Xuchuan Huang | 黄叙川

I am a junior at Yuanpei College of Peking University. I am advised by Prof. Yaodong Yang at PAIR Lab and PKU-Psibot Joint Lab. I am also a research assistant at NUS LinS Lab, advised by Prof. Lin Shao. My current research interests lie in embodied AI, especially dexterous manipulation, with the goal of building general-purpose embodied agents capable of completing diverse tasks in real world. I am always enthusiastic about exploring new ideas and opportunities. Feel free to reach out if you'd like to connect or collaborate! 🤩

E-mail  /  Github /  WeChat Account

We observe that current VLA models can be unified under a single framework: vision and language inputs are processed by a series of VLA modules, producing a chain of action tokens that progressively encode more grounded and actionable information, ultimately generating executable actions.

DexGraspVLA is a hierarchical framework for general-purpose dexterous grasping, which achieves over 90% zero-shot success in diverse real-world scenarios.

TerraX is a vision-language framework for terrain classification, featuring the TerraData dataset, TerraBench benchmark, and TerraCLIP model.

To empower progress alignment, ProgressGym is a benchmark for training AI to track, predict, and co-evolve with human moral progress.

We analyze over 130,000 Weibo comments on the 2023 Israel-Palestine conflict to examine how public opinion forms and spreads in Chinese social media. It reveals how factors like verified accounts, moral framing, and cultural similarity shape sentiment dynamics in Chinese online discourse.

We analyze 92,000+ multilingual tweets about Black Myth: Wukong to uncover how cultural, political, and emotional interpretations vary across linguistic communities. Using sentiment analysis and topic modeling, it highlights the complexities of global cultural exchange and digital geopolitics.