GIST Scholar
Not a member yet
30271 research outputs found
Sort by
Biophysical Investigations of DNA-Binding Proteins from Archaea to Humans
No full textAging and genome instability are fundamental biological challenges associated with the progressive accumulation of cellular stress and DNA damage. Proteins involved in DNA replication, repair, and gene regulation play essential roles in maintaining genomic integrity, particularly under extreme environmental conditions or during cellular senescence. Understanding the structural and functional properties of these proteins not only offers insight into fundamental molecular mechanisms but also provides avenues for therapeutic development. This dissertation explores the biophysical and structural characteristics of DNA-binding proteins across three systems: archaea, human DNA replication machinery, and transcription factors (TF). A multidisciplinary approach was employed, combining nuclear magnetic resonance (NMR) spectroscopy, molecular dynamics simulation, and other biophysical techniques. In Chapter 1, I examined the thermostability and DNA-binding mechanism of the single-stranded DNA-binding protein (SSB) from the hyperthermophilic archaeon Sulfolobus solfataricus (Sso). Using solution- state NMR spectroscopy, circular dichroism (CD), differential scanning calorimetry (DSC), and hydrogen- deuterium exchange (HDX), I characterized the secondary structure, DNA-binding interface, and structural flexibility of SsoSSB under high-temperature conditions. Although rational mutagenesis guided by HDX data and MutCompute predictions was performed, none of the tested variants exhibited improved thermostability. These findings provide insight into how thermophilic proteins maintain structural integrity and function at elevated temperatures, informing strategies for the rational design of thermostable proteins. Chapter 2 focuses on the molecular interaction between minichromosome maintenance protein 6 (MCM6) and Bloom syndrome helicase (BLM), both critical regulators of DNA replication and genome integrity. Through NMR-based chemical shift perturbation (CSP), paramagnetic relaxation enhancement (PRE), and fluorescence polarization assays (FPA), I mapped the interaction between the winged-helix domain of MCM6 and the minimal binding region of BLM. Structural modeling and electrostatic analysis identified complementary interfaces, while evolutionary conservation supported functional relevance. These results reveal a bipartite interaction that may coordinate origin licensing and fork resolution under replication stress. In Chapter 3, I examined how FOXO4, a forkhead TF, recognizes and differentiates between target and non-target DNA sequences. Emphasis was placed on the role of flexible structural elements, including the W2 region and the S2–W1 loop. Using TITAN line shape analysis, NMR spectroscopy, and Molecular dynamics simulations, I demonstrated that these regions contribute to binding plasticity, electrostatically driven scanning, and autoinhibitory conformations. Site-directed mutagenesis confirmed their role in modulating specificity and affinity. Comparative simulations across FOXO family members suggest a conserved strategy for regulating DNA interaction via dynamic conformational states. This dissertation presents a detailed structural and dynamic analysis of DNA-binding proteins operating across diverse biological contexts, including archaea, DNA replication machinery, and TFs. By integrating high-resolution experimental data with computational modeling, I provide mechanistic insights into protein thermostability, domain-specific interactions, and sequence recognition. These findings contribute to a deeper understanding of how structural flexibility and conserved motifs regulate protein function and may inform future efforts in protein engineering and molecular therapeutics.|세포 스트레스와 DNA 손상의 축적으로 인한 유전체 불안정성은 생물학적으로 중요한 과제이며, DNA 복제, 수선, 전사 조절에 관여하는 단백질들은 유전체 안정성을 유지하는 데 핵심적인 역할을 수행한다. 본 박사학위 논문은 세 가지 시스템 (고세균, DNA 복제 조절 기작, 전사 조절 인자)에 존재하는 핵산 결합 단백질들의 구조적 및 생물물리학적 특성을 규명하고자 하였으며, 핵자기 공명 분광법, 분자동역학 시뮬레이션 등, 다양한 생화학적 분석 기법을 통합적으로 활용하였다.
제1장에서는 온천에서 생존하는 고세균 Sulfolobus solfataricus의 단일 가닥 DNA 결합 단백질의 열안정성과 DNA 결합 메커니즘을 조사하였다. 용액상 핵자기 공명 분광법, 원편광 이색성 분광광도계, 시차 주사 열량계, 수소-중수소 교환 분석을 통해 SsoSSB의 이차 구조, 결합 표면, 유연성 등을 고온 환경에서 분석하였으며, 수소-중수소 교환 분석과 MutCompute 기반 예측을 바탕으로 설계된 돌연변이체들이 향상된 열안정성을 보이는 것을 확인하였다. 제1장은 고온 안정 단백질의 기능적 구조 유지 전략을 이해하는 데 기여한다.
