Loading...

학술행사

Korean AI Association

학술행사

  >   학술행사   >   국내학술대회

국내학술대회

연사 및 초록 (25일)
 
장영재 교수(KAIST)
 
Title:  제조 피지컬 AI & SDF
 
Abs:
본 세미나에서는 제조 피지컬 AI와 소프트웨어 중심적 공장 (SDF)에 대한 개념과 실제 사례를 소개한다.
AI의 물리적 요소를 고려한 시스템을 피지컬 AI라 정의하고 있다. 자율주행 자동차 및 로봇이 피지컬 AI의 대표적인 사례다.
그러나 제조 피지컬 AI는 거대 시스템과 다양한 복합 장비를 다룸으로 그 구성과 운영의 차별성이 필요하다.
이러한 차별점의 핵심이 바로 SDF이다. 소프트웨어 중심적 설계를 통해 공장을 하나의 거대한 AI시스템으로 설계하고 이를 기반으로 자율운영을 가능하게하는 것이 SDF의 본질이다.
실제 국내 사례를 통해 이들 개념 및 자율공장이 어떻게 구축되었는지 살펴볼 것이다.
 
Bio: 
장영재 교수는 산업 및 시스템 공학과에서 "스마트 팩토리" 및 "지능형 물류 및 공급사슬망 시스템" 관련 연구를 진행하고 있다. 또한 2025년 12월에 설립된  <카이스트 제조 피지컬 AI 연구소> 소장을 역임하고 있다.
카이스트 부임 전 미국 반도체 메모리 제조사인 마이크론 테크놀로지 (Micron Technology)에서 4년간 현장에서 공장 자동화 및 운영 관련 업무를 수행하였다. MIT 공대 기계 공학과에서 박사학위를 취득했다. 2020년에는 연구실 박사 졸업생 4명과 함께 <다임리서치>를 창업했다.
현재 국제저널인 Computers and Industrial Engineering (SCIE:2.62) 의 부편집장 (Associate Editor)를 맡고 있으며 International Journal of Production Research, IEEE Power Electronics 등 국제 저널 특별호의 편집장 및 부 편집장과,  반도체 운영 전문가들의 국제 학회인 (International Symposium on Semiconductor Intelligence) 및 Winter Simulatoin Conference MASM의 학회장을 역임했다. 
 
 
이건명 교수(충북대학교)
 
Title:  Overview of VLAs in Physical AI
 
Abs: 
With demos of humanoids such as Atlas, Optimus V3, and Unitree G1 at CES 2026, along with the unveiling of concrete humanoid commercialization roadmaps, interest in Physical AI is rapidly intensifying. This lecture provides an overview of Physical AI and examines the key building blocks of Vision–Language–Action (VLA) models, the most representative form of foundation models for Physical AI.
The talk also introduces major VLA models, including open-source systems such as OpenVLA, Octo, GR00T N1, Pi_0, Pi_0.5, Pi_0.6, SmolVLA, and X-VLA, as well as notable closed models such as Google Robotics and FigureAI Helix. Finally, it covers the open-source development ecosystem LeRobot and discusses the roles of world models and vision foundation models in enabling robust, scalable robot learning and control.
 
Bio: 
Keon Myung Lee is a professor in the School of Computer Science at Chungbuk National University. He earned his B.S., M.S., and Ph.D. in Computer Science from KAIST. He conducted postdoctoral research at INSA de Lyon in France and worked as a researcher at Park Scientific Instruments in the United States. He has also held visiting appointments as a visiting professor at University of Colorado Denver and as a visiting scholar at Indiana University. He currently serves as President of the Korean Institute of Intelligent Systems.
 
 
주한별 교수(서울대학교)
 
Title:  Towards Capturing Everyday Movements to Scale Up and Enrich Human Motion Data
 
Abs: 
Equipping AI and robotic systems with the ability to understand human behavior in everyday life is essential for enabling them to better assist people across a wide range of applications. However, the availability of high-quality human motion data for learning such knowledge remains extremely limited.

In this talk, I will present our lab’s efforts to scale and enrich 3D human motion datasets by capturing everyday human movements and natural human-object interactions. I will first introduce ParaHome, our new multi-camera system designed to capture human-object interactions in natural home environments. Next, I will present MocapEvery, a lightweight and cost-effective motion capture solution that uses two smartwatches and a head-mounted camera to enable full-body 3D motion capture across diverse settings. Finally, I will discuss our recent work that enables machines to model comprehensive affordances for 3D objects by leveraging pre-trained 2D diffusion models, allowing for unbounded object interaction capabilities.
 
Bio: 
Hanbyul Joo is an assistant professor at Seoul National University (SNU) in the Department of Computer Science and Engineering. Before joining SNU, Hanbyul was a Research Scientist at Facebook AI Research (FAIR), Menlo Park. Hanbyul received his PhD from the Robotics Institute at Carnegie Mellon University, Hanbyul is a recipient of the Samsung Scholarship and the Best Student Paper Award in CVPR 2018.
 
 
김유성 교수(성균관대학교)
 
Title:  Beyond Imitation: Self-Improving Vision–Language–Action Models
 
Abs:
Vision–Language–Action (VLA) models have rapidly advanced by learning robot behaviors from large-scale demonstration data.
However, in real-world physical environments, contact, uncertainty, and environmental variations are inevitable, revealing the limitations of imitation learning alone.
In this talk, we focus on how VLA models pretrained on demonstrations can move beyond imitation through experience gathered during execution.
We introduce reinforcement learning–based self-improvement approaches that leverage online interaction and human-in-the-loop feedback as learning signals.
Through fine-grained manipulation tasks, we discuss why self-improvement is a core capability of Physical AI.
 
Bio:
Yusung Kim is an Associate Professor at Sungkyunkwan University. He received his Ph.D. in Computer Science from KAIST. He was a Visiting Researcher at North Carolina State University and worked as a Senior Researcher at Samsung Electronics. His research focuses on reinforcement learning, imitation learning, and vision-language-action (VLA) models for embodied and physical AI systems.