Hao-Lun Hsu

Georgia Institute of Technology, Duke University

Papers

Total Citations

H-Index

About

Hao-Lun Hsu is a researcher advancing the safety and reliability of reinforcement learning (RL), with a focus on deep RL and offline decision-making. His work addresses critical challenges in deploying RL in real-world robotics and autonomous systems. In his highly cited 2022 paper, "Improving Safety in Deep Reinforcement Learning using Unsupervised Action Planning" (7 citations), Hsu introduced a novel technique that enhances safety during both training and testing phases of on-policy RL algorithms like trust region policy optimization. This contribution is pivotal for applications where unsafe actions can have severe consequences. More recently, in his 2024 paper "Steering Decision Transformers via Temporal Difference Learning" (2 citations), Hsu tackles the limitations of Decision Transformers in stochastic environments—a common hurdle in robotics. By integrating temporal difference learning, his work improves the robustness and adaptability of offline RL, enabling more reliable sequence modeling from demonstrations. Hsu’s research bridges the gap between theoretical RL advances and practical deployment, making him a notable figure in safe and efficient learning systems. His contributions are shaping the future of autonomous decision-making under uncertainty.