• Journal of the Korea Institute of Military Science and Technology
• /
• 제24권1호
• /
• pp.123-132
• /
• 2021

In order for the underwater vehicle to perform various tasks, it is important to control the depth, course, and roll of the underwater vehicle. To design such a controller, it is necessary to construct a dynamic model of the underwater vehicle and select the appropriate hydrodynamic coefficients. For the controller design, since the dynamic model is linearized assuming a limited operating range, the control performance in the steady state is well satisfied, but the control performance in the transient state may be unstable. In this paper, in order to overcome the problems of the existing controller design, we propose a A2C(Advantage Actor-Critic) based roll controller for underwater vehicle with stable learning performance in a continuous space among reinforcement learning methods that can be learned through rewards for actions. The performance of the proposed A2C based roll controller is verified through simulation and compared with PID and Dueling DDQN based roll controllers.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제15권3호
• /
• pp.853-873
• /
• 2021

In recent years, deep reinforcement learning (DRL) models are enjoying great interest as their success in a variety of challenging tasks. Deep Q-Network (DQN) is a widely used deep reinforcement learning model, which trains an intelligent agent that executes optimal actions while interacting with an environment. This model is well known for its ability to surpass skilled human players across many Atari 2600 games. Although DQN has achieved excellent performance in practice, there lacks a clear understanding of why the model works. In this paper, we present a visual analytics system for understanding deep Q-network in a non-blind matter. Based on the stored data generated from the training and testing process, four coordinated views are designed to expose the internal execution mechanism of DQN from different perspectives. We report the system performance and demonstrate its effectiveness through two case studies. By using our system, users can learn the relationship between states and Q-values, the function of convolutional layers, the strategies learned by DQN and the rationality of decisions made by the agent.

• The Journal of Korea Robotics Society
• /
• 제17권3호
• /
• pp.264-272
• /
• 2022

Human-robot cooperative tasks are increasingly required in our daily life with the development of robotics and artificial intelligence technology. Interactive reinforcement learning strategies suggest that robots learn task by receiving feedback from an experienced human trainer during a training process. However, most of the previous studies on Interactive reinforcement learning have required an extra feedback input device such as a mouse or keyboard in addition to robot itself, and the scenario where a robot can interactively learn a task with human have been also limited to virtual environment. To solve these limitations, this paper studies training strategies of robot that learn table balancing tasks interactively using deep reinforcement learning with human's facial expression feedback. In the proposed system, the robot learns a cooperative table balancing task using Deep Q-Network (DQN), which is a deep reinforcement learning technique, with human facial emotion expression feedback. As a result of the experiment, the proposed system achieved a high optimal policy convergence rate of up to 83.3% in training and successful assumption rate of up to 91.6% in testing, showing improved performance compared to the model without human facial expression feedback.

• Proceedings of the Korea Inteligent Information System Society Conference
• /
• 한국지능정보시스템학회 2001년도 The Pacific Aisan Confrence On Intelligent Systems 2001
• /
• pp.59-64
• /
• 2001

The robot soccer simulation game is a dynamic multi-agent environment. In this paper we suggest a new reinforcement learning approach to each agent's dynamic positioning in such dynamic environment. Reinforcement learning is the machine learning in which an agent learns from indirect, delayed reward an optimal policy to chose sequences of actions that produce the greatest cumulative reward. Therefore the reinforcement learning is different from supervised learning in the sense that there is no presentation of input pairs as training examples. Furthermore, model-free reinforcement learning algorithms like Q-learning do not require defining or learning any models of the surrounding environment. Nevertheless it can learn the optimal policy if the agent can visit every state- action pair infinitely. However, the biggest problem of monolithic reinforcement learning is that its straightforward applications do not successfully scale up to more complex environments due to the intractable large space of states. In order to address this problem. we suggest Adaptive Mediation-based Modular Q-Learning (AMMQL)as an improvement of the existing Modular Q-Learning (MQL). While simple modular Q-learning combines the results from each learning module in a fixed way, AMMQL combines them in a more flexible way by assigning different weight to each module according to its contribution to rewards. Therefore in addition to resolving the problem of large state effectively, AMMQL can show higher adaptability to environmental changes than pure MQL. This paper introduces the concept of AMMQL and presents details of its application into dynamic positioning of robot soccer agents.

• Journal of the Korea Society of Computer and Information
• /
• 제26권12호
• /
• pp.11-18
• /
• 2021

This paper shows the applicability of deep learning algorithm in predicting target position and getting correction value of impact point in order to improve the accuracy of naval gun firing. Predicting target position, the proposed model using LSTM model and RN structure is expected to be more accurate than existing method using kalman filter. Getting correction value of impact point, the another proposed model suggests a reinforcement model that manages factors which is related in ballistic calculation as data set, and learns using the data set. The model is expected to reduce error of naval gun firing. Combining two models, a ballistic calculation model for improving accuracy of naval gun firing based on deep learning algorithm was designed.

