https://www.youngju.dev/blog 천천히 올바르게. AI Researcher & DevOps Engineer Youngju's tech blog. GPU/CUDA, LLM, MLOps, Kubernetes AI workloads, distributed training, and data engineering. ko fjvbn2003@gmail.com (Youngju Kim) fjvbn2003@gmail.com (Youngju Kim) Sat, 27 Jun 2026 00:00:00 GMT https://www.youngju.dev/blog/ai-papers/2026-06-27-diffusion-policy-and-pi0.en https://www.youngju.dev/blog/ai-papers/2026-06-27-diffusion-policy-and-pi0.en Going beyond the limits of discrete action tokens, we examine two streams that generate actions as continuous values. Diffusion Policy generates actions via denoising, and π0 produces high-frequency continuous actions with flow-matching. We organize the ideas, architectures, control frequency, strengths, and limitations of both. Sat, 27 Jun 2026 00:00:00 GMT fjvbn2003@gmail.com (Youngju Kim) ai-papersroboticsdiffusion-policypi0flow-matchingimitation-learningmanipulation https://www.youngju.dev/blog/ai-papers/2026-06-27-diffusion-policy-and-pi0.ja https://www.youngju.dev/blog/ai-papers/2026-06-27-diffusion-policy-and-pi0.ja 離散行動トークンの限界を超えて、行動を連続値で生成する二つの流れを見ていきます。Diffusion Policyは行動をデノイジングで生成し、π0はflow-matchingで高周波の連続行動を作ります。両アプローチの考え方、アーキテクチャ、制御周波数、長所と限界を整理します。 Sat, 27 Jun 2026 00:00:00 GMT fjvbn2003@gmail.com (Youngju Kim) ai-papersroboticsdiffusion-policypi0flow-matchingimitation-learningmanipulation https://www.youngju.dev/blog/ai-papers/2026-06-27-diffusion-policy-and-pi0 https://www.youngju.dev/blog/ai-papers/2026-06-27-diffusion-policy-and-pi0 이산 액션 토큰의 한계를 넘어, 행동을 연속값으로 생성하는 두 흐름을 살펴봅니다. Diffusion Policy는 행동을 디노이징으로 생성하고, π0는 flow-matching으로 고주파 연속 액션을 만듭니다. 두 접근의 아이디어, 아키텍처, 제어 주파수, 강점과 한계를 정리합니다. Sat, 27 Jun 2026 00:00:00 GMT fjvbn2003@gmail.com (Youngju Kim) ai-papersroboticsdiffusion-policypi0flow-matchingimitation-learningmanipulation