Shallow-π: Knowledge Distillation for Flow-based VLAs
IROS 2026 Best Paper Award Finalist10 of 1,585 accepted (top 0.6%) · 4,348 submitted
Taehan Kim
AI/ML Research Engineer · Ludo Robotics
I'm a research engineer working on learning-based robot manipulation and humanoids. At Ludo Robotics, I'm building a foundation model for social humanoid interaction.
Before that, I spent four years at Samsung Research, working on vision-language-action (VLA) models, imitation-learning policies, and RL fine-tuning for precise manipulation and humanoid control. I also built the sound source localization system for Samsung Ballie.
Robots I've worked with
Work Experience
All work experience →
Aug 2026 - Present
Building a foundation model for social humanoid interaction.
Jul 2022 - Aug 2026
VLA models and RL fine-tuning for precise manipulation and humanoid control, imitation learning for bi-manual assembly, and sound source localization for Samsung Ballie.
Education
Yonsei University·B.S. in Mechanical Engineering
Mar 2014 - Aug 2021
Publications
All publications →
Diffusion-EDFs: Bi-equivariant Denoising Generative Modeling on SE(3) for Visual Robotic Manipulation
CVPR 2024 Highlight324 of 2,719 accepted (top 11.9%) · 11,532 submitted
Projects
All projects →
erAIser
Removing unwanted objects from video with segmentation, inpainting, and animation models. I implemented the VINet-based inpainting stage.
Multi-Sensor Fusion for Robotic Arms
A multi-modal VAE that fuses RGB-D images, joint states, and pressure data for reinforcement learning of a robotic arm (graduation thesis).
ICRA 2019 RoboMaster AI Challenge
Led Yonsei's team Roboin to 3rd place among 39 teams. Navigation (DWA), YOLOv3 perception, and RL-based behavior policies.