Skip to content

About Me

I'm Yunsung Lee, Head of Research on MaumAI's WoRV team and an Adjunct Professor at DGIST, where I teach undergraduate coursework in Physical AI.

Who I Am

I'm an AI researcher and engineer working in machine learning and computer vision. What keeps me in the field is the distance between a result on paper and a system people actually use. Most of my work lives in that gap.

What I Do

  • Research leadership: At MaumAI I lead WoRV (World Model for Robotics and Vehicle Control), building foundation models that tie language, vision, and action together for robots and autonomous vehicles.
  • PonderPounce: As a co-corresponding author, I work on episode memory and reasoning for robot control. PonderPounce-9B reached #1 on the official RoboMME Regular Leaderboard in September 2026 with 60.83% overall success, 16.32 percentage points ahead of the next model. The paper was accepted to the 8th Robot Learning Workshop @ NeurIPS 2026. Research, code, and demos →
  • Teaching: At DGIST's School of Undergraduate Studies I teach Physical AI (HSS118, Fall 2026).
  • Publishing: 10+ papers at venues including ICLR, ECCV, and CoRL, with 2.1K+ citations. The recent ones are on the homepage and the full list is in my CV.
  • Shipping products: Before WoRV I built agent memory, personalization, and multimodal features for wrtn, and earlier worked on multimodal dialogue at Scatter Lab and vision for education at Riiid. Those products reached 5M+ monthly active users.
  • Community: I review for CVPR, ICCV, NeurIPS, ECCV, ICLR, ICML, and ACL Rolling Review (ICML 2026 Gold Reviewer), and I give talks on robotics foundation models, world models, and research careers.

Research Interests

  • Vision-language-action models: Extending multimodal AI from seeing and reading to acting in the real world
  • Diffusion models: Generative modeling and training-free guidance
  • Embodied and agentic AI: Agents that perceive, reason, and act, from desktops to robots and vehicles
  • Robotics foundation models: Large-scale pretraining for skills that generalize across tasks and embodiments

Background

I earned my M.Sc. in Computer Science at Korea University, advised by Seungryong Kim and Jaegul Choo, and spent the first half of 2020 as a visiting scholar at Carnegie Mellon University's Language Technologies Institute. Along the way I interned with NAVER CLOVA's OCR team and Hyperconnect's ML team. The full record is in my CV.

Beyond Code

Outside of work:

  • Reading papers with PR12, TensorFlow Korea's advanced study group, going on six years now
  • Maintaining open-source tooling for open-world agents at open-world-agents
  • Mixing drinks; I hold Korea's national Craftsman Bartender license

Get in Touch

I'm glad to hear from researchers, students, and anyone building with embodied or multimodal AI.