William Liang

Papers

Total Citations

H-Index

About

William Liang is a leading researcher at the intersection of large language models (LLMs) and robotic manipulation. His work focuses on using LLMs not just as planners, but as powerful tools for reward design and sim-to-real transfer—bridging the gap between high-level reasoning and low-level physical control. Liang’s major contributions include pioneering the use of coding LLMs to automate reward function engineering, a traditionally labor-intensive bottleneck in reinforcement learning. His first-author paper, *Eureka: Human-Level Reward Design via Coding Large Language Models* (48 citations), demonstrated that LLMs could autonomously generate reward functions enabling complex tasks like dexterous pen spinning, achieving human-level performance. He further advanced the field with *LIV: Language-Image Representations and Rewards for Robotic Control* (24 citations), which unified vision-language representation learning with reward learning from action-free video. Most recently, his work *DrEureka: Language Model Guided Sim-To-Real Transfer* (7 citations) tackles the critical challenge of transferring simulated policies to real-world robots without manual tuning. Liang’s research is defining a new paradigm where language models serve as both the “brain” and the “trainer” for robotic systems, making him a rising star in embodied AI.