Hyeon-Min Shim

Implementation of an intelligent walking assistant robot for the elderly in outdoor environment

In this paper, we propose an architecture of walking assistant robot for the elderly which is usable at outdoor environment. It is equipped with a PXA255 embedded board, motors, a laser range finder, a CCD camera and a GPS receiver. A user operates it using haptic handle bar. It is useful to the elderly for more comfortable walking and path guidance.

Design and implementation of real-time security guard robot using CDMA networking

Progression of a mobile robot and communication techniques enables intelligent robot to be developed more versatilely. Through the ubiquitous robot control, paradigm of robot which performs specific tasks remotely has changing into intelligent robot to perform public and individual tasks. Especially, in the robot and security industry, security guard robot provides a variety of information to user and achieves its duty through the Web interface. The communication technique for these telepresence robot takes charge of the probability of various service in the robot industry. Last year, the interface for telepresence robot has developed over the Web or RF between robots, but these system has the demerits of limited distance or geographical limit to be established Internet link. In this paper, we propose the SG-robot (security guard robot) that can be operated and conduct surveillance the environment around itself anytime/anywhere using CDMA networking. SG-robot was able to solve those problems, conduct the surveillance task and communicate with between multi SG-robot and users over the cdma2000-1x communication network efficiently

EMG Pattern Classification by Split and Merge Deep Belief Network

In this paper; we introduce an enhanced electromyography (EMG) pattern recognition algorithm based on a split-and-merge deep belief network (SM-DBN). Generally, it is difficult to classify the EMG features because the EMG signal has nonlinear and time-varying characteristics. Therefore, various machine-learning methods have been applied in several previously published studies. A DBN is a fast greedy learning algorithm that can identify a fairly good set of weights rapidly—even in deep networks with a large number of parameters and many hidden layers. To reduce overfitting and to enhance performance, the adopted optimization method was based on genetic algorithms (GA). As a result, the performance of the SM-DBN was 12.06% higher than conventional DBN. Additionally, SM-DBN results in a short convergence time, thereby reducing the training epoch. It is thus efficient in reducing the risk of overfitting. It is verified that the optimization was improved using GA.

SG-Robot: CDMA network-operated mobile robot for security guard at home

The paradigm of robot which performs specific tasks remotely has changed into intelligent robot to perform public and individual tasks. Especially, in the robot and security industry, security guard robot provides a variety of information to the user and achieves its duty through the Web. The communication technique for these telepresence robots takes charge of the probability of various services in the robot industry. Last year, the interface for telepresence robot has developed over the Web or RF between robots, but these systems have the demerits of limited distance or geographical limit to be established to an internet link. In this paper, we propose the SG-Robot(Security Guard Robot) that can be operated and conduct surveillance of the environment around itself anytime/anywhere using CDMA networking. SG-Robot was able to solve those problems, conduct the surveillance task and communicate between multi SG-Robot and users over the CDMA2000-1x communication network efficiently.

Split and merge algorithm for deep learning and its application for additional classes

We present a novel GA based feature extractor for network optimal initialization.Our method selects more dominant feature extractor in merge phase using GA.Results show improvements recognition performance in comparison with DBNs.We also suggest a new approach for retraining additional classes as its application.Our approach for retraining can add output classes at lower error rate than DBNs. In this paper, we propose a novel split training and merge algorithm for deep learning. The proposed algorithm improves recognition accuracy and suggests a new approach for retraining. The algorithm is motivated by the genetic algorithm (GA) and is composed of two procedures. The first procedure initializes two individual networks using deep belief networks (DBNs), and the second procedure merges the two networks using the GA. Biases and weights of the network that is trained using DBNs are represented as a matrix between each layer, and each row of this matrix is used as a chromosome in the merge procedure. To evaluate the performance, we conduct two set of experiments. The first set is to recognize accuracy of the proposed algorithm, and the second set is for a new retraining approach. The results show that the proposed algorithm has a lower average error rate (6.84 ? 4.57%) than the DBNs, and it can add classes at a lower average error rate (9.06 ? 6.17% and 10.17 ? 4.51%) without pre-training using the restrict Boltzmann machines (RBMs) for existing classes data.