Hyunseok Kim

Electronics and Telecommunications Research Institute, Dong-A University

Papers

Total Citations

H-Index

About

Hyunseok Kim is a researcher advancing the frontiers of robot learning, with a focus on making autonomous systems more capable through demonstration, simulation, and reinforcement learning. His work centers on three key areas: behavioral cloning, reward engineering, and uncertainty-aware control. In his foundational 2020 paper on robotic behavioral cloning through task building, Kim demonstrated how robots can efficiently learn policies by directly mapping from human demonstrations to actions—a paradigm that expands the practical applications of robotics. That same year, he pioneered methods for learning control policies from robot simulators, showing how simulated environments can cost-effectively train physical robot actions, with reward shaping as a critical component. His most recent 2025 work introduces entropy-guided distributional reinforcement learning, a novel approach that controls uncertainty and mitigates overestimation bias in long-horizon robotic tasks. Across his publications—garnering over a dozen citations—Kim consistently addresses the core challenge of bridging simulation to reality while improving learning stability. His contributions are particularly valuable for researchers working on manipulation tasks and sample-efficient robot training.