D. Nguyen and B. Widrow, "Neural Networks for Self-Learning Control Systems," IEEE Cont. Sys., vol. 10, no. 4, pp.18-23, April, 1990.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....based approaches or backpropagation type solutions for dynamical systems involving neural networks. Among the many modelbased approaches, the N stage optimal control problem (Bryson Ho, 1969) has been studied in combination with several structures for the neural controller. For example, Nguyen Widrow (1990) applied backpropagation to the problem of backing up a trailer truck based on a neural network emulator for the plant. In Saerens et al. 1993) a full static state feedback controller has been studied in the context of optimal control, thereby relating the Lagrange multiplier sequence to the ....

Nguyen D. & Widrow B. (1990). Neural networks for self-learning control systems, IEEE Control Systems Magazine, 10(3), pp.18-23.

....[52] covered only the case when the recurrent network behavior relaxes to a fixed point. However, if a general temporal processing is needed, two main gradient based learning approaches exist for recurrent networks [24] 25] 36] Backpropagation through time (BPTT) 1] 5] 7] 24] 36] [46] and real time recurrent learning (RTRL) 2] 23] 24] 27] 28] 36] 38] 49] 56] The algorithm in [23] is an hybrid method. BPTT is a family of algorithms which extends the BP paradigm to dynamic networks. There are two main points of view to understand the BPTT algorithm. The first ....

D. H. Nguyen and B. Widrow, "Neural networks for self-learning control systems," IEEE Contr. Syst. Mag., pp. 18--23, Apr. 1990.

....the plant [2] A first illustration of a modelbased neural control strategy is Nguyen Widrow s backing up of a trailer truck. In order to be able to apply backpropagation for training the neural controller, a neural network model is trained first in order to emulate the trailer truck dynamics [16]. Among the many neural control strategies existing now are direct and indirect neural adaptive control, neural optimal control, reinforcement learning, model predictive control, internal model control, feedback linearization etc. Moreover a stability theory for multilayer recurrent neural ....

Nguyen D., Widrow B. (1990). Neural networks for self-learning control systems, IEEE Control Systems Magazine, 10(3), pp.18-23.

....error and control effort can be obtained. Since the error signals of back propagation can propagate through different time stages, this control methodology is called temporal back propagation, or simply TBP . As a matter of fact, the basic idea of TBP is similar to Nguyen and Widrow s approach [12] to construct a self learning 9 trajectory desired trajectory actual conditions initial measure error FC plant FC plant FC plant state state state state 0 1 2 m 1 parameter set SAN 0 SAN 1 m 1 SAN Figure 5: A trajectory adaptive network for control application. neural ....

D. H. Nguyen and B. Widrow. Neural networks for self-learning control systems. IEEE Control Systems Magazine, pages 18--23, April 1990.

....this chapter, the motivation behind this dissertation is discussed and the related literature is reviewed. 1.1 Motivation 1.1. 1 Neural network based models Artificial neural networks (NNs) are widely applied to a variety of practical problems [52, 96] The areas of applications include control [75, 80, 83, 108], signal processing [61] pattern recognition [71, 73] forecasting [106] modeling chemical processes [11, 62, 78] and manufacturing processes [3, 104] Many success stories have been reported. However, many concerns about the construction and the use of NNbased models have also been raised. ....

D. H. Nguyen and B. Widrow. Neural networks for self-learning control systems. IEEE Control Systems Magazine, pages 18--23, Apr. 1990.

.... Time (BPTT) algorithm. BPTT is an extension of the gradient based static optimization of an universal approximator whenever this latter is used in a dynamical context i.e. with the error computed over a given temporal horizon. This algorithm has been derived and used by Werbos (1990) Nguyen Widrow (1990), Parisini Zoppoli (1991) Plumer (1993) and Suykens, De Moor Vandewalle (1994) for the automatic tuning of neurocontroller aiming at minimizing a control cost either given over a certain temporal horizon or delayed in time. As the sections will show, the easiest and cleanest way to derive the ....

....The purpose here is to stabilize the process around x=0 with minimal cost on the control actions. We assume a statefeedback class of controller with the control law given by u(n) U[x(n) x0] where U is a differentiable vector function. Similar problems have been addressed by Werbos (1990) Nguyen Widrow (1990), Parisini Zoppoli (1991) Plumer (1993) Suykens, De Moor Vandewalle (1994) and Saerens, Renders Bersini (1995) The purpose of the learning algorithm is to train fuzzy controllers to provide the optimal control law for all initial conditions in a given operating region u(n) N[x(n) x0;w] ....

[Article contains additional citation context not shown here]

Nguyen D. & Widrow B., 1990, Neural networks for selflearning control systems. IEEE Control Systems Magazine, 10 (3), pp. 18-23.