제2장에서는 DNA 복제와 손상 수선에 핵심적인 역할을 하는 MCM6 단백질과 Bloom 증후군 헬리케이스 간의 상호작용을 구조적 수준에서 규명하였다. 핵자기 공명 분광법 화학적 이동 변화, 상자성 완화 증강, 형광 편광 분석을 활용하여 MCM6의 winged-helix domain과 Bloom 증후군 헬리케이스의 최소 결합 도메인 사이의 결합 인터페이스를 밝혔다. 구조 모델링, 전하 분석, 보존성 분석을 통해 이들 도메인의 기능적 상호작용이 복제 기작 및 복제 스트레스 상황에서 조절된다는 점을 제시하였다.
제3장에서는 전사인자인 FOXO4의 forkhead domain이 DNA 서열을 인식하는 특이성 메커니즘을 탐구하였다. 특히 유연한 구조 요소인 W2 영역과 S2–W1 루프의 역할에 집중하였으며, TITAN 선형 분석, 핵자기 공명 분광법, 분자동역학 시뮬레이션을 통해 이들 영역이 DNA 결합의 전하 기반 스캐닝 및 자동 억제 기전에 기여함을 확인하였다. 표적 서열 및 비표적 서열에 대한 결합 비교 분석과 FOXO 계열 단백질 간 시뮬레이션 비교를 통해 이들 조절 기작의 보존성을 제안하였다.
본 논문은 고세균에서 사람의 전사인자까지 다양한 생물학적 시스템에서의 핵산 결합 단백질들을 대상으로 구조적 유연성과 결합 특이성에 관한 종합적 생물물리학적 분석을 수행하였다. 고해상도의 실험 자료와 시뮬레이션 기반의 계산 접근법을 융합하여 단백질의 열안정성, 도메인 간 상호작용, 서열 인식 메커니즘에 대한 통합적인 기전을 제시하였으며, 이는 향후 단백질 공학 및 치료제 개발에 유용한 기반 지식을 제공한다.DoctorAbstract i
Contents ⅱ
List of Figures ⅳ
List of Tables ⅵ
Ⅰ. Chapter 1. Biophysical and Structural Investigation of SSB from Sulfolobus
solfataricus and Thermostability Enhancement 1
1. Introduction 2
2. Materials and Methods 4
2. 1. Sample preparation 4
2. 2. NMR experiments 5
2. 3. Secondary structure prediction and solution structure calculation 7
2. 4. Biophysical methods for thermal, conformational, and binding analysis 7
2. 5. MutCompute 8
3. Results 9
3. 1. Thermostability analysis of SsoSSB1–114 9
3. 2. NMR assignment with high-temperature secondary structure prediction 9
3. 3. Determination of high-temperature solution structure 10
3. 4. Assessment of temperature-dependent DNA interaction 12
3. 5. Comparison of backbone dynamics at high and room temperature 12
3. 6. Analysis of SsoSSB12–114 thermostability and ssDNA binding properties 13
3. 7. Evaluation of HDX-selected mutants by NMR and thermostability experiments 14
3. 8. Evaluation of AI-selected mutants by NMR and thermostability experiments 16
4. Discussion 18
5. Figures and Tables 21
Ⅱ. Chapter 2. Biophysical Analysis of the Molecular Interaction Between
Minichromosome Maintenance Protein 6 and Bloom Syndrome Helicase 57
1. Introduction 58
2. Materials and Methods 60
2. 1. Sample preparation 60
2. 2. NMR experiments 60
2. 3. Cohort analysis 61
2. 4. Biophysical binding analysis 62
3. Results 63
3. 1. MCM6 and BLM co-expression and prognostic significance in cancer 63
3. 2. MCM6 NTD1 assignment and BLM MBD CSP analysis 63
3. 3. MCM6 WHD directly binds to BLM MBD-C and D 64
3. 4. CSP analysis of MCM6 WHD upon titration with BLM MBD-C and D 65
3. 5. PRE analysis of MCM6 WHD interaction with BLM220–300- 66
3. 6. BLM220–300 assignment and CSP analysis upon titration with MCM6 WHD 67
4. Discussion 69
5. Figures and Tables 71
Ⅲ. Chapter 3. Biophysical Characterization of the FOXO4 Forkhead Domain and W 2
Region in DNA Binding Using NMR and Molecular Dynamics Simulation 94
1. Introduction 95
2. Materials and Methods 98
2. 1. Sample preparation 98
2. 2. NMR experiments 98
2. 3. TITAN analysis 98
2. 4. Isothermal titration calorimetry 99
2. 5. Molecular dynamics simulations 99
3. Results 101
3. 1. FOXO4 FHD DNA TITAN Analysis 102
3. 2. FOXO4 FHD ΔW2 DNA interaction 102
3. 3. Structural and functional effects of the FHD S2–W1 3E mutation 104
4. Discussion 107
5. Figures and Tables- 109
Ⅳ. Appendix. FOXO4 FHD–p53 TAD Interaction Inhibitor Discovery via
Fluorescence Polarization Screening and Machine Learning Approach 123
References 133
Korean abstract 138
Curriculum Vitae 139
Acknowledgments 14
Correction: Wing Shape Optimization of Underwater Floats With Motion Constraints (American Institute of Aeronautics and Astronautics Inc, AIAA)
No full textCorrection Notice Added Yi Chao and Jongchun Park as authors, in addition to the existing authors Donggun Lee and Seongim Choi. I would like to add an Acknowledgment section after the Conclusion. The content of this section is as follows: This work was supported by the Technology Innovation Program (or Industrial Strategic Technology Development Program - Development of high-accuracy prediction method of phase change, BOR, and pressure change in the cargo hold of LNG/LH2 ships in consideration of the operating environment, 20026368) funded by the Ministry of Trade, Industry & Energy (MOTIE, Korea). © 2025 by the American Institute of Aeronautics and Astronautics, Inc. All rights reserved.FALSEscopu
