LLM-based
Decision Making
Linking language understanding to planning, instruction following, and learned action.
Language · Planning · AgentsUNIST · AI GRADUATE SCHOOL & ELECTRICAL ENGINEERING
Machine Learning &
Intelligent Control.
We build the foundations of intelligent agents through reinforcement learning, language-based reasoning, and learning-driven control.
RESEARCH DIRECTIONS
Our research areas.
LLM-based
Physical
Intelligence
Learning from demonstrations and sensory input to control embodied systems across simulated and real environments.
Embodiment · Adaptation · ControlCore Reinforcement
Learning
Developing the algorithms and theory behind data-efficient, adaptive, and reliable sequential decisions.
Offline · Multi-agent · Safe RLApplied Reinforcement
Learning
Bringing learned decision policies to autonomous systems, industrial processes, and operational optimization.
Autonomy · Industry · OptimizationPUBLICATIONS
Lab publications.
12 publications
Focusing Influence Mechanism for Multi-Agent Reinforcement Learning
FIM
Focus on under-explored state dimensions and sustain collective influence to improve coordinated exploration under sparse rewards.
Escaping Path Mirages in Offline Goal-Conditioned Reinforcement Learning
BGP
Research on offline goal-conditioned reinforcement learning and the challenge of reaching long-horizon goals.
MARBO: Relational Belief Grounding for LLM Agents in Social Deduction Games
MARBO
Ground strategic actions and in-game speech in reliable beliefs about hidden roles and team relationships.
Interaction-Breaking Adversarial Learning Framework for Robust Multi-Agent Reinforcement Learning
IBAL
Train teams to maintain cooperation when adversarial perturbations disrupt interactions between agents.
LLM-Guided Communication for Cooperative Multi-Agent Reinforcement Learning
LMAC
Use an LLM to design communication protocols that help partially observable agents reconstruct the shared state.
Self-Improving Skill Learning for Robust Skill-based Meta-Reinforcement Learning
SISL
Refine reusable skills from noisy demonstrations for reliable adaptation to new, long-horizon tasks.
Strict Subgoal Execution: Reliable Long-Horizon Planning in Hierarchical Reinforcement Learning
SSE
Make long-horizon plans more reliable by separating feasible subgoals from unreachable ones.
Retaining Suboptimal Actions to Follow Shifting Optima in Multi-Agent Reinforcement Learning
S2Q
Keep alternative high-value actions available as the best joint strategy changes during training.
Wolfpack Adversarial Attack for Robust Multi-Agent Reinforcement Learning
WALL
Study coordinated attacks on cooperating agents and learn team policies that can withstand them.
Task-Aware Virtual Training: Enhancing Generalization in Meta-Reinforcement Learning for Out-of-Distribution Tasks
TAVT
Learn task representations and train on virtual tasks to generalize beyond the training distribution.
Exclusively Penalized Q-learning for Offline Reinforcement Learning
EPQ
Apply value penalties selectively to reduce estimation bias in offline reinforcement learning.
FoX: Formation-aware exploration in multi-agent reinforcement learning
FoX
Guide exploration through diverse team formations in partially observable cooperative tasks.
No publications match your search. Try another title, author, or topic.
ABOUT THE LAB
Machine Learning &
Intelligent Control.
Led by Prof. Seungyul Han, MLIC connects fundamental reinforcement learning with language-based agents, embodied intelligence, and practical control. Our research spans algorithms, theory, and applications.
AI Graduate School & Department of Electrical Engineering
Ulsan National Institute of Science and Technology