I am an M.S. student at KAIST's AMI (Advanced Machine Intelligence) Lab advised by Prof. Tae-Hyun Oh. I work on computer vision and multimodal learning with a focus on embodied AI. My interests span the perception-to-action stack, emphasizing 3D scene understanding and scene/task representations that support reliable long-horizon planning and execution.

Looking ahead, I aim to build human-like embodied systems that can generate high-level plans from instructions and perception, acquire new atomic skills, refine them at the action level from both observation and interaction using multimodal perceptual signals, and hierarchically compose these skills to execute the plans.

• DarkEQA: Benchmarking Vision-Language Models for Embodied Question Answering in Low-Light Indoor Environments
  Yohan Park, Hyunwoo Ha, Wonjun Jo, Tae-Hyun Oh
  Under review.
  

• Reactive Constraint Relaxation for Urban Environment Navigation
  Jinwoo Kim, Keonyoung Koh, Samuel Seungsup Lee, Yohan Park, Daehyung Park
  In International Conference on Robot Intelligence Technology and Applications, pp. 219–230 (Springer, 2024).
  Best Student Paper Award.
  

• Embodied AI (perception-to-action; language-conditioned behavior)
• 3D scene understanding and geometry-aware representations

• KAIST, Daejeon, South Korea [Mar 2026 – Current]

  • M.S. student, AMI (Advanced Machine Intelligence) Lab (Advisor: Prof. Tae-Hyun Oh)

• KAIST, Daejeon, South Korea [Feb 2019 – Aug 2025]

  • B.S., Electrical Engineering & Computer Science (Double major)

• Hansung Science High School, Seoul, South Korea [Mar 2016 – Feb 2019]

• Research Intern, AMI lab @ KAIST SoC [Apr 2025 – Feb 2026] Advised by Prof. Tae-Hyun Oh

  • Built DarkEQA, a controlled benchmark that disentangles low-light exposure loss and sensor noise, enabling evaluation of VLMs on EQA-relevant perceptual primitives
  • Developing a depth-aware perception module for embodied agents (ongoing; planned submission in 2026)

• Research Intern, RIRO lab @ KAIST SoC [Dec 2023 – Feb 2025] Advised Prof. Daehyung Park

  • Fine-tuned a semantic segmentation model for KAIST campus navigation
  • Designed and implemented a scalable, multimodal ROS navigation planning architecture for Boston Dynamics' Spot, integrating Behavior Trees to improve reliability and robustness
  • Built an RViz plugin for waypoint navigation, streamlining the saving, visualization, and transmission of waypoints and routes to Behavior Trees, improving usability for robotic path planning
  • Analyzed move_base’s local navigation plugin by investigating the ROS navigation stack and optimizing parameters for the global_planner and dwa_local_planner; replaced the latter with a 3rd-party local trajectory generation module

• Programming languages

• Frameworks

• Tools

• OS

English (Proficient), Chinese (Intermediate), German (Beginner), Korean (Native)

• 'Yohan (요한)' is the Korean translation of 'John,' and my legal Korean name is Yohan Park (박요한). Please feel free to call me John.

• Please checkoput my homepage! (It's "yohanpark.me")

💬 Also, please feel free to reach out to me: