- Instituto Tecnologico de Aeronautica
Division of Electronic Engineering
DCTA-ITA-IEE-IEES
12228-900-Sao Jose dos Campos-SP
Brazil
- BSc (UFU, Brazil, 1984), MSc (ITA, Brazil, 1988), PhD (UMIST, UK, 1994). See recent publications at: ftp://labattmot.... moreBSc (UFU, Brazil, 1984), MSc (ITA, Brazil, 1988), PhD (UMIST, UK, 1994).
See recent publications at:
ftp://labattmot.ele.ita.br/ele/cairo/My_Publications/
http://lattes.cnpq.br/0425874008159542edit
Este texto apresenta os fundamentos de Inteligência Artificial sob o ponto de vista de aplicações industriais, mormente no que concerne Controle e Automação de Processos. Busca condensar uma década de experiência dos autores neste campo e... more
Este texto apresenta os fundamentos de Inteligência Artificial sob o ponto de vista de aplicações industriais, mormente no que concerne Controle e Automação de Processos. Busca condensar uma década de experiência dos autores neste campo e incorpora diversos resultados obtidos pela equipe multidisciplinar do ITA. Este título foi co-editado pela FAPESP (Fundação da Amparo à Pesquisa do Estado de São Paulo).
... ai. A função de mérito g(x) pode, por sua vez, ser realizada por uma rede neural gRN(x,v), com pesos ajustáveis v. Seja k tal que zk ≥ zi, ∀ i ≠ k, ou seja, a ação ak possui o maior mérito no momento, dado por zk. O problema ...
The training of artificial neural networks can be seen as a hard optimization problem. Any algorithm used to solve this problem will have weak and strong features. In this article we consider the use of Asynchronous Teams to train... more
The training of artificial neural networks can be seen as a hard optimization problem. Any algorithm used to solve this problem will have weak and strong features. In this article we consider the use of Asynchronous Teams to train artificial neural networks. An Asynchronous Team is a general computational structure where several different algorithms run in parallel in different computers and are applied at the same time to solve the same optimization problem. During the computation, several intermediate solutions are analysed and used as new starting points for the different algorithms. We show that, when compared with the solutions obtained by each individual algorithm, the use of the Asynchronous Team (controlled "mixture" of different algorithms) leads to a better solution, that is, better trained artificial neural networks. As a simple example, an artificial neural network is trained to recognize a few simple characters.
Research Interests:
ABSTRACT This article shows how the problem of autonomous navigation of a small boat was formulated and solved. The boat is a catamaran equipped with two water wheels driven by DC motors. A look-up table controller is used to turn on and... more
ABSTRACT This article shows how the problem of autonomous navigation of a small boat was formulated and solved. The boat is a catamaran equipped with two water wheels driven by DC motors. A look-up table controller is used to turn on and off the DC motors. Firstly it is shown how the Kalman filter algorithm was applied to estimate in real-time the boat position and heading, using the measurements from a low-cost IMU (Inertial Measurement Unit), a standard GPS receiver and a digital compass. Then a mathematical model of the boat and simulation results for the sensor integration problem and for the boat controller, are discussed. Finally, the article shows how the proposed solution for the autonomous navigation problem was implemented and tested using an embedded computer and the sensors (IMU, GPS receptor and digital compass) aboard the boat.
Research Interests:
Research Interests:
Research Interests:
Research Interests:
ABSTRACT This article concerns the application of the Mixture of Local Expert Models (MLEM) to predict the daily and monthly price of the Sugar No. 14 contract in the New York Board of Trade. This technique can be seen as a forecasting... more
ABSTRACT This article concerns the application of the Mixture of Local Expert Models (MLEM) to predict the daily and monthly price of the Sugar No. 14 contract in the New York Board of Trade. This technique can be seen as a forecasting method that performs data exploratory analysis and mathematical modeling simultaneously. Given a set of data points, the basic idea is as follows: 1) a Kohonen neural network is used to divide the data into clusters of points, 2) several modeling techniques are then used to construct competing models for each cluster, 3) the best model for each cluster is then selected and called the Local Expert Model. Finally, a so-called Gating Network combines the outputs of all Local Expert Models. For comparison purposes, the same modeling techniques are also evaluated when acting as Global Experts, i. e., when the technique uses the entire data set without any clustering.
Research Interests:
Research Interests:
Abstract. This paper describes a software-in-the-loop simulation environment for the test and validation of a hierarchical guidance and control algorithm for a small fixed-wing unmanned aerial vehicle (UAV), utilizing the commercially... more
Abstract. This paper describes a software-in-the-loop simulation environment for the test and validation of a hierarchical guidance and control algorithm for a small fixed-wing unmanned aerial vehicle (UAV), utilizing the commercially available X-Plane flight simulator and ...
