• ETRI Journal
• /
• v.44 no.6
• /
• pp.1004-1019
• /
• 2022

For autonomous driving without high-definition maps, we present a model capable of generating multiple plausible paths from egocentric images for autonomous vehicles. Our generative model comprises two neural networks: feature extraction network (FEN) and path generation network (PGN). The FEN extracts meaningful features from an egocentric image, whereas the PGN generates multiple paths from the features, given a driving intention and speed. To ensure that the paths generated are plausible and consistent with the intention, we introduce an attentive discriminator and train it with the PGN under a generative adversarial network framework. Furthermore, we devise an interaction model between the positions in the paths and the intentions hidden in the positions and design a novel PGN architecture that reflects the interaction model for improving the accuracy and diversity of the generated paths. Finally, we introduce ETRIDriving, a dataset for autonomous driving, in which the recorded sensor data are labeled with discrete high-level driving actions, and demonstrate the state-of-the-art performance of the proposed model on ETRIDriving in terms of accuracy and diversity.

• International Journal of Internet, Broadcasting and Communication
• /
• v.7 no.2
• /
• pp.113-116
• /
• 2015

Generally, in the railway vehicle the driving force of gravity happens by the high-speed running and the repetitive impulse cause the degradation and the malfunction phenomenon shows differently because the durability of each component changes according to the internal and external causes. The maintenance of propulsion control device which is played the very important role as to the stable service of the railway vehicle is greatly important among them. Therefore maintenance training propulsion control device simulator is needed to maximize learning through repetition and improve the maintenance practical skills training. This paper designed the railway vehicle running device with a miniature for the railway vehicle maintenance training and developed a propulsion control device simulator equipped the imitation steering wheel.

• Korean Journal of Child Studies
• /
• v.13 no.2
• /
• pp.113-128
• /
• 1992

This study investigated methodological principles of childhood education in the Choson Dynasty. The study was conducted through review of literature, particularly through the analysis of the textbooks used in childhood education and the Confucianism Classics. The ultimate goal of childhood education in the Choson Dynasty was the moral completion of human beings through daily practice. The methodological principles of childhood education were the principles of learning by imitation, early education, the inseparability of knowing and doing, education for Ye(禮) in daily life, sex-role education, and teaching with consideration of the developmental characteristics of the children. The scientific validity of such methods of traditional education is verified in terms of its many points corresponding to important principles of modern education.

• Korean Journal of Child Studies
• /
• v.15 no.1
• /
• pp.23-35
• /
• 1994

This paper investigated children's learning process concerning age and gender, two of the most apparent social roles. The subjects of the study were 72 Korean preschool children between three to five years of age. The Study was designed to examine problems of developmental sequence and horizontal decalage in understanding of these social roles. Two five-step knowledge sequences of age growth and sex constancy were administered utilizing skill theory (Fischer, 1980). Each step in a sequence had different stories of age or sex role with increasing cognitive complexity, but the two sequences were equivalent in complexity. The children were tested under two different assessment conditions:high support condition of elicited imitation and low support condition of free play and the best story. The findings show that, the understanding of age and sex roles develops through the predicted five-step sequences. The understanding of age role seems to develop earlier than that of sex role although the developmental patterns of the two roles are similar. Variations in the children's performances under different conditions were dramatic, especially for the older children.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2012.11a
• /
• pp.400-402
• /
• 2012

서비스 로봇이 인간과의 상호작용을 하기 위해서는 상호작용에 필요한 동작들을 학습할 수 있는 방법이 필요하다. 본 연구에서는 서비스 로봇에게 동작 학습을 시키기 위한 모방학습 프레임 워크의 구축에 관한 연구를 진행한다. 모방학습에는 로봇이 행동을 수행하기 위한 모터 프리미티브가 필요하므로 본 연구에서는 선행자가 로봇을 직접 움직여 생성한 모터 프리미티브를 데이터 파일로 저장 하여 로봇에게 전송하면 로봇이 스스로 모터 프리미티브를 수행하도록 하는 모터 프리미티브 생성 시스템을 설계한다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2012.11a
• /
• pp.403-405
• /
• 2012

본 연구에서는 인간과 서비스 로봇과의 상호작용에 필요한 동작의 학습을 위한 방법으로 모방학습을 활용한다. 로봇을 모방학습 시키기 위해서는 모방학습을 위한 특정한 구조가 필요하다. 따라서 선행자가 로봇을 직접 움직여 생성한 모방학습 데이터를 저장하고, 그 데이터를 학습알고리즘에 적용시킨 뒤 로봇이 학습 내용을 수행하도록 하는 모방학습 프레임워크를 제안한다. 모방학습 프레임워크의 프로그래밍은 본 연구에서 사용하는 휴머노이드 로봇인 나오에서 제공하는 그래픽기반 환경 개발도구인 콜레그래프를 활용한다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2022.05a
• /
• pp.428-429
• /
• 2022

