Mengyuan Yan
Papers
13
Total Citations
1,037
H-Index
10
About
Mengyuan Yan is a robotics and machine learning researcher whose work spans robot manipulation, 3D scene understanding, and the integration of large language models with physical systems. She is perhaps best known for her contribution to "Do As I Can, Not As I Say" (2022), a landmark paper with over 500 citations that demonstrated how large language models can be grounded in real-world robotic affordances to execute complex, natural-language instructions — a foundational advance in embodied AI. Her MeteorNet architecture (2019, 225 citations) pioneered deep learning on dynamic 3D point cloud sequences, providing robots with richer tools for perceiving and reasoning about moving environments. Yan has also made significant contributions to robot manipulation of deformable objects, developing self-supervised state estimation techniques that enable reliable handling of flexible materials like ropes and cables. Her research on sim-to-real transfer, reinforcement learning at scale for real-world deployment, and knot-planning using topological motion primitives further reflects her commitment to bridging theoretical machine learning with practical robotics. Collectively, her work addresses some of the most challenging open problems in building robots that can perceive, reason, and act effectively in unstructured real-world environments.
Research Focus
Key Achievements
Top Papers
- 1Do As I Can, Not As I Say: Grounding Language in Robotic Affordances516 citations · 2022
- 2MeteorNet: Deep Learning on Dynamic 3D Point Cloud Sequences225 citations · 2019
- 3
- 4
- 5Jump-Start Reinforcement Learning17 citations · 2022
- 6MeteorNet: Deep Learning on Dynamic 3D Point Cloud Sequences16 citations · 2019
- 7
- 8
- 9Learning Topological Motion Primitives for Knot Planning11 citations · 2020
- 10