Papers
16
Total Citations
436
H-Index
10
About
Siyuan Huang is a prominent researcher at the intersection of embodied AI, robotics, and 3D scene understanding, whose work is reshaping how intelligent systems perceive, reason about, and interact with the physical world. His most celebrated contribution, SceneDiffuser (2023, 165 citations), introduced a unified diffusion-based generative model for 3D scene-conditioned generation, optimization, and planning — a landmark advance that established a new paradigm for physics-aware, goal-oriented scene understanding. Beyond scene modeling, Huang has made substantial contributions to dexterous robotic manipulation, notably through GenDexGrasp (2023, 59 citations), which tackled the long-standing challenge of generalizable multi-hand grasping, and work on simultaneous multi-object grasping that mirrors human hand dexterity. His research also bridges foundation models and embodied intelligence, with projects like Instruct2Act and COME-robot demonstrating how large language and vision-language models can be grounded in robotic action. His YouRefIt dataset advanced embodied reference understanding through multimodal language and gesture cues. Across more than ten highly cited publications spanning robotics, 3D vision, and multimodal AI, Huang has established himself as a leading voice in building truly generalist, physically grounded robotic agents.
Research Focus
Key Achievements
Top Papers
- 1Diffusion-based Generation, Optimization, and Planning in 3D Scenes165 citations · 2023
- 2GenDexGrasp: Generalizable Dexterous Grasping59 citations · 2023
- 3
- 4YouRefIt: Embodied Reference Understanding with Language and Gesture32 citations · 2021
- 5
- 6Grasp Multiple Objects With One Hand26 citations · 2024
- 7Overview of image-based 3D reconstruction technology22 citations · 2024
- 8
- 9An Embodied Generalist Agent in 3D World15 citations · 2023
- 10Closed-Loop Open-Vocabulary Mobile Manipulation with GPT-4V10 citations · 2025