Gordon Wells

Optimal control of a water distribution network in a supervisory control system

This paper deals with the use of optimal control techniques in water distribution networks. An optimal control tool, developed in the context of a European research project is described and the application to the city of Sintra (Portugal) is presented. ( 2000 Elsevier Science Ltd. All rights reserved.

Vision-based robot positioning using neural networks

Abstract Most vision-based robot positioning techniques rely on analytical formulations of the relationship between the robot pose and the projected image coordinates of several geometric features of the observed scene. This usually requires that several simple features such as points, lines or circles be visible in the image, which must either be unoccluded in multiple views or else part of a 3D model. Featurematching algorithms, camera calibration, models of the camera geometry and object feature relationships are also necessary for pose determination. These steps are often computationally intensive and error-prone, and the complexity of the resulting formulations often limits the number of controllable degrees of freedom. We provide a comparative survey of existing visual robot positioning methods, and present a new technique based on neural learning and global image descriptors which overcomes many of these limitations. A feedforward neural network is used to learn the complex implicit relationship between the pose displacements of a 6-dof robot and the observed variations in global descriptors of the image, such as geometric moments and Fourier descriptors. The trained network may then be used to move the robot from arbitrary initial positions to a desired pose with respect to the observed scene. The method is shown to be capable of positioning an industrial robot with respect to a variety of complex objects with an acceptable precision for an industrial inspection application, and could be useful in other real-world tasks such as grasping, assembly and navigation.

Dynamic control of a robot arm using CMAC neural networks

Abstract Neural identification and control techniques are well-suited to the problem of controlling robot dynamics. This paper describes the use of CMAC networks for the adaptive dynamic control of an orange-harvesting robot. Among the various neural-network paradigms available, the CMAC model was chosen in this case because of its fast convergence and on-line adaptation capability. The solution of this dynamic control problem with CMAC is an encouraging demonstration of “experience-based”, as opposed to model-based, control techniques and is a good example of the use of on-line learning in adaptive neural control.

Assessing Image Features for Vision-Based Robot Positioning

The development of any robotics application relying on visual information always raises the key question of what image features would be most informative about the motion to be performed. In this paper, we address this question in the context of visual robot positioning, where a neural network is used to learn the mapping between image features and robot movements, and global image descriptors are preferred to local geometric features. Using a statistical measure of variable interdependence called Mutual Information, subsets of image features most relevant for determining pose variations along each of the six degrees of freedom (dofs) of camera motion are selected. Four families of global features are considered: geometric moments, eigenfeatures, Local Feature Analysis vectors, and a novel feature called Pose-Image Covariance vectors. The experimental results described show the quantitative and qualitative benefits of performing this feature selection prior to training the neural network: Less network inputs are needed, thus considerably shortening training times; the dofs that would yield larger errors can be determined beforehand, so that more informative features can be sought; the order of the features selected for each dof often accepts an intuitive explanation, which in turn helps to provide insights for devising features tailored to each dof.

Selection of image features for robot positioning using mutual information

The authors and Venaille (1996) developed a prototype for visual robot positioning, based on global image descriptors and neural networks. Now, a procedure to automatically select subsets of image features most relevant to determine pose variations along each of the six degrees of freedom (DOFs) has been incorporated into the prototype. This procedure is based on a statistical measure of variable interdependence, called mutual information. Three families of features are considered in this paper: geometric moments, eigenfeatures and pose-image covariance vectors. The experimental results described show the quantitative and qualitative benefits of carrying out this feature selection prior to training the neural network: fewer network inputs need to be considered, thus considerably shortening training times; the DOFs that would yield larger errors can be determined beforehand, so that more informative features can be looked for; the ordering of the features selected for each DOF often admits a very natural interpretation, which in turn helps to provide insights for devising features tailored to each DOF.

APPLICATION OF NEURAL NETWORKS TO IMAGE-BASED CONTROL OF ROBOT ARMS

Abstract This paper describes experiments with the use of neural networks for image-based positioning of industrial robots. Firstly, an improved neural “visual-servoing” method based on geometric image characteristics is developed and analyzed. Subsequently, a more flexible and robust scheme is shown, based on global descriptors of the image.

Neural Approaches to Robot Control: Four Representative Applications

This paper reviews neural network techniques for achieving adaptivity in both manipulator and mobile robots. It is structured in two parts. First, the different learning approaches are classified according to the amount of training information they require: quantitative (supervised approaches), qualitative (reinforcement-based approaches) and none (unsupervised approaches). Afterwards, the adequacy of each approach for solving specific problems in robot control is illustrated through four working industrial prototypes developed by the authors in the frame of two Esprit projects. The problems tackled are the inverse kinematics and inverse dynamics of robot manipulators, visual robot positioning and mobile robot navigation.

Neural networks for robot control

Abstract This paper describes the use of neural networks in diferent domains of robot control. Three robot control problems, relevant to a broad range of robotics applications, are analyzed, with a review of the state of the art and a description of current research by the authors, highlighting the advantages of the use of neural networks with respect to conventional techniques.

NEURAL NETWORKS FOR ROBOT CONTROL

This paper describes the use of neural networks in diferent domains of robot control. Three robot control problems, relevant to a broad range of robotics applications, are analyzed, with a review of the state of the art and a description of current research by the authors, highlighting the advantages of the use of neural networks with respect to conventional techniques.