Yihao Feng

The University of Texas at Austin, Salesforce (United States)

Papers

Total Citations

H-Index

About

Yihao Feng’s research lies at the intersection of reinforcement learning, robotics, and statistical machine learning, with a focus on sample efficiency, lifelong learning, and reliable policy evaluation. His most impactful work, the LIBERO benchmark (2023, 15+ citations), has become a foundational tool for studying knowledge transfer in lifelong robot learning, addressing how generalist agents can adapt across diverse manipulation tasks without catastrophic forgetting. Feng also advanced off-policy evaluation (OPE) with his work on Kernel Bellman Statistics (2020, 7 citations), providing a principled, accountable framework for assessing new policies from historical data—critical for high-stakes domains like healthcare and autonomous systems. His Metric Residual Network (2023) further tackles sample efficiency in goal-conditioned reinforcement learning, enabling robots to learn complex navigation and manipulation tasks with fewer interactions. By bridging theoretical rigor with practical robotics benchmarks, Feng’s contributions are shaping how researchers build adaptive, safe, and data-efficient learning agents. His work is particularly influential for students and researchers interested in lifelong learning, transfer in robotics, and trustworthy policy evaluation.