Guiliang Liu

Papers

Total Citations

H-Index

About

Guiliang Liu is an emerging researcher working at the intersection of reinforcement learning, inverse reinforcement learning (IRL), and robotic manipulation. His work addresses one of the most pressing challenges in deploying AI agents in real-world environments: ensuring that autonomous systems can learn and respect constraints without requiring those constraints to be explicitly programmed. In his 2022 paper "Learning Soft Constraints From Constrained Expert Demonstrations," Liu advances IRL methodology by recognizing that expert agents often operate under implicit constraints, not merely reward-maximizing objectives — a subtle but critical distinction for safe AI deployment. Complementing this, his benchmarking work on constraint inference in IRL provides the research community with standardized tools to evaluate how well RL agents can infer real-world constraints from demonstrations, an invaluable contribution for reproducible progress in safe reinforcement learning. More recently, his 2025 work on one-shot bimanual robotic manipulation from video demonstrations signals a compelling expansion into physical robotics, tackling the difficult problem of teaching robots complex dual-arm tasks with minimal supervision. Though early in his citation trajectory, Liu's research agenda positions him as a thoughtful contributor to the critical field of safe and efficient robot learning.