In this paper, we propose an adaptive gait by which a quadruped walking robot can walk against external disturbances. This adaptive gait mechanism makes it possible for a quadruped walking robot to change its gait and accommodate external disturbances form various external environmental factors. Under the assumption that external disturbances can be converted to an external force acting on the body of a quadruped walking robot, we propose a new criterion for the stability margin of a waling robot by using an effective mass center based on the zero moment point under unknown external force. And for a solution of an adaptive gait against external disturbances, an method of altitude control and reflexive direction control is suggested. An algorithmic search method for an optimal stride of the quadruped mehtod, the gait stability margin of a quadruped walking robot is optimized in changing its direction at any instance for and after the reflexive direction control. To verify the efficiency of the proposed approach, some simulaton results are provided.

When artificial potential field approach is used to avoid obstacle, the problem can be occurred in case that manipulator selects the path which across over an obstacle among paths. In thiscase manipulator can't reach the desired goal form obstacle. This problem is a case of structual local minimum. so this paper proposes the method to solve structual local minimum in this case. The method is that the manipulator goes via temporary goal. This paper proposes that visual region concept to select the temporary goal. The temporary goal is selected on the border of the visual region. To prove its effectiveness, two simulation examples are done by two link manipulator in two dimension and by three link manipulator in three dimension.

In this paper, we discuss the errors in selectivity estimation using the multilevel grid file (MLGF). We first demonstrate that the estimatio errors stem from the uniformity assumption that records are uniformly distributed in their belonging region represented by an entry in a level of an MLGF directory. Bsed on this demonstration, we then investigate five factors affecting the accuracy of estimation: (1) the data distribution in a region (2) the number of records stored in an MLFG (3) the page size, (4) the query region size, and (5) the level of an MLFG directory. Next we present the tendancy of estimation errors according to the change of values for each factor through experiments. The results show that the errors decrease when (1) the distribution of records in a region becomes closer to the uniform one, (2) the number of records in an MLFG increases, (3) the page size decreases, (4) the query region size increases, and (5) the level of an MLFG directory employed as data distribution information becomes lower. After the definition of the granule ratio, the core formula representing the basic relationship between the estimation errors and the above five factors, we finally examine the change of estimation errors according to the change of the values for the granule ratio through experiments. The results indicate that errors tend to be similar depending on the values for the granule ratio regardless of the various changes of the values for the five factors. factors affecting the accuracy of estimation:

In this paper, for the single s-a-E fault detected in a triangular cellular permutation network (TCPN), we propose a method which can tolerate a fault by reconfiguring the netowrk and analyze the possibilities of the reconfiguration. The network is set up through iterative decomposition of a permutation into the right or left coset. For the s-a-E fault of a cell which is to be transpositioned for an increasing order mapping, we cna reconfigure it merely by switching te decomposition scheme from right coset to left coset or vice versa. Also for a decreasing order mapping, we make a detour around the faulty cell. Reconfiguring with the redundant connectivity of a TCPN, we could realize form 17% to 90% of the permutation for the number of inputs from 4 to 40. REconfiguration of the network by exchanging the first input with the last input and the first output with the last output resulted in more than 99% realization of the permutation. Also with the exchange of all inputs and outputs with neighboring cells, we could have 100% realization of the permutation.

Back-propagation(BP) algorithm needs a lot of time to train the artificial neural network (ANN) to get high accuracy level in classification tasks. So there have been extensive researches to process back-propagation algorithm on parallel processors. This paper prsents a linear systolic array which calculates forward-backward propagation of BP algorithm at the same time using effective space-time transformation and PE structure. First, we analyze data flow of forwared and backward propagations and then, represent the BP algorithm into data dapendency graph (DG) which shows parallelism inherent in the BP algorithm. Next, apply space-time transformation on the DG of ANN is turn with orthogonal direction projection. By doing so, we can get a snakelike systolic array. Also we calculate the interval of input for parallel processing, calculate the indices to make the right datas be used at the right PE when forward and bvackward propagations are processed in the same PE. And then verify the correctness of output when forward and backward propagations are executed at the same time. By doing so, the proposed system maximizes parallelism of BP algorithm, minimizes th enumber of PEs. And it reduces the execution time by 2 times through making idle PEs participate in forward-backward propagation at the same time.

In this paper, the bit-level 1-dimensional systolic array for modular multiplication is designed. First of all, the parallel algorithm and data dependence graph from walter's method based on montgomery algorithm suitable for array design for modular multiplication is derived. By the systematic procedure for systolic array design, four 1-dimensional systolic arrays are obtained and then are evaluated by various criteria. As it is modified the array which is derived form [0,1] projection direction by adding a control logic and it is serialized the communication paths of data A, optimal 1-dimensional systolic array is designed. It has constant I/O channels for expansile module and it is easy for fault tolerance due to unidirectional paths. It is suitable for RSA cryptosystem which deals iwth the large size and many consecutive message blocks.

