• Journal of IKEEE
• /
• v.25 no.4
• /
• pp.644-649
• /
• 2021

Self-Imitation Learning is a simple off-policy actor-critic algorithm that makes an agent find an optimal policy by using past good experiences. In case that Self-Imitation Learning is combined with reinforcement learning algorithms that have actor-critic architecture, it shows performance improvement in various game environments. However, its applications are limited to reinforcement learning algorithms that have actor-critic architecture. In this paper, we propose a method of applying Self-Imitation Learning to Deep Q-Network which is a value-based deep reinforcement learning algorithm and train it in various game environments. We also show that Self-Imitation Learning can be applied to Deep Q-Network to improve the performance of Deep Q-Network by comparing the proposed algorithm and ordinary Deep Q-Network training results.

• The Journal of Korea Robotics Society
• /
• v.8 no.1
• /
• pp.20-28
• /
• 2013

Large workspace and strong grasping force are required when a robot manipulates big and/or heavy objects. In that situation, bimanual manipulation is more useful than unimanual manipulation. However, the control of both hands to manipulate an object requires a more complex model compared to unimanual manipulation. Learning by human demonstration is a useful technique for a robot to learn a model. In this paper, we propose an imitation learning method of bimanual object manipulation by human demonstrations. For robust imitation of bimanual object manipulation, movement trajectories of two hands are encoded as a movement trajectory of the object and a force trajectory to grasp the object. The movement trajectory of the object is modeled by using the framework of dynamic movement primitives, which represent demonstrated movements with a set of goal-directed dynamic equations. The force trajectory to grasp an object is also modeled as a dynamic equation with an adjustable force term. These equations have an adjustable force term, where locally weighted regression and multiple linear regression methods are employed, to imitate complex non-linear movements of human demonstrations. In order to show the effectiveness our proposed method, a movement skill of pick-and-place in simulation environment is shown.

• Journal of Institute of Control, Robotics and Systems
• /
• v.14 no.10
• /
• pp.1038-1046
• /
• 2008

This paper presents a framework to generate human-like movements of a humanoid in real time using the movement primitive database of a human. The framework consists of two processes: 1) the offline motion imitation learning based on an Evolutionary Algorithm and 2) the online motion generation of a humanoid using the database updated bγ the motion imitation teaming. For the offline process, the initial database contains the kinetic characteristics of a human, since it is full of human's captured motions. The database then develops through the proposed framework of motion teaming based on an Evolutionary Algorithm, having the kinetic characteristics of a humanoid in aspect of minimal torque or joint jerk. The humanoid generates human-like movements far a given purpose in real time by linearly interpolating the primitive motions in the developed database. The movement of catching a ball was examined in simulation.

• Korean Journal of Cognitive Science
• /
• v.23 no.4
• /
• pp.517-551
• /
• 2012

We are imitating animals. True imitation can be defined as a learning to do an act from seeing it done by others. We have been building culture by imitating others' skills and knowledge with high fidelity. In this regard, it is important to ask how the faculty of imitation has evolved and how imitation behaviors develop ontogenetically. It is also interesting to see whether nonhuman animals can imitate truly or not and how different imitation learning is among human and non-human animals. In this paper, first I review empirical data from imitation studies with human and nonhuman animals. Comparing different species, I highlight their different levels of copying fidelity and explain the reason why they are showing the difference. Then I review recent studies on neurobiological mechanisms underlying imitation. The initial neurobiological studies on imitation in humans suggested a core imitation circuitry composed of mirror neuron system [inferior parietal lobule(IPL) and inferior frontal gyrus(IFG)] and the posterior part of the superior temporal sulcus(pSTS). More recent studies on the neurobiology of imitation, however, has gone beyond the studies on the core mechanisms. Finally, I try to find out implications of psychology and biology of imitation for cultural evolution. I argue for a memetic approach to cultural evolution, along the lines with a recent study on measuring memes by mirror neurons system.

• The Journal of Korea Institute of Information, Electronics, and Communication Technology
• /
• v.14 no.4
• /
• pp.239-246
• /
• 2021

In the field of computer graphics and HCI, there are many studies on systems that create characters and interact naturally. Such studies have focused on the user's response to the user's behavior, and the study of the character's behavior to elicit positive emotions from the user remains a difficult problem. In this paper, we develop a prototype of an interaction system to elicit positive emotions from users according to the movement of virtual characters using artificial intelligence technology. The proposed system is divided into face recognition and motion generation of a virtual character. A depth camera is used for face recognition, and the recognized data is transferred to motion generation. We use imitation learning as a learning model. In motion generation, random actions are performed according to the first user's facial expression data, and actions that the user can elicit positive emotions are learned through continuous imitation learning.

• Annual Conference of KIPS
• /
• 2019.05a
• /
• pp.485-486
• /
• 2019

Simulated driving behavior is an important aspect of realistic simulation systems. To simulate natural driving behavior, this paper proposes an imitation learning method based on active learning that combines demonstration and experience. Driving demonstrations are collected from human drivers in a driving simulator. A driving behavior policy is learned from these demonstrations. The driving demonstration dataset is augmented with new demonstrations that the original demonstrations did not contain, in the form of behaviors from another driving behavior policy learned from experience. The final driving behavior policy is learned from an augmented demonstration dataset.

• Informatization Policy
• /
• v.22 no.4
• /
• pp.91-104
• /
• 2015

This study tests the effects of differential association, definitions, differential reinforcement and imitation from social learning theory in the explanation of university students' crimes in social network services. In addition, this study tests the interaction effects between social learning factors and other factors such as low self-control, subcultural environment, and crime opportunity for the integrated approach. Using data from 486 university students in Seoul, results show that both definition and imitation have significant influences on crimes, even though differential association and differential reinforcement factors have no significant influences on crimes in social network services. Results also reveal that there are significant interaction effects between definition and subcultural environment, which meana that definition has a strong effect on crimes in high subcultural environment. In addition, it is found that reinforcement has also a strong effect on crimes in high crime opportunity and that interaction effect between imitation and low self-control is significant, which means that imitation has a strong effect on crimes in low self-control students.

• Journal of Educational Research in Mathematics
• /
• v.12 no.4
• /
• pp.473-491
• /
• 2002

This article analyzes mathematics education from dialectical materialism acknowledging the objectivity of knowledge. The thesis that knowledge is objective advances to the recognition that knowledge will be internalized, and an idea of zone of proximal development(ZPD) is established as a practice program of internalization. The lower side of ZPD, i.e. the early stage of internalization takes imitation in a large portion. And in the process of internalization the mediational means play an important role. Hereupon the role of mathematics teacher, the object of imitation, stands out significantly. In this article, treating the contents of study as follows, I make manifest that teaching and learning in mathematics classroom are united dialectically: I hope to findout the method of teaching-learning to mathematical knowledge from the point of view that mathematical knowledge is objective; I look into how analysis into units, as the analytical method of Vygotsky, has been developed from the side of mathematical teaching-learning; I discuss the significance of mediational means to play a key role in attaining the internalization in connection with ZPD and re-illuminate imitation. Based on them, I propose how the role of mathematics teachers, and the principle of organization to mathematics textbook should be.

• Journal of the Korea Computer Graphics Society
• /
• v.29 no.1
• /
• pp.23-29
• /
• 2023

In this paper, we introduce a method to obtain high-quality results at a low cost for simulating musculoskeletal characters based on data from the reference motion through motion capture on two-legged walking through reinforcement learning. We reset the motion data of the reference motion to allow the character model to perform, and then train the corresponding motion to be learned through reinforcement learning. We combine motion imitation of the reference model with minimal metabolic energy for the muscles to learn to allow the musculoskeletal model to perform two-legged walking in the desired direction. In this way, the musculoskeletal model can learn at a lower cost than conventional manually designed controllers and perform high-quality bipedal walking.

• KIPS Transactions on Software and Data Engineering
• /
• v.9 no.9
• /
• pp.281-290
• /
• 2020

The Vision-and-Language Navigation(VLN) task is a complex intelligence problem that requires both visual and language comprehension skills. In this paper, we propose a new learning model for visual-language navigation agents. The model adopts a hybrid learning that combines imitation learning based on demo data and reinforcement learning based on action reward. Therefore, this model can meet both problems of imitation learning that can be biased to the demo data and reinforcement learning with relatively low data efficiency. In addition, the proposed model uses a novel path-based reward function designed to solve the problem of existing goal-based reward functions. In this paper, we demonstrate the high performance of the proposed model through various experiments using both Matterport3D simulation environment and R2R benchmark dataset.