....adaptive non linear approximation. When gradient based optimisation algorithms are used, derivatives are required which entails in turn the need for an action derivable cost and, in the case of model based control, the knowledge of the process Jacobians. Back Propagation Through Time (BPTT) 4] 5][12][13] 17] 19] 21] algorithm is the temporal extension of backpropagation algorithm when the error to minimize is, like it is indeed the case for optimal control, given over a temporal horizon or delayed in time. A clean way to derive BPTT is by introducing Lagrange multipliers and performing a ....

....the analytical equations. Besides all the Jacobians can be approximated by their sign following a further simplification strategy already discussed in [17] In their parking of the truck backer upper, Nguyen and Widrow used a second neural net for emulating the process and deriving the Jacobians [12]. In [5] we have shown that BPTT, when taken in its simplified version, works better and faster than the original one so that, in this paper, we will restrict the use of BPTT to its simplified version. A last important assumption of our problem is that we consider the Rendez Vous to occur after ....

Nguyen D. & Widrow B., 1990, Neural networks for self-learning control systems.IEEE Control Systems Magazine, 10 (3), pp. 18-23.

....from R 2m to R n . The control system architecture for this control strategy is shown in Fig.4.1 and the corresponding simulation results are given in Fig.4.2. 5. SELF LEARNING CONTROL (SLC) In this section, the self learning control scheme which was originally proposed by Nguyen and Widrow [3] is elaborated upon. The method differs from the other methods in that this scheme is goal directed. The controller is trained to keep the plant output at a desired and previously defined level. The approach utilizes the well known backpropagation method in the training of the controller. The ....

....point in the weight updating is the succession that appears in the controller training structure. In fact, the weight change at the K th stage influences the (K 1) th stage. This obviously requires the saving of each individual weight change evaluated at the corresponding stage. Nguyen and Widrow [3], in their work, say that the weight changes could be applied immediately as they are evaluated because they are the accumulated effects that improve the performance of the controller and this does not affect the final performance significantly. Having trained the neural controller, it is ....

. Nguyen, D. H., Widrow, B., "Neural Networks for Self-Learning Control Systems," IEEE Control Systems Magazine, pp. 18-23, April 1990.

....value (a success or failure signal) The trailer backing problem is gaining attention as a simple to understand yet difficult to solve learning control problem. Approaches such as the Cerebellar Model Articulated Controller (CMAC) 25] adaptive fuzzy systems [15] backpropagation through time [18], and fuzzy BOXES [28] have all been applied to versions of this problem. 3.2 Our Solution Kohonen [14] has proposed a physiologically plausible method of cooperative and competitive organization for connectionist systems that allows them to self organize around a set of input vectors. Several ....

D. Nguyen and B. Widrow. Neural networks for self-learning control systems. IEEE Control Systems Magazine, 10(3):18--23, 1990.

....as well as success. 5 Related work The trailer backing problem is gaining attention as a simple to understand yet difficult to solve learningcontrol problem. Approaches such as the Cerebellar Model Articulated Controller (CMAC) 8] adaptive fuzzy systems [5] backpropagation through time [6], and fuzzy BOXES [11] have all been applied to versions of this problem. It is difficult to directly compare results across many of these systems and with our results. The most obvious difference between this study and those of most other authors writing on this problem, is that our system was ....

D. Nguyen and B. Widrow. Neural networks for self-learning control systems. IEEE Control Systems Magazine, 10(3):18--23, 1990.

....into a stable state. 4 Related work The trailer backing problem is gaining attention as a simple to understand yet difficult to solve learningcontrol problem. Approaches such as the Cerebellar Model Articulated Controller (CMAC) 18] adaptive fuzzy systems [9, 6] backpropagation through time [14], fuzzy BOXES [20] genetic algorithms [10] and our own ROLNNET [5] have all been applied to versions of this problem. It is difficult to directly compare 0 0.2 0.4 0.6 0.8 1 0 200 400 600 800 1000 Graph of case 1. Average performance for 100 runs of 1000 trials each. Initial position: ....

D. Nguyen and B. Widrow. Neural networks for selflearning control systems. IEEE Control Systems Magazine, 10(3):18--23, 1990.

....systems have recently received much attention, but most of the work has concentrated on simulated systems. Approaches such as the Cerebellar Model Articulated Controller [ Shelton and Peterson, 1992 ] adaptive fuzzy systems [ Kong and Kosko, 1992 ] backpropagation through time [ Nguyen and Widrow, 1990 ] and fuzzy BOXES [ Woodcock et al. 1991 ] have all been applied to versions of the trailer backing problem. Our method generally learns faster or with less supervision than these other methods while remaining practical for implementation with minimal computing hardware. Our method combines ....

Nguyen, D. and B. Widrow: 1990, `Neural networks for self-learning control systems'. IEEE Control Systems Magazine 10(3), 18-23.