• The Journal of Korea Robotics Society
• /
• 제19권1호
• /
• pp.117-129
• /
• 2024

This paper introduces collective navigation through a narrow gap using a curriculum-based deep reinforcement learning algorithm for a swarm of unmanned aerial vehicles (UAVs). Collective navigation in complex environments is essential for various applications such as search and rescue, environment monitoring and military tasks operations. Conventional methods, which are easily interpretable from an engineering perspective, divide the navigation tasks into mapping, planning, and control; however, they struggle with increased latency and unmodeled environmental factors. Recently, learning-based methods have addressed these problems by employing the end-to-end framework with neural networks. Nonetheless, most existing learning-based approaches face challenges in complex scenarios particularly for navigating through a narrow gap or when a leader or informed UAV is unavailable. Our approach uses the information of a certain number of nearest neighboring UAVs and incorporates a task-specific curriculum to reduce learning time and train a robust model. The effectiveness of the proposed algorithm is verified through an ablation study and quantitative metrics. Simulation results demonstrate that our approach outperforms existing methods.

• Journal of Fisheries and Marine Sciences Education
• /
• 제24권6호
• /
• pp.763-780
• /
• 2012

The purpose of this study was to develop the teaching & learning model to increase the key competencies in the university. For this, the researcher suggested the draft of teaching & learning model in the university based on literature analysis and survey for key competencies education to professors and students. After that, the draft of the model was revised through the formative test by experts. And then, the revised model was verified through delphi survey and a final model was fixed. As the result of this study, features of the final model are as follows. First, this study identified that a core values of teaching and learning model for the key competencies education in the university is reinforcement of learner's thinking skills and enhancement of field oriented on learning contents and reinforcement of professor's role as a mentor, a facilitator, a motivator, and an informant. Second, The core learning activity's process of the teaching and learning model for the key competencies education in the university consists of the following in order: team-building, exploring problem, gathering and analysing informations, arranging of leaning outcomes, comprehensive arrangement and self-examination.

• Journal of Korean Association for Spatial Structures
• /
• 제21권4호
• /
• pp.73-80
• /
• 2021

A mid-story isolation system was proposed for seismic response reduction of high-rise buildings and presented good control performance. Control performance of a mid-story isolation system was enhanced by introducing semi-active control devices into isolation systems. Seismic response reduction capacity of a semi-active mid-story isolation system mainly depends on effect of control algorithm. AI(Artificial Intelligence)-based control algorithm was developed for control of a semi-active mid-story isolation system in this study. For this research, an practical structure of Shiodome Sumitomo building in Japan which has a mid-story isolation system was used as an example structure. An MR (magnetorheological) damper was used to make a semi-active mid-story isolation system in example model. In numerical simulation, seismic response prediction model was generated by one of supervised learning model, i.e. an RNN (Recurrent Neural Network). Deep Q-network (DQN) out of reinforcement learning algorithms was employed to develop control algorithm The numerical simulation results presented that the DQN algorithm can effectively control a semi-active mid-story isolation system resulting in successful reduction of seismic responses.

• KIPS Transactions on Software and Data Engineering
• /
• 제10권4호
• /
• pp.143-152
• /
• 2021

It is very important to learn behavioral policies that allow multiple agents to work together organically for common goals in various real-world applications. In this multi-agent reinforcement learning (MARL) environment, most existing studies have adopted centralized training with decentralized execution (CTDE) methods as in effect standard frameworks. However, this multi-agent reinforcement learning method is difficult to effectively cope with in a dynamic environment in which new environmental changes that are not experienced during training time may constantly occur in real life situations. In order to effectively cope with this dynamic environment, this paper proposes a novel multi-agent reinforcement learning system, C-COMA. C-COMA is a continual learning model that assumes actual situations from the beginning and continuously learns the cooperative behavior policies of agents without dividing the training time and execution time of the agents separately. In this paper, we demonstrate the effectiveness and excellence of the proposed model C-COMA by implementing a dynamic mini-game based on Starcraft II, a representative real-time strategy game, and conducting various experiments using this environment.

• Nuclear Engineering and Technology
• /
• 제55권10호
• /
• pp.3515-3524
• /
• 2023

A reinforcement learning framework is proposed for the control problem of outlet steam pressure of the once-through steam generator(OTSG) in this paper. The double-layer controller using Proximal Policy Optimization(PPO) algorithm is applied in the control structure of the OTSG. The PPO algorithm can train the neural networks continuously according to the process of interaction with the environment and then the trained controller can realize better control for the OTSG. Meanwhile, reinforcement learning has the characteristic of difficult application in real-world objects, this paper proposes an innovative pretraining method to solve this problem. The difficulty in the application of reinforcement learning lies in training. The optimal strategy of each step is summed up through trial and error, and the training cost is very high. In this paper, the LSTM model is adopted as the training environment for pretraining, which saves training time and improves efficiency. The experimental results show that this method can realize the self-adjustment of control parameters under various working conditions, and the control effect has the advantages of small overshoot, fast stabilization speed, and strong adaptive ability.