Part I. Synthesis and biological evaluation of 5HT2B antagonists for liver fibrosis Part II. Identification of new KDM5B inhibitors for colorectal cancer
No full textThis work has been divided into two major parts. Part I. Synthesis and biological evaluation of 5HT2B antagonists for liver fibrosis Part II. Identification of new KDM5B inhibitors for colorectal cancer Part I. Synthesis and biological evaluation of 5HT2B antagonists for liver fibrosis Liver fibrosis is characterized by excessive accumulation of extracellular matrix components, leading to the distortion of liver architecture and function. This condition poses serious health risks by impairing the ability of the liver to detoxify and metabolize substances and can be a precursor to severe liver diseases such as cirrhosis and hepatocellular carcinoma. Notably, recent studies have shown that the expression of the 5-hydroxytryptamine receptor 2B (5HT2B) is induced during the activation of hepatic stellate cells. Antagonism of 5HT2B stimulates the apoptosis of activated hepatic stellate cells and inhibits their proliferation while concurrently regressing hepatocyte proliferation. These findings underscore the therapeutic potential of targeting 5HT2B in liver fibrosis. In this study, we present compound 19c, which demonstrates promising efficacy both in vitro and in vivo. 19c showed robust in vitro activity with an IC50 value of 1.05 nM and limited blood-brain barrier penetration. Furthermore, 19c did not significantly inhibit hERG and Cytochrome P450 enzymes. 19c markedly reduced fibrotic deposition, with a dose-dependent decrease in fibrosis stage and area in the CCl4-induced liver fibrosis mouse model. Additionally, treatment with 19c led to downregulation of key fibrosis-related genes, including α- SMA, Timp1, Col1a1, and Col3a1, in a dose-dependent manner. Taken together, these results suggest that 19c has the potential to be a novel anti-fibrotic agent. Part II. Identification of new KDM5B inhibitors for colorectal cancer Colorectal cancer (CRC) represents a significant and prevalent cause of cancer-related mortality globally, posing substantial challenges related to therapeutic resistance and metastatic progression. Evidence has identified KDM5B (histone lysine demethylase 5B, also known as JARID1B (Jumonji/AT-rich interactive domain-containing protein 1B)) as a promising therapeutic target for colorectal cancer. Notably, KDM5B expression was significantly elevated in CRC tissues, and its downregulation led to a markable decrease in CRC cell proliferation. Based on these findings, we designed and synthesized a series of arylhydrazone compounds as potential KDM5B inhibitors. Among these, compound 9c exhibited compelling in vitro activity, achieving an IC50 value of less than 1 nM. Furthermore, 9c demonstrated 30% oral bioavailability, and displayed selective cytotoxicity in multiple colorectal cancer cell lines while preserving the viability of normal cells. Compound 9c effectively reduced tumor volume and weight in syngeneic MC38 tumor model, without causing body weight loss.DoctorAbstract i
Contents 1
List of Schemes 3
List of Tables 4
List of Figures 5
Part I. Synthesis and biological evaluation of 5HT2B antagonists for liver fibrosis 6
I. Introduction 7
1.1.1. Liver fibrosis 7
1.1.2. 5-Hydroxytryptamine receptor 2B and liver fibrosis 8
1.2.1. Obesity 9
1.2.2. Tryptophan hydroxylase 1 and obesity 10
II. Results and Discussion 11
2.1.1. Designing strategy 11
2.1.2. Modification in Part A 12
2.1.3. Modification in Part B 14
2.1.4. Amino acid derivatives 16
2.1.5. BBB, hERG, CYP, selectivity, and pharmacokinetic profiles 18
2.1.6. In vivo Evaluation of 19c 19
2.2.1. Designing Stretagy and Linker position modification 21
2.2.2. Linker structure modification. 26
2.2.3. CRBN ligand part modification 30
III. Conclusion 32
3.1. Synthesis and biological evaluation of 5HTR2B antagonists for liver fibrosis 32
3.2. Design and synthesis of TPH1-PROTAC 32
IV. Experimental procedures 33
4.1.1 Chemistry 33
4.1.2. Biology 60
Part II. Identification of new KDM5B inhibitors for colorectal cancer 63
I. Introduction 64
1.1. Histone post-translational modifications (PTMs) 64
1.2. Inhibition of histone lysine demethylase 5B (KDM5B) 65
