More information

International Journal of Advanced Research in Computer Science and Software Engineering

International Journal of Advanced Research in Computer Science and Software Engineering Volume 3, Issue 4, April 2013 ISSN: 2277 128X International Journal of Advanced Research in Computer Science and Software Engineering Research Paper Available online at: www.ijarcsse.com Application of

More information

ECE Introduction to Artificial Neural Network and Fuzzy Systems

ECE Introduction to Artificial Neural Network and Fuzzy Systems ECE 39 - Introduction to Artificial Neural Network and Fuzzy Systems Wavelet Neural Network control of two Continuous Stirred Tank Reactors in Series using MATLAB Tariq Ahamed Abstract. With the rapid

More information

Machine Learning for Large-Scale Data Analysis and Decision Making A. Neural Networks Week #6

Machine Learning for Large-Scale Data Analysis and Decision Making A. Neural Networks Week #6 Machine Learning for Large-Scale Data Analysis and Decision Making 80-629-17A Neural Networks Week #6 Today Neural Networks A. Modeling B. Fitting C. Deep neural networks Today s material is (adapted)

More information

Trajectory Planning, Setpoint Generation and Feedforward for Motion Systems

Trajectory Planning, Setpoint Generation and Feedforward for Motion Systems 2 Trajectory Planning, Setpoint Generation and Feedforward for Motion Systems Paul Lambrechts Digital Motion Control (4K4), 23 Faculty of Mechanical Engineering, Control Systems Technology Group /42 2

More information

EE04 804(B) Soft Computing Ver. 1.2 Class 2. Neural Networks - I Feb 23, Sasidharan Sreedharan

EE04 804(B) Soft Computing Ver. 1.2 Class 2. Neural Networks - I Feb 23, Sasidharan Sreedharan EE04 804(B) Soft Computing Ver. 1.2 Class 2. Neural Networks - I Feb 23, 2012 Sasidharan Sreedharan www.sasidharan.webs.com 3/1/2012 1 Syllabus Artificial Intelligence Systems- Neural Networks, fuzzy logic,

More information

Mechanical Engineering Department - University of São Paulo at São Carlos, São Carlos, SP, , Brazil

Mechanical Engineering Department - University of São Paulo at São Carlos, São Carlos, SP, , Brazil MIXED MODEL BASED/FUZZY ADAPTIVE ROBUST CONTROLLER WITH H CRITERION APPLIED TO FREE-FLOATING SPACE MANIPULATORS Tatiana FPAT Pazelli, Roberto S Inoue, Adriano AG Siqueira, Marco H Terra Electrical Engineering

More information

Online Learning in High Dimensions. LWPR and it s application

Online Learning in High Dimensions. LWPR and it s application Lecture 9 LWPR Online Learning in High Dimensions Contents: LWPR and it s application Sethu Vijayakumar, Aaron D'Souza and Stefan Schaal, Incremental Online Learning in High Dimensions, Neural Computation,

More information

Artificial Neural Network and Fuzzy Logic

Artificial Neural Network and Fuzzy Logic Artificial Neural Network and Fuzzy Logic 1 Syllabus 2 Syllabus 3 Books 1. Artificial Neural Networks by B. Yagnanarayan, PHI - (Cover Topologies part of unit 1 and All part of Unit 2) 2. Neural Networks

More information

Motion Control of a Robot Manipulator in Free Space Based on Model Predictive Control

Motion Control of a Robot Manipulator in Free Space Based on Model Predictive Control Motion Control of a Robot Manipulator in Free Space Based on Model Predictive Control Vincent Duchaine, Samuel Bouchard and Clément Gosselin Université Laval Canada 7 1. Introduction The majority of existing

More information

Hybrid Direct Neural Network Controller With Linear Feedback Compensator

Hybrid Direct Neural Network Controller With Linear Feedback Compensator Hybrid Direct Neural Network Controller With Linear Feedback Compensator Dr.Sadhana K. Chidrawar 1, Dr. Balasaheb M. Patre 2 1 Dean, Matoshree Engineering, Nanded (MS) 431 602 E-mail: sadhana_kc@rediff.com

More information

STATE GENERALIZATION WITH SUPPORT VECTOR MACHINES IN REINFORCEMENT LEARNING. Ryo Goto, Toshihiro Matsui and Hiroshi Matsuo

STATE GENERALIZATION WITH SUPPORT VECTOR MACHINES IN REINFORCEMENT LEARNING. Ryo Goto, Toshihiro Matsui and Hiroshi Matsuo STATE GENERALIZATION WITH SUPPORT VECTOR MACHINES IN REINFORCEMENT LEARNING Ryo Goto, Toshihiro Matsui and Hiroshi Matsuo Department of Electrical and Computer Engineering, Nagoya Institute of Technology

More information