Showing papers presented at "International Conference on Control, Automation, Robotics and Vision in 2000"

Evolutionary Design of Fuzzy Control Systems - A Hybrid Approach

Abstract: A control and actuator system for a press brake having a frame, a bed, a ram, and a pair of hydraulic cylinders for reciprocating the ram, utilizes a jackscrew arrangement in conjunction with positive mechanical stops on the ram pistons to support the ram beneath the cylinders to enable the bottom travel limit of the ram to be preset. The top travel limit of the ram is preset by means of vertically adjustable actuator rods on the ram, which engage actuator stems on valves associated with each cylinder to stop upward travel and hold the ram in position. Tilt compensation is provided at the top and bottom ram limits by independent adjustment of the jackscrews and actuator rods, obviating the need for a complex tape and pulley driven differential valve arrangement. The novel hydraulic circuit provided for powering the cylinders utilizes pilot-driven control valves, and provides for direct venting of the system high-volume hydraulic pump when not in use to maximize system efficiency.

Designing Optimal Neuro-Fuzzy Architectures for Intelligent Control

Abstract: The integration of Artificial Neural Network (ANN) learning and Fuzzy Logic (FL) approximate reasoning in one architecture, to overcome individual limitations and achieve synergetic effects through hybridization of these techniques, has in recent years contributed to a large number of Neuro-Fuzzy (NF) architectures. NF techniques override the classical control methods in many aspects, such as algorithm simplicity, system robustness and the ability to handle imprecision and uncertainty. In this paper we present the state-of-art NF models that have evolved in the past few years. We further attempt to assess the strengths and weakness of each NF architecture and selection criteria for IC applications. Finally we present our vision of an optimal NF architecture and future research directions.

Merging maps in 2D structure and motion estimation

Abstract: We present a system for doing automatic surveying or structure and motion analysis given 1D images of a 2D surrounding. The system automatically identiies and tracks the image of new points and solves the structure and motion problem. One key feature of the system is the ability to hypothesize, test and incorporate simple constraints on the scene, e.g. that two conngurations of points are the same. This is essential when dealing with long image sequences (thousands of images). In this paper we develop and test the theory for automatic geometric reasoning. Ideas on hypothesis generation and testing are presented. It is also shown how to update the uncertainty representation of the database.

Comparing performances of supervised classifiers

Abstract: This paper describes a hierarchical architec- ture constructed using the neural gas algorithm. The learn- ing algorithm used is a mixed unsupervised/supervised method with most of the learning being unsupervised. The proposed architecture generates multiple classifications for every data sample presented, and these are registered as confidence values. This enables to employ different fusion techniques to combine the individual classifications pro- duced by the proposed architecture. Excellent recognition rates for two benchmark datasets are presented. In this approach, the combined system should take ad- vantage of the strengths of the individual classifiers and reduce their weaknesses in order to improve the overall classification accuracy. In this paper, we concentrate on the second approach for the decision combination problem. If the individual classifiers have different accuracies, performance of the combined system should be improved by giving the combiner the ability to bias the outputs based on a priori knowledge about the reliability of the individual classifiers. This fact has been demonstrated, for instance, in the technique of boosting (?). In this paper, we present a classifier based on the neural gas algo- rithm (12) with a hierarchical architecture (HONG) (13) whose individual classifiers are combined by different fusion techniques. The algorithm used is a mixed un- supervised/supervised method with most of the learning being unsupervised. Performance of the basic method is compared with that of other techniques on two well-known benchmark data sets.