II. Results and Discussion 66
2.1. (E)-2-Benzoylpyridine hydrazone derivatives 66
2.2. Structure-activity relationship study of 4d derivatives 69
2.3. Pharmacokinetics, hERG inhibition, normal cell cytotoxicity, and selectivity profiles of
promising compounds 74
2.4. Anti-cancer efficacy of compound 9c 76
III. Conclusion 77
IV. Experimental procedures 78
V. References 99
List of Schemes
Scheme 1 12
Scheme 2 14
Scheme 3. 16
Scheme 4 Preparation of intermediates 22
Scheme 5 Preparation of Linker position A compounds 23
Scheme 6 Preparation of Linker position B compounds 24
Scheme 7 Linker modification (1) 26
Scheme 8 Linker modification (2) 27
Scheme 9 CRBN ligand modification 30
Scheme 10 Synthesis of (E)-2-benzoylpyridine hydrazone derivatives. 66
Scheme 11 Synthesis of 4d derivatives 69
Scheme 12 Synthesis of 9c derivatives 7
Fusion of Generative AI Techniques and Machine Learning Models to Generate and Investigate Biosignals for Glucose Sensors
No full textThe research presents a cutting-edge and an inexpensive technology to predict hematological parameters on the amperometric data set in a hand-held glucometer. The data set contains peak current (Ip in μA), time corresponding to the current (Tp in sec), hematocrit volume (Hv in %), glucose concentration (Gc in mg/dL), and blood viscosity at 12 s–1 (Vis_12 in cP) and 120 s–1 (Vis_120 in cP) shear rates. We deciphered an interconnection between the blood glucose concentration and hemoglobin level through the hematocrit volume of the blood by utilizing machine learning (ML) models. The ML models such as linear regression (LR), support vector regressor (SVR), decision tree (DT), random forest regressor (RFR), extreme gradient boosting regressor model (XGBoost), light gradient boosting regressor (Light GBM), and artificial neural network (ANNs) predicted Gc, Hv, Vis_12, Vis_120, Hgb, and Occ with an acceptable accuracy corroborated through statistical metrics, namely, R-squared (R2) score, mean squared error (MSE), and root-mean squared error (RMSE). The ML models were trained with 80% of the data set and validated with the remaining 20%. Furthermore, the reliability of the models were tested via relative error (RE), K-fold cross-validation technique, and 95% of confidence interval in the domain of predictive analytics. Moreover, five thousand synthetic data sets were generated by utilizing generative artificial intelligence (Gen AI) models such as Generative Adversarial Network (GAN), Variational Auto-Encoder (VAE), and Gaussian copula (Gcop), a multivariate distribution technique. Synthetic data sets were assessed by training the developed machine learning models on the synthetic data set and testing them on the original data set. This approach enabled validation of model performance by comparing the original data with the predicted outputs. The statistical metrics of the models trained and tested on the original data set were compared with the trained data set and tested on the synthetic data set. While XGBoost outperformed other models on the original data set, Light GBM surpassed all models, including XGBoost, on the Gcop-generated data set, making it the most reliable model for synthetic data applications. Our limitations lie toward the viscosity prediction on the Gcop-generated synthetic data set as corroborated through SHAP analysis. Conclusively, we are futuristically propelled to refine the generative process to produce feasible values for the viscosity variables. © 2025 The Authors. Published by American Chemical SocietyTRUEsciescopu
Semantic Hierarchy-Guided Adversarial Attack for Autonomous Driving
No full textAutonomous vehicles employ semantic segmentation as a foundational component for perception and scene understanding, upon which driving decisions can be informed. Despite their performance, these deep learning models remain susceptible to subtle input perturbations that can cause severe deviation in model output. To enhance algorithmic robustness by examining such vulnerabilities, researchers have investigated adversarial examples, which are visually imperceptible yet can severely degrade model performance. However, traditional attacks produce arbitrary misclassifications that ignore semantic relationships, making the attack less effective. This letter introduces a semantic hierarchy-guided adversarial attack (SHAA), a white-box adversarial attack against semantic segmentation for autonomous driving. By combining semantic hierarchy and adaptive momentum-based updates across the image, SHAA produces semantically nontrivial yet highly effective perturbations. The SHAA method exposes deeper vulnerabilities with a higher attack success rate in semantic segmentation than existing methods, aiding the design of a more resilient perception system for autonomous vehicles.FALSEsciescopu
