Haonan Yu

Horizon Robotics (China), Ministry of Industry and Information Technology, Purdue University West Lafayette

Papers

Total Citations

H-Index

About

Haonan Yu is a researcher at the forefront of embodied AI, focusing on the intersection of hierarchical reinforcement learning, robot perception, and natural language grounding. His most impactful work, "Hierarchical Reinforcement Learning By Discovering Intrinsic Options" (2021, 24 citations), introduces HIDIO, a method that enables robots to autonomously discover reusable skills—or "options"—in a self-supervised manner, dramatically improving learning efficiency in sparse-reward environments. This contribution addresses a fundamental challenge in scaling reinforcement learning to complex, real-world tasks. Yu also advances robust robotic perception. His work on DAT-SLAM (2023, 9 citations) tackles the fragility of visual SLAM in challenging indoor conditions, such as poor lighting or feature-sparse scenes, by fusing template-matching visual odometry for more reliable localization. Earlier, in "Driving Under the Influence (of Language)" (2017, 8 citations), he pioneered a unified framework for grounding natural language semantics in robotic driving, enabling robots to learn, generate, and comprehend spatial language from human annotations—a key step toward human-robot communication. With additional contributions in active object recognition, Yu’s work consistently bridges perception, learning, and language, making him a notable figure in creating more autonomous and interactive robots.