DiffDAIL: Diffusion-Enhanced Vision-Guided Imitation Learning with Discrete Latent Representations = DiffDAIL: 이산 잠재 표현을 활용한 확산 강화 비전 가이드 모방학습

다국어 초록 (Multilingual Abstract)

Robotic imitation learning often struggles with high-dimensional sensory in-puts, noisy demonstrations, and computational inefficiency in policy learning. We present Discrete Action Imitation Learning enhanced with Diffusion (DiffDAIL), a vision-guided framework that combines discrete action encoding and diffusion-based representation learning to address these challenges. By discretizing latent embeddings, DiffDAIL reduces computational cost, while a diffusion-based denoising loop improves robustness and captures heterogeneous action distributions. Transformer-based architecture further models long-range dependencies in sequential decision-making tasks. Experimental results across benchmark manipulation tasks demonstrate that DiffDAIL outperforms conventional VAE-based and diffusion methods, achieving higher task success rates and improved data efficiency. These results highlight DiffDAIL as a scalable, robust, and resource-efficient approach for vision-guided robotic imitation learning. Keywords: Imitation Learning, Vision-Guided Robotics, Discrete Latent Encoding, Diffusion Models, Transformer Networks, Sequential Decision Making.

목차 (Table of Contents)

• 1 INTRODUCTION 2
• 1.1 BACKGROUND AND MOTIVATION 2
• 1.2 PROBLEM STATEMENT 3
• 1.3 RESEARCH OBJECTIVES AND CONTRIBUTIONS 4
• 1.4 THESIS OUTLINE 5
• 2 RELATED WORKS 6
• 2.1 IMITATION LEARNING: EVOLUTION, ACHIEVEMENTS, AND CHALLENGES 6
• 2.2 REPRESENTATION LEARNING FOR ROBOT TRAJECTORIES 8
• 2.3 TRANSFORMERS FOR SEQUENTIAL DECISION-MAKING 9
• 2.4 DIFFUSION MODEL FOR GENERATIVE POLICY LEARNING 12
• 2.5 COMPARATIVE ANALYSIS 13
• 3 PRELIMINARIES 14
• 3.1 BEHAVIORAL CLONING 14
• 3.2 DISCRETE REPRESENTATION LEARNING WITH VECTOR QUANTIZATION 15
• 3.3 DIFFUSION METHODS 15
• 3.4 TRANSFORMER SEQUENTIAL MODELLING 16
• 4 METHODOLOGY 18
• 4.1 DATA ENCODING 19
• 4.2 VECTOR QUANTIZATION 20
• 4.3 ACTION ENCODER-DECODER 22
• 4.3.1 FORWARD DIFFUSION 22
• 4.3.2 TRANSFORMER-BASED ACTION ENCODER 22
• 4.3.3 REVERSE DIFFUSION(DENOISING) 23
• 4.3.4 ACTION DECODER 24
• 4.4 ALGORITHM 26
• 5 EXPERIMENTAL RESULTS 27
• 5.1 BENCHMARKS AND BASELINES 27
• 5.2 EVALUATION RESULTS ON LIBERO BENCHMARKS 29
• 5.3 ADDITIONAL EXPERIMENTS 30
• 5.4 COMPUTATIONAL EFFICIENCY 34
• 5.5 ABLATION STUDIES 36
• 6 CONCLUSION 38