I Want to Break Free: Enabling User-Applied Active Locomotion in In-Car VR through Contextual Cues
No full textWe explore the feasibility of active user-applied locomotion in virtual reality (VR) within in-car environments, diverging from previous in-car VR research that synchronized virtual motion with the car's movement. Through a two-step study, we examined the effects of locomotion methods on user experience in dynamic vehicle environments and evaluated contextual cues designed to mitigate sensory mismatch caused by vehicle motion. The first study evaluated five locomotion methods, identifying joystick-based navigation as the most suitable for in-car use due to its low physical demand and stability. The second study focused on designing and testing contextual cues that translate physical sensations of vehicle motion into virtual effects without limiting the user's freedom of movement, with results demonstrating their effectiveness in reducing motion sickness and enhancing presence. We conclude with initial insights and design considerations for expanding upon our findings in regards to enabling active locomotion in in-car VR. © 2025 Copyright held by the owner/author(s)
Ligand-Shell Cooperativity in a Bilayer Silica-Sandwiched Mixed-Metals Nanocatalyst Design for Absolute Selectivity Switch
No full textUnlike homogeneous metal complexes, achieving absolute control over reaction selectivity in heterogeneous catalysts remains a formidable challenge due to the unguided molecular adsorption/desorption on metal-surface sites. Conventional organic surface modifiers or ligands and rigid inorganic and metal-organic porous shells are not fully effective. Here, we introduce the concept of "ligand-porous shell cooperativity" to desirably switch reaction selectivity in heterogeneous catalysis. We present a nanocatalyst design strategy consisting of bilayer silica-sandwiched 2D mixed metal islands. The intimate 2D/2D nanoscale interfacing between porous silica layers and flat island-like mixed-metal sites, combined with organic ligands, creates a nanoconfined microenvironment that enables reliable control of molecular orientation-dependent reactivity, affording the desired product in 100% selectivity. This design simultaneously leverages the hydrophobicity and flexibility of organic ligands and the nanoscale geometric rigidity of the pores inside the inorganic silica shell. Our strategy is effective with simple amorphous silica, random Cu-alloy, and commonly used metal-coordinating ligands. We demonstrate the applicability in industrially significant reactions: selective hydrogenation of alkynes, alpha,beta-unsaturated esters/aldehydes, and nitroarenes. Our findings offer the valuable scope of a multicomponent compact nanoscale design strategy in next-generation switchable, sustainable, and recyclable catalysis.FALSEsciescopu
Exploring the efficient catalytic activity of mixed-phase palladium selenides in oxygen reduction reaction
No full textMixed-phase palladium selenides (PdxSey) were synthesized using a single organometallic precursor, offering precise control over the phase distribution and enabling a comprehensive study of phase-dependent electrocatalytic performance. Rietveld refinement of the X-ray diffraction data supported by inductively coupled plasma mass spectrometry and X-ray photoelectron spectroscopy analyses revealed a progressive transition from Se-rich (PdSe2) to Se-deficient (Pd34Se11) phases with increasing synthesis temperature. This phase transformation is critical for enhancing the catalytic activity of materials. Among the synthesized catalysts prepared at different synthesis temperatures, PdxSey-1000 synthesized