Jihwan Yoon

Research Scientist, Robot Intelligence Lab (Korea University)

A portrait of Jihwan Yoon

Education

Korea University
Integrated Ph.D course in Artificial Intellegence, advised by Sungjoon Choi and Kyungjae Lee.
Spring 2024 -
Korea University
B.S. in Mechanical Engineering
Spring 2018 - Winter 2024

Employment

Publications

LEGO: Latent-space Exploration for Geometry-aware Optimization of Humanoid Kinematic Design Learns a compact, geometry-preserving latent space of humanoid upper-body designs from existing mechanical designs and uses human motion data - via motion retargeting and Procrustes analysis - as the optimization objective, enabling automated discovery of new robot kinematics with minimal human design input.
ICRA, 2026
Learning Social Navigation from Positive and Negative Demonstrations and Rule-Based Specifications Social navigation framework that learns a density-based reward from positive and negative demonstrations and augments it with rule-based objectives for obstacle avoidance and goal reaching. A sampling-based lookahead controller produces safe yet adaptive supervisory actions, which are distilled into a compact student policy suitable for real-time operation with uncertainty estimates.
ICRA, 2026
Hierarchical Vision Language Action Model Using Success and Failure Demonstrations Hierarchical vision-language-action model that learns from both success and failure demonstrations: a high-level System 2 performs feasibility-guided tree search over scene-graph subgoals using success/failure probabilities, while a low-level System 1 executes the selected actions - turning failure data into a structured signal for robust manipulation.
CoRL 2025 Workshop on Safe and Robust Robot Learning
CoRe: A Hybrid Approach of Contact-Aware Optimization and Learning for Humanoid Robot Motions An automated pipeline that converts text-generated human motions into physically executable humanoid motions: text-to-motion generation, robot-specific retargeting, optimization-based motion refinement, and a contact-aware RL phase - reducing foot sliding, unnatural floating, and excessive joint accelerations before RL training.
Humanoids, 2025