Hassen Mekki

A robust adaptive control using neural network

Feedback linearisation is an approach applied to non-linear control design and has attracted a great deal of research interest in recent years. The common assumptions are that the full state is measurable, and that the system is exactly linearly parameterised and feedback linearisable (input/state or input/output). With few exceptions, the robustness issue is not addressed. In practical implementation of exactly linearising control laws, the chief drawback is that they are based on exact cancellation of non-linear terms. If there is any uncertainty in the knowledge of the non-linear functions, the cancellation is not exact and the resulting inputoutput equation is not linear. The aim of this paper discusses the use of Neural Network (NN)-based adaptive control to get asymptotically exact cancellation.

VARIABLE STRUCTURE NEURAL NETWORKS FOR ADAPTIVE ROBUST CONTROL USING EVOLUTIONARY ARTIFICIAL POTENTIAL FIELDS

A novel neural network architecture, is proposed and shown to be useful in approximating the unknown nonlinearities of dynamical systems. In the variable structure neural network, the number of basis functions can be either increased or decreased with time according to specified design strategies so that the network will not overfit or underfit the data set. Based on the Gaussian radial basis function (GRBF) variable neural network, an adaptive state feedback controller is presented. The location of the centers of the GRBFs is analyzed using a new method inspired from evolutionary artificial potential fields method combined with a pruning algorithm. Using this method we can guarantee a minimal number of neuron. It is in noted, that both the recruitment and the pruning is made by a single neuron. Consequently, the recruitment phase does not perturb the network and the pruning does not provoke an oscillation of the output response. The weights of neural network are adapted using a Lyapunov approach. Moreover, the stability of the system can be analyzed and guaranteed by introducing the supervisory controller and modified adaptation law with projection.

Path planning for 3D visual servoing: For a wheeled mobile robot

In this paper, we are interested in 3D visual servoïng path planning and path tracking. In fact, in the 3D visual servoïng task, there is no control in the image space and the object may get out of the camera field of view during servoing. To solve this problem, we have used a new approach based on a flatness concept. The 3D visual servoïng suffer from another major problem, is to determine the relative pose of the camera and the object. Generally, the pose estimation is made by correspondences between points of one image and points of the space that is the 2D-3D correspondence. In our work we have used a 3D visual sensor called Kinect. To show the efficiency of the proposed algorithm, we have implemented it on a wheeled Koala robot.

Stochastic approximation based adaptive neural control for a class of nonlinear systems

This article presents an adaptive multilayer neural network-based controller that feedback-linearizes the system for a class of single-input single-output (SISO) and multi-input multi-output (MIMO) continuous-time nonlinear systems. Control action is used to achieve tracking performances for state-feedback linearizable unknown nonlinear system. The control structure consists of a feedback linearization portion provided by neural networks (NN). In the standard problem of feedback-based control, the cost to minimize is a function of the output derivatives. When the cost function depends on the output error, the gradient method cannot be applied to adjust the neural network parameters. In this context, the stochastic approximation approach allows computation of the cost function derivatives. In order to show the feasibility and performance of this control scheme, two applications are chosen as nonlinear case studies.

KUKA robot control based Kinect image analysis

In this work, we have investigated the processes required for visual extracting and the remote control of KUKA KR-125 industrial robot manipulator. For this purpose, the robot controller communicates with the external system via anEthernet cable IEEE 802.3. The exchanged data are transmitted thanks to TCP/IP Protocol. To do this, we performed a client/server application with all relevant motions control. Second, we set up a Kinect in the robotproximity for the detection of objects (recognition of form, determination of position etc...) and finally we applied it to a practical example: we have programmed the robot to be able to stack object thanks to the reliability of the visual processing.

Robust visual servoing using global features based on random process

This paper presents new approach illustrating robust visual servoing based on global visual features: random distribution of limited set of pixels luminance. Our approach aims to improve the real-time performance of the visual servoing scheme. In fact, the use of our new features reduces the computation time of the visual servoing task and removes matching and tracking process. Concerning the control scheme, we present new approach based on the second-order error-dynamics instead of the first-order error-dynamics. The main goal of this approach is to generate new control law able to improve mobile robot robustness with respect to kinematic modelling errors during visual servoing scheme. The new control law ensures the convergence of the mobile robot to its desired pose even in the presence of modelling errors. Experimental results are presented to validate our approaches and to demonstrate its efficiency.

Real-Time Visual Servoing Based on New Global Visual Features

This chapter proposes a new approach to achieve real-time visual servoing tasks. Our contribution consists in the definition of new global visual features as a random distribution of limited set of pixels luminance. The new method, based on a random process, reduces the computation time of the visual servoing scheme and removes matching and tracking process. Experimental results validate the proposed approach and show its robustness regarding to the image content.

Face recognition system using bag of features and multi-class SVM for robot applications

Face recognition system is used for the identification and verification of a face from a video or digital image. In the first phase, Viola Jones algorithm is used to detect and crop face region automatically from image/video frame. The second phase is to recognize the face of a person, in our proposed method Bag of Word technique used to extract features from an image which uses SURF for interest point selection, the proposed technique is tested on different state of the art face recognition databases and found more accurate and fast for face recognition. Our proposed method recognizes more than two person faces, multi-class support vector machine is used to classify the face image and assign a class label based on the learning of the classifier. The algorithm for the detection and recognition of faces is implemented in MATLAB application, and achieves a high accuracy rate of 99.21, which will be tested on the mobile robot.

Improving Mobile Robot Robustness in Visual Servoing Application

This paper proposes a novel method to improve mobile robot robustness with respect to kinematic modeling errors during visual servoing task. Instead of using first-order error-dynamics, as it is usually done, we use the second-order error-dynamics leading to a new control law. The main aim of this approach is to guarantee a robust visual servoing scheme. In fact, the new control law ensures the convergence of the mobile robot to its desired pose even in the presence of modeling errors. Experimental results are presented to validate our approach and to demonstrate its efficiency.

2D visual servoïng of wheeled mobile robot by neural networs

We are interested in this paper in the 2D visual servoïng for a mobile robot type Koala using radial basis function (RBF) neural network (NN). Seen that the interaction matrix, expressing the relationship between the camera motion and the consequent changes on the visual features, contains parameters to be estimated (depth) and requires a calibration phase of the camera. In more, the model of the robot can contain uncertainties engendered the movement with sliding. An online identification, using NN was proposed to overcome these problems. The RBF NN is used to estimate the block formed by the interaction matrix and the model inverts of the robot. The considered images are described by objects given by four points. Seen that the variables number of the estimated function is important, what can cause a problem of the use of an excessive number of RBFs. As remedy, we used a new approach consists in considering that a single point is sufficient to solve the problem of the 2D visual servoïng of the mobile robot.