.... features like planification (geometric approach) Can87, CL90, FT87, Lat91, LTJ90] fuzzy and expert systems (rule based approach with structured information) HLG91, OMO 91, vdRvNLD90] neural nets (coding the knowledge into a black box with learning abilities) Kos92, Lee91, Mel90, NSA90, NW90a, NW90b, Wil93] Of course, conventional solutions have a definite advantage: there are theorems which state clearly when the solution exists and when it does not exist. But they are either implicit solutions or awkward solutions in the sense that they involve complex computations, or are very ....

.... new methods have been developed: planification (geometric approach) fuzzy and expert systems (rule based approach with structured information) neural nets (coding the knowledge into a black box with learning abilities) AM92, Can87, CL90, FT87, Lat91, Lau90, LTJ90, Lee91, Mel90, NSA90, NW90a, NW90b, vdRvNLD90] 4 Learning to control a system If we see the control of a system as a skill, we can distinguish two different parts: acquisition of the skill and refinement of the skill. These may be related to more classical notions such as system identification and control (acquisition) ....

D. Nguyen and B. Widrow. Neural networks for self-learning control systems. IEEE Control Systems Magazine, 10-3:18--23, 1990.

....the output of the i Gamma 1 th layer as a subset) networks with local feedback interconnections. 3.1. 1 Truck backer upper The next section deals with the truck backer upper , a neural network controller steering a trailer truck while backing up (no forward movement allowed) to a loading dock [NW90] For this application, the neurons used are Adalines and they are connected together to form a two layer feedforward network. An Adaline has n 1 inputs (the n inputs of the network and one additional input set to 1 which adds a constant bias to the weighted sum) and a single output. The output ....

D. Nguyen and B. Widrow. Neural networks for self-learning control systems. IEEE Control Systems Magazine, 10-3:18--23, 1990.

....supervised control , direct inverse control , neural adaptive control and back propagation of utility) of neural networks in control, as proposed by Werbos [58, 9] are also directly applicable schemes for ANFIS. Particularly we have employed a similar method of the back propagation through time [32] or unfolding in time to achieve a self learning fuzzy controller with four rules that can balance an inverted pendulum in an near optimal manner [12] It is 38 expected that the advances of neural network techniques in control can promote those of ANFIS as well, and vice versa. The active role ....

D. H. Nguyen and B. Widrow. Neural networks for self-learning control systems. IEEE Control Systems Magazine, pages 18--23, April 1990. 41

....model criterion J gradient rJ possible state of system SET of EXAMPLES control backpropagation e u y Our learning structure is very simple. The on line structure allows to refine the neural controller in front of real behavior of the robot. In comparison, D. H. Nguyen and B. Widrow proposed in [3] to use a network to emulate the system. Figure 2: Learning off line with criterion neural network criterion J gradient rJ measured state real robot state current control backpropagation e(t) u(t) y(t 1) Then, they can backpropagate a criterion through the emulator net and trained the controller ....

D. Nguyen, B. Widrow "Neural Networks for Self-Learning Control Systems",IEEE Work. on Industrial Applications of Neural Networks, pp 18-23,1991

....descent weight update by Fig. 4. Learning on line with criterion (On line undirect backpropagation) The backpropagation of the criterion gradient permits the network to learn with the real system. This is the difference between our method and the one proposed by D. H. Nguyen and B. Widrow in [7]. They use a network to emulate the system. Then, they can backpropagate a criterion through the emulator net and trained the controller without a gradient. Nethertheless, the real robot cannot be used in their method because the criterion cannot be propagated throught it. Consequently, they ....

D. H. Nguyen, B. Widrow "Neural Networks for Self-Learning Control Systems", IEEE Workshop on Industrial Applications of Neural Networks, pp 18-23, Ascona Ticino,1991

....added until the trailer was placed at a distance of 6 scale feet from the target and at an angle of Gamma45 ffi to 45 ffi and the hitch angle was set from Gamma20 ffi to 20 ffi . These increasingly difficult lessons are consistent with the training schemes used by other authors (e.g. [17]) After a total of 1000 training trials, the learned responses were tested in simulation and on TBMin. A sample simulation run after training is depicted in Figures 8, 9, and 10. The angle of the trailer and of the hitch are depicted graphically for a sample simulation run in Figure 11 and for a ....

....systems (e.g. 1] The trailer backing problem has not been studied as long or as extensively as the polebalancing problem, but it too is becoming widely studied. Approaches such as the Cerebellar Model Articulated Controller (CMAC) 21] adaptive fuzzy systems [13] backpropagation through time [16, 17], and fuzzy BOXES [26] have all been applied to this problem. 7.1 Comparison with related work While many researchers have studied the pole balancing or trailer backing problems (or both, e.g. 26] it is very difficult to directly compare results across many of these systems and we find it ....