Abstract. Hardware in the Loop (HIL) Simulator is an important step in system design & development. In the present work, a lateral and longitudinal control system for Unmanned Aerial Vehicle (UAV) target was designed and implemented in a... more
Abstract. Hardware in the Loop (HIL) Simulator is an important step in system design & development. In the present work, a lateral and longitudinal control system for Unmanned Aerial Vehicle (UAV) target was designed and implemented in a dedicated platform. Then, ...
This paper presents an adaptive inverse control approach for the positional control of an unconstrained multibody system with flexible appendages. The approach is called Feedback-Error-Learning and it is based on the output of a feedback... more
This paper presents an adaptive inverse control approach for the positional control of an unconstrained multibody system with flexible appendages. The approach is called Feedback-Error-Learning and it is based on the output of a feedback controller with fixed parameters to adapt a ...
Research Interests:
Research Interests: Reinforcement Learning, Robot Learning, Atmospheric Modeling, Vehicle Dynamics, Aerodynamics, and 15 moreControl Systems, Control system, Optimization, Protocols, Robots, Tracking, Learning Automata, Local Area Networks, Computational Efficiency, Mathematical Model, Control Strategy, Path Tracking, Training Algorithm, UDP, and Attitude Control
Research Interests:
Research Interests:
Abstract. This article is concerned with the remote control of real experiments via internet with the aim of sharing expensive laboratory resources by making them available at any time to remote users.
Abstract This work concerns the Mixture of Local Experts Models (MLEM) technique, which can be seen as a Data Mining tool that performs data exploratory analysis and mathematical modeling simultaneously.
Resumo-Este artigo apresenta uma abordagem para o problema de controle do deslocamento vertical do plasma no tokamak ETE baseado em Rede Neural Artificial (RNA). O ETE é um tokamak de pequena razão de aspecto que deverá operar com plasmas... more
Resumo-Este artigo apresenta uma abordagem para o problema de controle do deslocamento vertical do plasma no tokamak ETE baseado em Rede Neural Artificial (RNA). O ETE é um tokamak de pequena razão de aspecto que deverá operar com plasmas alongados, que são mais susceptíveis aos deslocamentos verticais. Um controlador neural esta sendo desenvolvido baseado em RNA, com o objetivo de controlar a posição vertical do plasma.
Abstract This paper concerns the application of two neural network architetures to solve the problem of visual interception of a stationary 3D target by a stereo (binocular) vision system. The neural networks are trained using competitive... more
Abstract This paper concerns the application of two neural network architetures to solve the problem of visual interception of a stationary 3D target by a stereo (binocular) vision system. The neural networks are trained using competitive and unsupervised learning. The first neural architecture uses two Fuzzy-ART pattern-clustering neural networks which are combined with a linear “representation layer” to act as a fast open-loop neural controller and provides a rapid and coarse positioning of the visual system.
Resumo-Este artigo apresenta uma abordagem para controle de um Levitador Magnético (MagLev) utilizando controle analógico. Este MagLev foi construído com o intuito de testar sistemas de controle para um experimento de fusão nuclear, neste... more
Resumo-Este artigo apresenta uma abordagem para controle de um Levitador Magnético (MagLev) utilizando controle analógico. Este MagLev foi construído com o intuito de testar sistemas de controle para um experimento de fusão nuclear, neste caso, sistemas de controle clássicos analógicos (Proporcionais-Derivativos (PD), Proporcionais-Integrais (PI) e Proporcionais-Integrais-Derivativos
In this paper we propose a modelling technique designed to combine the results of different experts (forecasting techniques, in our case) where each expert model (called Local Expert) is developed using only part of the data set. Many... more
In this paper we propose a modelling technique designed to combine the results of different experts (forecasting techniques, in our case) where each expert model (called Local Expert) is developed using only part of the data set. Many expert models are developed for the same part of the data set and only the best expert for each part is then used.
Resumo—Os sensores desempenham um importante papel em um sistema autônomo porque extraem do ambiente as informaçoes que o sistema necessita para a determinaçao da açao a ser tomada. Assim, o projetista deve analisar o custo-benefıcio para... more
Resumo—Os sensores desempenham um importante papel em um sistema autônomo porque extraem do ambiente as informaçoes que o sistema necessita para a determinaçao da açao a ser tomada. Assim, o projetista deve analisar o custo-benefıcio para determinar os tipos de sensores a serem utilizados. Em se tratando de distância, um exemplo de sensor com um baixo custo e fácil interpretaçao dos dados é o sonar.
Resumo. Neste artigo são apresentados os resultados alcançados com o aprendizado por reforço, do inglês Reinforcement Learning (RL) de jogadores artificiais de jogo da velha em um tabuleiro com 3X3 casas. O aprendizado por reforço é uma... more
Resumo. Neste artigo são apresentados os resultados alcançados com o aprendizado por reforço, do inglês Reinforcement Learning (RL) de jogadores artificiais de jogo da velha em um tabuleiro com 3X3 casas. O aprendizado por reforço é uma técnica de aprendizagem onde os agentes recebem recompensas e punições a partir da realização de uma determinada ação.
