Seonghyun Kim

Electronics and Telecommunications Research Institute

Papers

Total Citations

H-Index

About

Seonghyun Kim is a researcher advancing the frontiers of deep reinforcement learning for robotics, with a focus on bridging the gap between simulation and real-world application. His work centers on developing cost-efficient training methodologies for robotic control policies, particularly through the innovative use of reward engineering. Kim’s key contributions include demonstrating how learning control policies in simulated environments can dramatically reduce the expense and risk of training physical robots, a critical step toward scalable automation. His most cited paper, “Learning Control Policy with Previous Experiences from Robot Simulator” (2020, 4 citations), introduces a framework that leverages prior simulated experiences to accelerate policy learning. In a related study, “Learning Robot Manipulation based on Modular Reward Shaping” (2020, 2 citations), he explores modular reward structures to enhance deep reinforcement learning in discontinuous action spaces, drawing parallels to human-level performance in complex tasks like Atari games. Though his citation counts are modest, Kim’s work represents foundational steps in making robot learning more accessible and efficient, with potential implications for manufacturing, service robotics, and autonomous systems. His research continues to inspire new approaches to reward design and simulation-to-reality transfer.