최근에 자율 주행 자동차에 관련한 관심이 증가하면서 다양한 연구들이 도출되고 있다. 특히, 자율 주행 자동차를 시뮬레이터에서 검증하는 방법은 실 환경과 비교할 때 상대적으로 안전한 성능 검증 방법으로 많이 활용되고 있다. 시뮬레이터의 핵심 기술은 실 환경과 가상 시뮬레이션 환경의 차이를 줄이는 데 있다. 본 논문에서는 Generative Adversarial Imitation Learning(GAIL)[1] 기반으로 자율 주행 자동차 시뮬레이터 내에서 다수의 가상 동적 객체들의 움직임을 제어하는 방법을 제안한다. GAIL은 생성기와 판별기로 구성된다. 생성기는 강화학습 정책 생성기와 전문가 정책 생성기를 포함한다. 판별기는 보상 학습기를 포함한다. GAIL 기반으로 가상 자동차 및 가상 보행자를 제어함으로써 동영상에서의 이동경로를 학습해서 표현할 수 있다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.11a
• /
• pp.571-574
• /
• 2023

애니메이션은 주어진 키프레임(key frame)에 맞추어 움직이기 때문에, 다른 객체와 상호작용할 때 상대편 물체의 위치나 방향을 애니메이션에 맞추어 변환해야 한다. 이 논문에서는 모방학습으로 애니메이션을 학습하고, 모션 블렌딩(motion blending) 기법으로 객체 간 상호작용을 학습하여 새로운 애니메이션을 생성하는 방법을 제안한다. 에이전트(agent)는 오브젝트의 상태를 관측하고 주어진 모션들을 블렌딩하는 방법으로 다양한 행동을 취하고 목적에 대한 보상을 받는다. 에이전트가 행동하는 과정에서 모션 블렌딩 비율에 대한 가중치를 계산하는 함수를 설계하고, 생성되는 애니메이션이 사람이 취할 수 있는 동작에 가깝도록 회전 각도 clamping 함수와 보상 시스템을 설계하여 반영한다. 모방학습 기반 모션 블렌딩 기법은 객체의 변화에 상호작용하는 애니메이션을 기존 강화학습 기반 애니메이션 생성 기법보다 적은 학습량으로 생성할 수 있음을 확인했다.

• Journal of Distribution Research
• /
• v.15 no.4
• /
• pp.61-85
• /
• 2010

Introduction: Diffusion is process by which an innovation is communicated through certain channel overtime among the members of a social system(Rogers 1983). Bass(1969) suggested the Bass model describing diffusion process. The Bass model assumes potential adopters of innovation are influenced by mass-media and word-of-mouth from communication with previous adopters. Various expansions of the Bass model have been conducted. Some of them proposed a third factor affecting diffusion. Others proposed multinational diffusion model and it stressed interactive effect on diffusion among several countries. We add a spatial factor in the Bass model as a third communication factor. Because of situation where we can not control the interaction between markets, we need to consider that diffusion within certain market can be influenced by diffusion in contiguous market. The process that certain type of retail extends is a result that particular market can be described by the retail life cycle. Diffusion of retail has pattern following three phases of spatial diffusion: adoption of innovation happens in near the diffusion center first, spreads to the vicinity of the diffusing center and then adoption of innovation is completed in peripheral areas in saturation stage. So we expect spatial effect to be important to describe diffusion of domestic discount store. We define a spatial diffusion model using multinational diffusion model and apply it to the diffusion of discount store. Modeling: In this paper, we define a spatial diffusion model and apply it to the diffusion of discount store. To define a spatial diffusion model, we expand learning model(Kumar and Krishnan 2002) and separate diffusion process in diffusion center(market A) from diffusion process in the vicinity of the diffusing center(market B). The proposed spatial diffusion model is shown in equation (1a) and (1b). Equation (1a) is the diffusion process in diffusion center and equation (1b) is one in the vicinity of the diffusing center. $$\array{{S_{i,t}=(p_i+q_i{\frac{Y_{i,t-1}}{m_i}})(m_i-Y_{i,t-1})\;i{\in}\{1,{\cdots},I\}\;(1a)}\\{S_{j,t}=(p_j+q_j{\frac{Y_{j,t-1}}{m_i}}+{\sum\limits_{i=1}^I}{\gamma}_{ij}{\frac{Y_{i,t-1}}{m_i}})(m_j-Y_{j,t-1})\;i{\in}\{1,{\cdots},I\},\;j{\in}\{I+1,{\cdots},I+J\}\;(1b)}}$$ We rise two research questions. (1) The proposed spatial diffusion model is more effective than the Bass model to describe the diffusion of discount stores. (2) The more similar retail environment of diffusing center with that of the vicinity of the contiguous market is, the larger spatial effect of diffusing center on diffusion of the vicinity of the contiguous market is. To examine above two questions, we adopt the Bass model to estimate diffusion of discount store first. Next spatial diffusion model where spatial factor is added to the Bass model is used to estimate it. Finally by comparing Bass model with spatial diffusion model, we try to find out which model describes diffusion of discount store better. In addition, we investigate the relationship between similarity of retail environment(conceptual distance) and spatial factor impact with correlation analysis. Result and Implication: We suggest spatial diffusion model to describe diffusion of discount stores. To examine the proposed spatial diffusion model, 347 domestic discount stores are used and we divide nation into 5 districts, Seoul-Gyeongin(SG), Busan-Gyeongnam(BG), Daegu-Gyeongbuk(DG), Gwan- gju-Jeonla(GJ), Daejeon-Chungcheong(DC), and the result is shown