Hokyun Im Profile Picture

Hokyun Im

I am an MS/Ph.D student majoring in Artificial Intelligence at Yonsei University, advised by Prof. Youngwoon Lee.

My research focuses on the intersection of Robotics and AI. I am particularly interested in building intelligent agents that can reason and make decisions in the real world by leveraging Foundation Models and Reinforcement Learning.

Previously: Research intern at Microsoft Research Asia (MSRA), MLCS Lab (Prof. Jongeun Choi), and CLVR Lab (Prof. Joseph J. Lim).

GitHub X LinkedIn CV

Publications

TwinVLA: Data-Efficient Bimanual Policy through Twin Single-Arm Vision-Language-Action Models

ICLR 2026

Hokyun ImEuijin JeongAndrey KolobovJianlong FuYoungwoon Lee

We introduce TwinVLA, a vision-language-action (VLA) model for bimanual manipulation that fuses pretrained single-arm VLA models. This design reduces reliance on scarce bimanual data while achieving comparable performance.

Projects

End-to-End Visual Indoor Drone Navigation with Foundation Model

Project Lead, Yonsei Drone (2024)

Previously, distance-sensing sensors such as depth cameras were necessary for drone navigation. With the advancement of foundation models, we successfully accomplished indoor navigation using only a front-facing RGB image by leveraging the CrossFormer model.

Autonomous Outdoor Drone Navigation(KRAC 2023 2nd Place)

Yonsei Drone (2023)

Fully autonomous outdoor drone navigation using GPS to travel long distances, avoid obstacles, reach the destination to deliver items, and return safely.

I led the development of building approach algorithms, and worked on safe landing using Sim2Real RL.

Other Projects & Activities

Graduate
Tabletop Sim
Tabletop Simulation

dm_control based bimanual tabletop simulation / dataset for testing bimanual robot policy.

Undergraduate
Yonsei Drone
Yonsei Drone

A student-led drone research society. I served as the 4th president.

YAI
YAI

The leading academic society for AI at Yonsei University. I was a 11th member.

TensorFlow.js
VAE Visualization

A toy project that allows interactive, real-time manipulation of the latent representation of a VAE on a webpage. Implemented using TensorFlow.js.

HTML Games
HTML Games

HTML games developed in JavaScript during spare time in the military. Quite fun to play haha.

PID Pendulum
PID Pendulum

A simple pendulum simulator designed as an educational tool. It allows students to explore and understand the principles of PID control.