Hengyuan Hu

Papers

Total Citations

H-Index

About

Hengyuan Hu is a researcher working at the intersection of reinforcement learning (RL), imitation learning, and commonsense reasoning for robotics. His major contributions focus on bridging the gap between data-efficient imitation learning and the generalization capabilities of RL. In his highly cited work, "Imitation Bootstrapped Reinforcement Learning" (2024, 13 citations), Hu proposes a novel framework that uses a small set of expert demonstrations to bootstrap RL, significantly improving sample efficiency in robotic control tasks—a critical step toward deploying RL in real-world settings where data is scarce. He also explores the challenge of grounded commonsense reasoning in "Toward Grounded Commonsense Reasoning" (2024, 12 citations; 2023, 3 citations), where he investigates how robots can understand nuanced, context-dependent human expectations—such as not disassembling a Lego sports car during a tidying task. By leveraging large language models (LLMs) to infuse robots with common sense, Hu’s work addresses a fundamental bottleneck in autonomous decision-making. His research has already garnered over 28 citations, reflecting its growing impact on both the RL and robotics communities. Hu’s work is particularly notable for its practical focus on making robots more adaptable and intelligent in unstructured human environments.