Catálogo de publicaciones - libros

Precise and real-time segmentation of color images is a crucial aspect of many applications, where high accuracy and quick response of a system is required. There is a number of algorithms available, but they are either fast or accurate, but not both. This paper describes a modification to one of the fastest color segmentation methods based on constant thresholds. Our idea is to use variable thresholds and merge the results achieving higher precision of color segmentation, comparable with adaptive methods. Using variable thresholds requires multiple passes of the algorithm which is time consuming, but thanks to our modification the processing time can be reduced by up to 50%. The original algorithm as well as the proposed modification are described. The performance of the modified method is tested in a real-time visual servoing application.

Neural networks are a popular technique for learning the adaptive control of non-linear plants. When applied to the complex control of android robots, however, they suffer from serious limitations such as the moving target problem, i.e. the interference between old and newly learned knowledge. However, in order to achieve lifelong learning, it is important that robots are able to acquire new motor skills without forgetting previously learned ones. To overcome these problems, we propose a new framework for motor learning, which is based on consolidation. The framework contains a new rehearsal algorithm for retaining previously acquired knowledge and a growing neural network. In experiments, the framework was successfully applied to an artifical benchmark problem and a real-world android robot.

This piece of research introduces POOCA (Progressive Optimisation of Organised Colonies of Ants) as an appealing variant of the established ACO (Ant Colony Optimisation) algorithm. The novelty of POOCA lies on the combination of the co-operation inherent in ACO with the spread of activation around the winner node during SOM (Self-Organising Map) training. The principles and operation of POOCA are demonstrated on examples from robot navigation in unknown environments cluttered with obstacles: efficient navigation and obstacle avoidance are demonstrated via the construction of short and – at the same time - smooth paths (i.e. optimal, or near-optimal solutions); furthermore, path convergence is speedily accomplished with restricted numbers of ants in the colony. The aim of this presentation is to put forward the application of POOCA to combinatorial optimisation problems such as the traveling salesman, scheduling etc.

This paper proposes a novel algorithm for mobile robots to select a group leader and to be guided in order to perform a specific work. The concepts of mobile ad hoc network (MANET) and object entropy are adopted to design the selection of a group leader. A logical robot group is created based on the exchange of and messages in a robot communication group whose organization depends on a transmission range. A group leader is selected based on the transmission of message from a robot who initiates to make a logical robot group. The proposed algorithm has been verified based on the computer-based simulation. The performance metric such as the number of message in order to make a logical robot group and to select a group leader is defined and verified by using the computer-based simulation.

We present a new approach to path planning based on the properties of the minimum enclosing ball (MEB) in a reproducing kernel space. The algorithm is designed to find paths in high-dimensional continuous spaces and can be applied to robots with many degrees of freedom in static as well as dynamic environments. In the proposed method a sample of points from free space is enclosed in a kernel space MEB. In this way the interior of the MEB becomes a representation of free space in kernel space. If both start and goal positions are interior points in the MEB a collision-free path is given by the line, contained in the MEB, connecting them. The points in work space that satisfy the implicit conditions for that line in kernel space define the desired path. The proposed algorithm was experimentally tested on a workspace cluttered with random and non random distributed obstacles. With very little computational effort, in all cases, a satisfactory free collision path could be calculated.

We describe a method for localizing a mobile robot in its working environment using a vision system and Virtual Reality Modeling Language (VRML). The robot identifies the landmarks located in the environment, using image processing and neural network pattern matching techniques, and then it performs self-positioning based on vision information and a well-known localization algorithm. The correction of position error is performed using the 2-D scene of the vision and the overlay with the VRML scene. Through an experiment, the self-positioning algorithm has been implemented to a prototype robot and also it performed autonomous path tracking.

This paper describes a neural network control for a visual guidance system of a mobile robot to follow a guideline. Without complicated geometric reasoning from the image of a guideline to the robot-centered representation of a bird’s eye view in conventional studies, the proposed system transfers the input of image information into the output of a steering angle directly. The neural network controller replaces the nonlinear relation of image information to a steering angle of robot on the real ground. For image information, the feature points of guideline are extracted from a camera image. In a straight and curved guideline, the driving performances by the proposed technology are measured in simulation and experimental test.

A new notion of cone-realization for discrete-time linear systems with delays is proposed. Necessary and sufficient conditions for the existence of cone-realizations of discrete-time linear systems with delays are established. A procedure is proposed for computation of a cone-realization for a given proper rational matrix (z). It is shown that there exists a ()-cone realization of (z) if and only if there exists a positive realization of () = () where , and are non-singular matrices generating the cones , and respectively.

This paper deals with the problem of global stability for a class of neural networks with time-varying delays. A new sufficient condition for global stability is proposed by using some slack matrix variables to express the relationship between the system matrices. The restriction on the derivative of the delay function to be less than unit is removed. A numerical example shows that the result obtained in this paper improves the upper bound of the delay over some existing ones.

This paper presents a method to determine the initial rotor position for Direct Torque Controlled (DTC) Permanent Magnet Synchronous Motor (PMSM) using Artificial Neural Network (ANN).The inductance variation is a function of the rotor position and stator current for PMSM. A high frequency and low magnitude voltage is applied to the stator windings and examined the effects the stator currents by using ANN for initial rotor position detection.