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 MLCS Lab (Prof. Jongeun Choi), and CLVR Lab (Prof. Joseph J. Lim).

GitHub X LinkedIn CV

Publications

Latent Policy Steering through One-Step Flow Policies

Preprint

Hokyun Im • Andrey Kolobov • Jianlong Fu • Youngwoon Lee

Paper Project page Code

We introduce Latent Policy Steering (LPS), a method that enables high-fidelity policy improvement by backpropagating original-action-space Q-gradients through a differentiable one-step MeanFlow policy to optimize a latent actor. By eliminating proxy latent critics and using the MeanFlow policy as a behavior-constrained generative prior, LPS achieves robust, state-of-the-art performance in offline RL with minimal tuning.

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

ICLR 2026

Hokyun Im • Euijin Jeong • Andrey Kolobov • Jianlong Fu • Youngwoon Lee

Paper Project page Code

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.