Abstract. This article is concerned in the advantages of to use an arrangement of sonars with one transmitter and two receivers for a more trustworthy description of an environment through the methodology occupancy grid. A previous... more
Abstract. This article is concerned in the advantages of to use an arrangement of sonars with one transmitter and two receivers for a more trustworthy description of an environment through the methodology occupancy grid. A previous article presented the mobile robot ROMEO III which was designed to navigate on a known environment. The robot was capable of following a desired trajectory by combining data from odometry sensors with a map of its known environment.
Resumo-Este artigo apresenta o projeto, a implementação e os testes de validação do hardware e software do robô móvel construído no ITA chamado de ROMEO II. Este robô móvel contém um microcontrolador 80C32 (da família INTEL 8051), 2... more
Resumo-Este artigo apresenta o projeto, a implementação e os testes de validação do hardware e software do robô móvel construído no ITA chamado de ROMEO II. Este robô móvel contém um microcontrolador 80C32 (da família INTEL 8051), 2 motores de passo e 4 sensores de colisão. Através de sua porta serial, o robô recebe de um computador desktop uma descrição em alto nível da trajetória que deve ser executada.
Abstract Concerns the use of neural nets in the feedback control of systems described by ordinary differential or difference equations. The controller is assumed to be of fixed structure but with free parameters that must be tuned to meet... more
Abstract Concerns the use of neural nets in the feedback control of systems described by ordinary differential or difference equations. The controller is assumed to be of fixed structure but with free parameters that must be tuned to meet some previously chosen performance specifications. The supervisor consists of a neural net producing the actions and evaluating the success of failure of the selected action. An action is a point in the parameter space that is used to adjust a controller of fixed structure.
Abstract This paper discusses three structures for neural control of a flexible link using the Feedback-Error-Learning technique. This technique aims to acquire the inverse dynamic model of the plant and uses a neural network acting as an... more
Abstract This paper discusses three structures for neural control of a flexible link using the Feedback-Error-Learning technique. This technique aims to acquire the inverse dynamic model of the plant and uses a neural network acting as an adaptive controller to improve the performance of a conventional non-adaptive feedback controller. The non-collocated control of a flexible link is characterized as a non-minimum phase system, which is difficult to be controlled by most control techniques.
Resumo Este artigo apresenta o resultado da comparação do desempenho de dois tipos de classificadores neurais com aprendizagem supervisionada para o problema de reconhecimento de caracteres. As técnicas de redes neurais utilizadas foram:... more
Resumo Este artigo apresenta o resultado da comparação do desempenho de dois tipos de classificadores neurais com aprendizagem supervisionada para o problema de reconhecimento de caracteres. As técnicas de redes neurais utilizadas foram: rede neural tipo feedforward Perceptron Multicamada (MLP) com 1 camada escondida sendo treinada com o algoritmo backpropagation ea rede Support Vector
Resumo–Este artigo descreve um sistema para o comando e monitoração remota da plataforma robótica móvel Romeo III através da internet. Os componentes principais do sistema são: um microcomputador conectado à internet executando um... more
Resumo–Este artigo descreve um sistema para o comando e monitoração remota da plataforma robótica móvel Romeo III através da internet. Os componentes principais do sistema são: um microcomputador conectado à internet executando um programa do tipo “web browser” padrão; um microcomputador conectado à internet executando um programa do tipo “web server” conectado com o sistema a ser monitorado via comunicação do tipo sem fio (wireless).
Abstract: This article deals with the relative efficiency evaluation of a set of companies operating in the telecommunication field, by the use of both Data Envelopment Analysis (DEA) and Kohonen Neural Networks (KNN). We use a DEA Model... more
Abstract: This article deals with the relative efficiency evaluation of a set of companies operating in the telecommunication field, by the use of both Data Envelopment Analysis (DEA) and Kohonen Neural Networks (KNN). We use a DEA Model to evaluate the relative efficiency of each company present on the database and KNN with a bi-dimensional topology aiming at finding clusters on the database. KNN results are not deterministic, since the estimated weights are randomly initialized.
ABSTRACT This paper presents a scheme of multiple neural networks with a new strategy of combination. This combination can obtain an accumulative learning: the knowledge is increased by adding more neural networks to the system. This... more
ABSTRACT This paper presents a scheme of multiple neural networks with a new strategy of combination. This combination can obtain an accumulative learning: the knowledge is increased by adding more neural networks to the system. This scheme is applied to flexible link control via Feedback-Error-Learning strategy, here called Multi-Network-Feedback-Error-Learning. Three different neural control approaches are combined to acquire a better inverse dynamic model of the plant.