Block data transfer poses a serious problem is a pipelined bus where each data transfer step is pipelined. In this paper, we describe the design and implementation of a variable data block transfer protocol for a pipelined bus of a shared-memory multiprocessor. The proposed method maintains compatibility with the existing protocol for the pipelined bus and ensures fairness and effectiveness by preventing starvation. We present flow charts of requester and responder during a block transfer in the pipelined bus that uses the proposed protocol. The proposed protocol was implemented for the TICOM-III HiPi+Bus.

This paper proposes an image compression algorithm that adopts projection scheme on wavelet transform domain of image signal. Wavelet decomposed image is encoded by the result of projection along one direction out of eight which approximates the coefficients most closely to the originally transformed coefficients. These projectrion data are vector quantized using separate codebooks depending on the decomposition level and orientation of decomposed of image. Experimental results reveals that proposed scheme shows excellent performance in PSNR manner and also shows good subjective quality.

Computer vision-based gesture recognition systems consist of image segmentation, object tracking and decision. However, it is difficult to segment an object from image for gesture in computer systems because of vaious illuminations and backgrounds. In this paper, we describe a method to learn features for segmentation, which improves the performance of computer vision-based hand-gesture recognition systems. Systems interact with a user to acquire exact training data and segment information according to a predefined plan. System provides some models to the user, takes pictures of the user's response and then analyzes the pictures with models and a prior knowledge. The system sends messages to the user and operates learning module to extract information with the analyzed result.

In this paper, we propose a face recognition algorithm using the hidden markov model and neural networks (HMM-NN). In the preprocessing stage, we find edges of a face using the locally adaptive threshold (LAT) scheme and extract features based on generic knowledge of a face, then construct a database with extracted features. In the training stage, we generate HMM parameters for each person by using the forward-backward algorithm. In the recognition stage, we apply probability vlaues calculated by the HMM to subsequent neural networks (NN) as input data. Computer simulation shows that the proposed HMM-NN algorithm gives higher recognition rate compared with conventional face recognition algorithms.

In the stochastic structure for doing exact calculationk, an input number must be changed to a pulse stream. Because the performance of random number generator (RNG) is controlled by its initial condition, we suggested newly modified cellular automata (MCA) which is uses a counter for boundary condition. We compared newly suggested MCA RNG to previously reported RNGs using the AND gate passing outputs which have the same meaning of multiplication in the stochastic calculation. In order to use stochastic we studied about the method, one large RNG can generate many small random numbers. In this method, RNG must have large drive capabilities for many input comparator. So we studied about drive capabilities using BiCMOS circuit and CMOS circit by SPICE.

In this paper, we present a new learning method for the fuzzy ARTMAP which is effective for the noisy input patterns. Conventional fuzzy ARTMAP employs only fuzzy AND operation between input vector and weight vector in learning both top-down and bottom-up weight vectors. This fuzzy AND operation causes excessive update of the weight vector in the noisy input environment. As a result, the number of spurious categories are increased and the recognition ratio is reduced. To solve these problems, we propose a new method in updating the weight vectors: the top-down weight vectors of the fuzzy ART system are updated using weighted average of the input vector and the weight vector itself, and the bottom-up weight vectors are updated using fuzzy AND operation between the updated top-down weitht vector and bottom-up weight vector itself. The weighted average prevents the excessive update of the weight vectors and the fuzzy AND operation renders the learning fast and stble. Simulation results show that the proposed method reduces the generation of spurious categories and increases the recognition ratio in the noisy input environment.

Korean has a large set of characters, and has a two-dimensional formation: each character is composed of graphemes in two dimensions. Whereas connections between characters are rare, connections inside a grapheme and between graphemes happen frequently and these connections generate many cursive strokes. To deal with the large character set and the cursive strokes, using the graphemes as a recognition unit is an efffective approach, because it naturally accommodates the structural characteristics of the characters. In this paper, we propose a grapheme-based on-line recognition method for cursive korean characters. Our method uses a TDNN recognition engine to segment cursive strokes into graphemes and a graph-algorithmic postprocessor based on korean grapheme composition rule and viterbi search algorithm to find the best recognition score path. We experimented the method on freely hand-written charactes and obtained a recognition rate of 94.5%.

In this paper, we popose a neural network that detects edges of moving objects in an image using a neural network of hierarchical spatial filter with orientation selectivity. We modify the temporal difference network by adding a self loop to each neuraon cell to reduce the problems of phantom edge detected by the neural network proposed by kwon yool et al.. The modified neural network alleviates the phantom edges of moving objects, and also can detect edges of miving objects even for the noisy input. By computer simulation with real images, the proposed neural netowrk can extract edges of different orientation efficiently and also can reduce the phantom edges of moving objects.