D. Nguyen and B. Widrow. Neural networks for self-learning control systems. IEEE Control Systems Magazine, 10(3):18--23, 1990.

....X k=0 ku(k)k 2 ; where u(k) is the control action at time step k. By a proper selection of , a compromise between trajectory error and control efforts can be obtained. Use of back propagation through time to train a neural network for backing up a tractor trailer system is reported in [69]. The same technique was used to design an ANFIS controller for balancing an inverted pendulum [28] Note that back propagation through time is usually an off line learning algorithms in the sense that the parameters will not be updated till the sequence (k = 1 to m) is over. If the sequence is ....

D. H. Nguyen and B. Widrow. Neural networks for self-learning control systems. IEEE Control Systems Magazine, pages 18-- 23, April 1990.

....a goal. The car is penalized for hitting the edge of the track and failing to reach the goal. A feature of the problem is that it can have a large number of states. Nguyen and Widrow used neural networks to learn to reverse a simulated articulated lorry into a loading bay (Nguyen and Widrow 1989, Nguyen and Widrow 1990). Thrun uses a robot navigation task to illustrate the trade off between exploration and exploitation in machine learning (Thrun 1992) Tham and Prager have approached controlling a simulated robot arm as a reinforcement learning task (Tham and Prager 1992) The arm is penalized for collisions by ....

....and its derivatives are transformed into an error suitable for adapting the networks parameters. 4.1. 4 Back propagation through time Nguyen and Widrow used two neural networks in a control scheme that learned to reverse a simulated articulated lorry into a loading bay (Nguyen and Widrow 1989, Nguyen and Widrow 1990). Figure 4.2 shows the task. One network formed a model of the lorry, another mapped the lorry s state into a control action (the angle of steer of the cab) Figure 4.2: The articulated lorry and loading bay. The networks were trained using a technique known as back propagation through time (BTT) ....

Nguyen, D. and Widrow, B. (1990). Neural networks for self-learning control systems, IEEE Control Systems Magazine pp. 18--23.

.... example simple supervised control [1] 19] 22] system identification [11] 35] and inverse system identification [18] 20] 23] In the realm of robot control, several authors have utilized neural networks to train robots to reach targets or follow prescribed trajectories [4] 8] 24] [26], 30] 34] 36] Neural networks are appealing to many researchers because their parallel distributed nature tends to make them highly redundant and thus fault tolerant. 0 Affiliations: P. Gaudiano (E mail: gaudiano cns.bu.edu; WEB: http: cnsweb. bu.edu) is at Boston University, Dept. of ....

D. Nguyen and B. Widrow, "Neural networks for self-learning control systems," IEEE Control Systems Magazine, vol. 10, no. 3, pp. 18--23, 1990.

.... example simple supervised control [1] 22] 25] system identification [11] 37] and inverse system identification [21] 23] 26] In the realm of robot control, several authors have utilized neural networks to train robots to reach targets or follow prescribed trajectories [4] 8] 27] [29], 32] 36] 38] Neural networks are appealing to many researchers because their parallel distributed nature tends to make them highly redundant and thus fault tolerant. The relative success of classical control theory has made it possible for many of these research endeavors to focus on ....

D. Nguyen and B. Widrow, "Neural networks for self-learning control systems," IEEE Control Systems Magazine, vol. 10, pp. 18--23, 1990.

....the plant derivative, efficient neuro control learning has been obtained (Zhang et al. 1995) Indeed this method can not be generalised to many systems. Back propagation through time neuro control learning To avoid the problem of the specialised inverse control learning a number of researchers (Nguyen and Widrow, 1990; Narendra and Parthasarathy, 1990; Jordan and Rumelhart, 1992) have independently developed the back propagation through time (Jordan and Rumelhart, 1992) neuro control leaning method. The problem of the specialised inverse control learning is that the performance error e y = r Gamma y is not ....

....neuro control leaning method. The problem of the specialised inverse control learning is that the performance error e y = r Gamma y is not reliable because it is not directly related to the neuro controller output u. A reliable error for the neuro control learning is e u = u Gammau. Therefore (Nguyen and Widrow, 1990; Narendra and Parthasarathy, 1990; Jordan and Rumelhart, 1992) propose to emulate e u by using the back propagated performance e y through a forward model of the system (see figure 0.8) This way the error e y , back propagated toward the input layer of the feed forward model, should correspond ....

Nguyen, D. H. and B. Widrow (1990). Neural Networks for Self-learning Control Systems. IEEE Control Systems Magazine 10, 18--23.

No context found.

D. Nguyen and B. Widrow, "Neural Networks for Self-Learning Control Systems," IEEE Cont. Sys., vol. 10, no. 4, pp.18-23, April, 1990.

No context found.

Nguyen D., Widrow B., "Neural networks for self-learning control systems," IEEE Control Systems Magazine, 10(3), pp.18-23, 1990.

First 50 documents

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC