Haotian Fu

Papers

Total Citations

H-Index

About

Haotian Fu is a rising researcher advancing the frontiers of reinforcement learning (RL), with a focused expertise in hierarchical and meta-reinforcement learning. His work tackles a core challenge in AI: enabling agents to learn and adapt efficiently across diverse, long-horizon tasks. Fu’s seminal contribution, the **MGHRL (Meta Goal-Generation for Hierarchical Reinforcement Learning)** framework, addresses the limitations of standard meta-RL methods that struggle with wide task distributions. By introducing a meta-learned goal generator, his approach allows agents to dynamically set and pursue sub-goals, significantly improving transfer and adaptation. This work, published in both 2019 and 2020, has garnered early citations, signaling its growing influence. Building on this, his 2022 paper on **Meta-Learning Parameterized Skills** proposes a novel algorithm that learns transferable, parameterized skills. These skills are synthesized into a new action space, enabling efficient learning in complex, long-horizon tasks. Fu’s approach leverages off-policy meta-RL combined with trajectory-centric smoothness, a technical innovation that enhances stability and performance. Though early in his career, Haotian Fu’s research is carving a clear path toward more general, adaptable, and sample-efficient AI systems, making his work essential reading for students and researchers in RL and robotics.