Welcome! I'm Young-Chae Son, a Ph.D. student at Dongguk University’s Interactive Robotics Lab, supervised by Prof. Soochul Lim.
I work on Vision-Language Models (VLMs) and Vision-Language-Action (VLA) models for task planning and manipulation on real robotic systems. My research asks how a robot can make reliable decisions when visual observations alone do not reveal enough about the current task state.
Rather than simply adding more sensory inputs, I am interested in determining which information is relevant to the task at hand and how it should influence robot behavior.
A See Selectively, Act Adaptively decomposes bimanual robot manipulation into view-selective perception and interaction-aware action routing to improve robust, long-horizon task performance.
A hierarchical VLM–VLA framework that leverages thermal information to adapt task plans to environmental conditions, enabling safer, temperature-aware robotic manipulation.
Robot task automation framework based on large language models (LLMs) that utilizes both visual and physical information.
Funded Projects
Development of Intelligent Autonomous Manipulation Technology for Humanoids with Reduced Dependency on Real-World Data Jul 2025 – Present · Ministry of Science and ICT (MSIT)
Keywords: Humanoid Manipulation, Data-Efficient Learning, Robot Foundation Models
Development and real-robot validation of a data-efficient humanoid manipulation model that reasons about intent and context to generate physical-interaction-based actions from small-scale real-world data
Use of no more than 10% of the real-world data required by conventional imitation learning
Demonstration of at least five service tasks in everyday environments
Robot Motion Generation AI based on Multimodal Vision/Tactile Information Driven by Language Model Mar 2025 – Present · National Research Foundation of Korea (NRF)
Keywords: Multimodal Robot Learning, Vision/Tactile Perception, Language Models
Development of a Robot Foundation Model that generates object manipulation actions for autonomous goal achievement based on physical understanding, by integrating physical properties (force, tactile, and robot pose information) with 2D/3D visual information and natural language instructions
Construction of multimodal datasets for training a foundation model capable of understanding physical properties, including force and tactile data
Development of an RFM that performs language-instructed object manipulation tasks based on an understanding of the physical properties of both rigid and deformable objects
Collecting Large-scale Robot Manipulation Data in Physical Environment Aug 2024 – Dec 2024 · National Information Society Agency (NIA)
Keywords: Robot Data Collection, Manipulation, NVIDIA Isaac Gym
Construction of training data for large-scale AI models and acquisition of high-quality data
Construction of over 216,000 data pairs extracted from multi-sensor videos of manipulation involving more than 200 object types, and over 100,000 Korean/English sentences (over 1 million tokens) describing robot object manipulation
Collection of robot manipulation data (images, 3D point clouds, force/torque, and tactile data) in real-world settings and in Isaac Gym-based simulation environments configured to resemble everyday environments
Personal Projects
Task Integration with a VLM Planner and VLM Monitor Young-Chae Son, in collaboration with Yoon-Ji Choi, Dae-Kwan Ko, Jung-Woo Lee
[Video]
Composing reusable VLA skills for long-horizon manipulation, with a VLM planner that sequences skills and a VLM monitor that checks each step during execution.
Real2Sim Digital Twin for VLA Data Collection Young-Chae Son [Video]
Keywords: Real2Sim Digital Twin · Robot Teleoperation · VLA Data Augmentation
A digital twin of a real tabletop scene in NVIDIA Isaac Sim, teleoperated through a Meta Quest Pro with eye tracking to collect multi-view VLA training data.
Experience & Activities
Patent
Multimodal Robot System for Task-Context-Based Selective Sensory Utilization
Apr 2026
Korean Patent Application, No. 10-2026-0073734
Method, Apparatus, System, and Computer-Readable Recording Medium for Robot Control Using Thermal Image Information
Apr 2026
Korean Patent Application, No. 10-2026-0067117
Honors & Awards
Global Outstanding Talent Scholarship (Full Tuition Waiver for Ph.D.), Dongguk UniversitySep 2025 – Present
Best Paper Award, BK21FOUR AIMS Center Winter Conference, Dongguk University2026
Grand Prize (1st Place), Transfer Student Mentoring Testimonial, Tech University of Korea2022
Grand Prize (1st Place), English Presentation Contest, Tech University of Korea2022