• Convergence Security Journal
• /
• v.18 no.3
• /
• pp.3-9
• /
• 2018

The Deep Deterministic Policy Gradient (DDPG) algorithm is an algorithm that learns by using artificial neural network s and reinforcement learning. Among the studies related to reinforcement learning, which has been recently studied, the D DPG algorithm has an advantage of preventing the cases where the wrong actions are accumulated and affecting the learn ing because it is learned by the off-policy. In this study, we experimented to control the bicycle autonomously by applyin g the DDPG algorithm. Simulation was carried out by setting various environments and it was shown that the method us ed in the experiment works stably on the simulation.

• Journal of the Korean Society for Precision Engineering
• /
• v.20 no.6
• /
• pp.62-70
• /
• 2003

This paper deals with a approximation and tracking control of hydraulic servo system using a real time recurrent neural networks (RTRN) with 2-dimensional iterative learning rule. And it was driven that 2-dimensional iterative learning rule in discrete time. In order to control the trajectory of position, two RTRN with same network architecture were used. Simulation results show that two RTRN using 2-D learning algorithm is able to approximate the plant output and desired trajectory to a very high degree of a accuracy respectively and the control algorithm using two same RTRN was very effective to control trajectory tracking of electro-hydraulic servo system.

• Transactions of the Korean Society of Mechanical Engineers A
• /
• v.24 no.3 s.174
• /
• pp.653-660
• /
• 2000

The automatization by industrial robot of today is merely rely on to the simple position repeating works, but requirements of research and development to the force control which would adapt positively to various restriction or contacting works to environment. In this paper, a learning control algorithm using, neural networks is proposed for the position and force control by a direct-drive robot. The proposed controller is the feedback controller to which the learning function of neural network is added on to and has a character of improving controller's efficiency by learning. The effectiveness of the proposed algorithm is demonstrated by the experiment on the hybrid position and force control of a parallelogram link robot with a force sensor.

• Journal of the Korea Society of Computer and Information
• /
• v.23 no.10
• /
• pp.23-31
• /
• 2018

Recently, there have been many studies on machine learning. Among them, studies on reinforcement learning are actively worked. In this study, we propose a controller to control bicycle using DDPG (Deep Deterministic Policy Gradient) algorithm which is the latest deep reinforcement learning method. In this paper, we redefine the compensation function of bicycle dynamics and neural network to learn agents. When using the proposed method for data learning and control, it is possible to perform the function of not allowing the bicycle to fall over and reach the further given destination unlike the existing method. For the performance evaluation, we have experimented that the proposed algorithm works in various environments such as fixed speed, random, target point, and not determined. Finally, as a result, it is confirmed that the proposed algorithm shows better performance than the conventional neural network algorithms NAF and PPO.

• The Transactions of the Korean Institute of Electrical Engineers D
• /
• v.49 no.11
• /
• pp.610-616
• /
• 2000

This paper presents an active noise control algorithm via Walsh transform domain controller learned by genetic algorithm. Typical active noise control algorithms such as the filtered-x lms algorithm are based on the gradient algorithm. Gradient algorithm have two major problems; local minima and eigenvalue ratio. To solve these problems, we propose a combined algorithm which consist of genetic learning algorithm and discrete Walsh transform called Walsh Transform Domain Genetic Algorithm(WTDGA). Analyses and computer simulations on the effect of Walsh transform to the genetic algorithm are performed. The results show that WTDGA increase convergence speed and reduce steady state errors.

• 제어로봇시스템학회:학술대회논문집
• /
• 2001.10a
• /
• pp.176.4-176
• /
• 2001

In this paper, a novel type of iterative learning controller is studied. The proposed learning algorithm utilizes not only the error signal of the previous iteration but also the delayed error signal of the current iteration. The delayed error signal is adopted to improve the convergence speed. The convergence condition is examined and the result shows that the proposed learning algorithm shows the fast convergence speed under the same convergence condition of the traditional iterative learning algorithm. The simulation examples are presented to confirm the validity of the proposed ILC algorithm.

• Proceedings of the Earthquake Engineering Society of Korea Conference
• /
• 2000.10a
• /
• pp.509-514
• /
• 2000

A structural control algorithm using CMAC(Cerebellar Model Articulation Controller) neural network is proposed Learning rule for CMAC is derived based on cost function. Learning convergence of CMAC is compared with MLNN(Multilayer Neural Network). Numerical examples are shown to verify the proposed control algorithm. Examples show that CMAC can be applicable to structural control with fast learning speed.

• Transactions of the Korean Society of Mechanical Engineers A
• /
• v.21 no.11
• /
• pp.1819-1826
• /
• 1997

Direct-drive robots are suitable to the position and force control with high accuracy, but it is difficult to design a controller because of the system's nonlinearity and link-interactions. This paper is concerned with the study of the force control of direct-drive robots. The proposed algorithm consists of feedback controllers and a neural network. After the completion of learning, the output of feedback controller is nearly equal to zero, and the neural network controller plays an important role in the control system. Therefore, the optimum retuning of parameters of feedback controllers is unnecessary. In other words, the proposed algorithm does not require any knowledge of the controlled system in advance. The effectiveness of the proposed algorithm is demonstrated by the experiment on the force control of the parallelogram link-type direct-drive robot.

• Proceedings of the Korean Nuclear Society Conference
• /
• 1996.11a
• /
• pp.73-78
• /
• 1996

In this study, we proposed a fuzzy gain scheduler with intelligent learning algorithm for a reactor control. In the proposed algorithm, we used the gradient descent method to learn the rule bases of a fuzzy algorithm. These rule bases are learned toward minimizing an objective function, which is called a performance cost function. The objective of fuzzy gain scheduler with intelligent learning algorithm is the generation of adequate gains, which minimize the error of system. The condition of every plant is generally changed as time gose. That is, the initial gains obtained through the analysis of system are no longer suitable for the changed plant. And we need to set new gains, which minimize the error stemmed from changing the condition of a plant. In this paper, we applied this strategy for reactor control of nuclear power plant (NPP), and the results were compared with those of a simple PI controller, which has fixed gains. As a result, it was shown that the proposed algorithm was superior to the simple PI controller.

• Journal of the Korean Society of Industry Convergence
• /
• v.18 no.4
• /
• pp.224-230
• /
• 2015

In this paper, it is presented a learning controller for repetitive walking control of biped walking robot. We propose the iterative learning control algorithm which can learn periodic nonlinear load change ocuured due to the walking period through the intelligent control, not calculating the complex dynamics of walking robot. The learning control scheme consists of a feedforward learning rule and linear feedback control input for stabilization of learning system. The feasibility of intelligent control to biped robotic motion is shown via dynamic simulation with 25-DOF biped walking robot.