Compacting Vision-Language Semantics into an Actionable Space for Robot Navigation
Under review, 2026
paper soon
Jintae Park
sonjt00@kaist.ac.kr
I'm an M.S./Ph.D. student at KAIST AI, advised by Professor Jaegul Choo.
My current research interests lie in spatial reasoning for action policies, so that robots and agents can read a scene and act on it.
- Spatial Reasoning
- Action Policies
- Robotics
- Agents
Publications
ReDesign: Recovering Editable Design Structures from Images via Agentic Decomposition
European Conference on Computer Vision (ECCV), 2026
datasetODPG: Outfitting Diffusion with Pose Guided Condition
International Conference on Computer Vision Theory and Applications (VISAPP), 2025 · Oral
Ongoing Research
Tackling Long-Horizon Robot Action Sparsity
In progress, 2026
Long-horizon VLA policies fail where training data covers robot and object poses too sparsely. A modular loop finds these bottlenecks, synthesises recovery trajectories there, and keeps patching the sparsity so long-horizon rollouts stay robust.
Experience
-
Samsung Electronics Research collaboration · Main participant
Building vision models for automated defect detection and root cause analysis
-
KIST Center for Artificial Intelligence Korea Institute of Science and Technology · Student researcher
Enhancing spatial reasoning of navigation models
-
KAIST AI Undergraduate research intern · Prof. Jaegul Choo
-
Republic of Korea Air Force Mandatory military service
Education
-
KAIST AI Integrated M.S./Ph.D. · Advisor: Prof. Jaegul Choo · GPA 4.1 / 4.3
-
Korea University B.S. in Business, Computer Science (double major) · GPA 4.29 / 4.5 (Major GPA 4.46)
Awards
- Grand Prize, 25th Campus CEO Startup Competition, Korea University
- 1st place, research poster session, 2024 Korea University AI Forum
- 3rd place, Startup Competition, Korea University Technology Commercialization Center