at 1000 °C exhibited exceptional oxygen reduction reaction (ORR) performance, achieving a half-wave potential of 0.931 V and demonstrating remarkable durability, with only a 7 mV shift in half-wave potential after 20,000 accelerated durability tests. The superior catalytic activity and stability of the mixed-phase PdxSey-1000 compounds are attributed to the synergistic interactions between distinct crystalline phases, such as PdSe2, Pd17Se15, Pd4Se, and surface Pd. These interactions modulate electronic structures, fine-tune intermediate binding energies, and facilitate efficient pathway transitions during the ORR process. Theoretical calculations revealed that the coexistence of these phases optimized specific reaction steps, leveraging the unique strengths of each phase to minimize the energy barriers and enhance the overall reaction kinetics. This study highlights the transformative potential of leveraging the synergistic effects in mixed-phase compounds. It demonstrates their potential as high-performance and durable alternatives to traditional electrocatalysts. These findings pave the way for the rational design of multi-phase materials for advanced energy-conversion applications. © 2025 Elsevier B.V.FALSEsciescopu
Frustrated phonon with charge density wave in vanadium Kagome metal
No full textThe formation of a star-of-David charge density wave superstructure, resulting from the coordinated displacements of vanadium ions on a corner-sharing triangular lattice, has garnered significant attention to comprehend the influence of electron–phonon interaction within geometrically intricate lattice of Kagome metals, specifically AV3Sb5 (where A represents K, Rb, or Cs). However, understanding of the underlying mechanism behind charge density wave formation, coupled with symmetry-protected lattice vibrations, remains elusive. Here, from femtosecond time-resolved X-ray scattering experiments, we reveal that the phonon mode, associated with cesium ions’ out-of-plane motion, becomes frustrated in the charge density wave phase. Furthermore, we observed the photoinduced emergence of a metastable charge density wave phase, facilitated by alleviating the frustration. By not only elucidating the longstanding puzzle surrounding the intervention of phonons but introducing the phononic frustration, this research offers insights into the competition between phonons and periodic lattice distortions, a phenomenon widespread in other correlated quantum materials including layered high-temperature superconductors. © The Author(s) 2025.TRUEsciescopu
Imitation Learning for Autonomous Parking
No full textWith the rise of automated systems, there is a growing demand for agents that can make human-like decisions. One of the ways to train such agents is imitation learning (IL), which can converge to the optimal policy faster by referring to expert demonstrations. In this study, IL algorithms were applied as an auxiliary learning method for reinforcement learning algorithm, and the performance of IL algorithms was compared on various expert data. The experiments analyzed the characteristics of the expert data required by each IL algorithm, and the performance di↵erences are clear. In particular, the accuracy of the expert data provided to the Behavioral Cloning algorithm and the diversity of the expert data provided to Generative Adversarial Imitation Learning algorithm were emphasized. These findings contribute practical guidance for training agents through IL.MasterAbstract (English) i
Abstract (Korean) ii
List of Contents iii
List of Tables v
List of Figures vi
1 Introduction 1
1.1 Introduction 1
1.2 Motivation 3
1.3 Contribution 3
2 Background 4
2.1 Reinforcement Learning 4
2.2 Imitation Learning 6
2.2.1 Behavioral Cloning 7
2.2.2 Generative Adversarial Imitation Learning 7
2.3 ML-Agents 8
3 Method 10
3.1 PPO+BC+GAIL 10
3.2 Evaluation Metric 11
4 Experiment 12
4.1 Experimental Setup 12
4.1.1 MDP Definition 13
4.1.2 Expert Data Collection 16
4.2 Experimental Result 17
4.2.1 Experiment on Algorithm 17
4.2.2 Experiment on expert data 19
– iii –
5 Conclusion 23
5.1 Discussion 23
5.2 Limitation 23
5.3 Future Works 24
Summary 25
References 26
A Abbreviations 29
B Configuration 30
Acknowledgements 31
– iv