Tae-Dok Eom

Heterogeneous local model networks for time series prediction

The approaches of local modeling have emerged as one of the promising methods of time series prediction. By use of the divide-and-conquer method, local models can exploit state-dependent features to approximate a subset of training data accurately. However, the generalization performance of local model networks is subject to the proper selection of model parameters. In this paper, we present a new method for local model construction for the noisy time series prediction. The proposed method uses the principal component analysis (PCA) and cross-validation technique to construct an optimal input vector for each local model. A heuristic learning rule is also proposed to update the mixture of experts network structure, which determines the confidence level of local prediction model. The proposed method has been tested with noisy Mackey-Glass time series and Sunspot series.

Generalized asymmetrical bidirectional associative memory for multiple association

A classical bidirectional associative memory (BAM) suffers from low storage capacity and abundance of spurious memories though it has the properties of good generalization and noise immunity. In this paper, Hamming distance in recall procedure of usual asymmetrical BAM is replaced with modified Hamming distance by introducing weighting matrix into connection matrix. This generalization is validated to increase storage capacity, to lessen spurious memories, to enhance noise immunity, and to enable multiple association using simulation work.

Hierarchical object recognition algorithm based on Kalman filter for adaptive cruise control system using scanning laser

Not merely running at a designated constant speed as classical cruise control, the adaptive cruise control (ACC) maintains a safe headway distance when the front is blocked by other vehicles. One of the most essential part of ACC system is the range sensor which can measure the position and speed of all objects in front continuously, ignore all irrelevant objects, distinguish vehicles in different lanes and lock on to the closest vehicle in the same lane. In this paper, the hierarchical object recognition algorithm (HORA) is proposed to process raw scanning laser data and acquire valid distance to target vehicle. HORA contains two principal concepts. First, the concept of life quantifies the reliability of range data to filter off the spurious detection and preserve the missing target position. Second, the concept of conformation checks the mobility of each obstacle and tracks the position shift. To estimate and predict the vehicle position a Kalman filter is used. Repeatedly updated covariance matrices determine the bound of valid data. The algorithm is emulated on computer and tested on-line with our ACC vehicle.

New skill learning paradigm using various kinds of neurons

Modeled from human neurons, various types of artificial neurons are developed and applied to the control algorithm. In this paper, the weights and structure of feedforward neural network controller are updated using new skill learning paradigm which consists of supervisory controller, chaotic neuron filter and associative memory. The pattern of system nonlinearity along the desired path is extracted while supervisory controller guarantees stability in the sense of the boundedness of the tracking error. Next the pattern is divided into small segments and encoded to bipolar codes depending on the existence of critical points. Comparing the encoded pattern with the pre-stored neural parameters and pattern pairs through the associative memory, the most similar one is obtained. Also, the chaotic neuron filter is used to add perturbation to neural parameters when the training of feedforward neural network is not successful with the pre-stored parameters. Finally the memory is updated with new successful parameters and pattern pairs. Simulation is performed for simple two-link robot in case of the slight modification of the desired trajectory.

The problem of stability in the application of neural network to continuous-time dynamic systems

Using a neural network to identify a function in the dynamic equation brings about additional difficulties which are not generic in other function approximation problems. First, training samples can not be arbitrarily chosen due to hard nonlinearity, so are apt to be nonuniform over the region of interest. Second, the system may become unstable while attempting to obtain the samples. This paper deals with these problems in continuous-time systems and suggests an effective solution, which provides stability and uniform sampling by the virtue of a supervisory controller. The supervisory control algorithm can be applied to robot system dynamics. The algorithm can be applied to an n-th order robot system, a simulation result is given for a simple two link robot.

Stable nonlinear controller design for a Takagi-Sugeno fuzzy model

This paper proposes another adaptive control scheme for nonlinear systems using a Takagi-Sugeno fuzzy model. Takagi-Sugeno fuzzy models have been widely used to identify the structures and parameters of unknown or partially known plants, and to control nonlinear systems. This scheme shows a good approximation capability by the fuzzy blending of local dynamics. Since a Takagi-Sugeno fuzzy model is a nonlinear system in nature, and its parameters are not linearly parameterized, it is difficult to design an adaptive controller using conventional design methods for adaptive controllers which are derived from linearly parameterized systems. In this paper, the functional form of the local dynamics are assumed to be known, but the corresponding parameters are unknown. This additional information about system nonlinearity makes it possible to design an adaptive controller for a nonlinearly parameterized system. The control law is similar to that of a conventional adaptive control technique, while its parameter-update rule is based on the local search method. A parameter-update law is derived so that the time-derivative of the Lyapunov function is negative in the region of interest. Simulation results have shown that this adaptive controller is capable of a good performance.

Developing Soccer-Playing Robots Based on the Centralized Approach

ABSTRACTThis paper presents the design procedure for soccer-playing robots based on the centralized approach. Using a fast vision system, we obtain the configuration of each robot, then the host computer computes the desired motion and commands each robot directly via RF communication. The robot soccer game has a lot of challenging problems such as coordination between robots, motion planning of robots, visual recognition of objects, and so on. To implement such functions, we think that the centralized approach may be more powerful than the distributed approach. We describe the technical tips for developing the robots in detail here and explain our strategy for getting the scores.

Information processing using chaos with application to mobile robot navigation problems

Throughout this study on information processing using an artificial neural network (ANN) and chaos we are attempting to devise a memory model that resembles human behavioral characteristics. For that purpose we construct a framework of the macroscopic model of the responding process in biological systems. Incoming stimuli are applied to the sensory receptors and preprocessed. A pattern-matching block allows one of the chaotic memories to find a feasible response in an associative way. After the chaotic memory is stabilized on one of the stable equilibrium points or limit cycles, its performance is evaluated. Since chaotic memory and the performance evaluation block form a feedback loop, they can handle features of the information blocks and store newly updated information blocks. Two kinds of chaotic memories are established in this paper: one is a 1-D map in which many information blocks can be stored as unstable periodic orbits, and the other is the famous Lozi attractor with rich dynamics. Simulations are performed for the mobile robot navigation problem in each case.