Wenli Xiao

Wenli Xiao

Robotics PhD Student at CMU School of Computer Science

I'm a PhD student at CMU Robotics, advised by Prof. Guanya Shi. My vision is to build scalable general intelligent robots in the real world. I am generally interested in dexterous manipulation and humanoid robots.

I am currently doing a research internship at NVIDIA GEAR Lab, working with Dr. Jim Fan and Prof. Yuke Zhu, where I am leveraging industrial-level resources for scaling up robot foundation model learning.

News

  • 2025 Oct
    Received CMU RI Presidential Fellowship!

Research Projects

Self-Improving Vision-Language-Action Models with Data Generation via Residual RL
Co-Lead

Authors

Wenli Xiao*, Haotian Lin*, Andy Peng, Haoru Xue, Tairan He, Yuqi Xie, Fengyuan Hu, Jimmy Wu, Zhengyi Luo, Linxi "Jim" Fan†, Guanya Shi, Yuke Zhu†

Self-Improving Vision-Language-Action Models with Data Generation via Residual RL

Under Review
TL;DR: PLD (Probe, Learn, Distill) is a plug-and-play recipe for Vision-Language-Action (VLA) post-training. It is model agnostic, supporting both autoregressive and diffusion architectures, and can push success rates to 99%.
HOVER: Versatile Neural Whole-Body Controller for Humanoid Robots
Co-Lead

Authors

Tairan He*, Wenli Xiao*, Toru Lin, Zhengyi Luo, Zhengjia Xu, Zhenyu Jiang, Jan Kautz, Changliu Liu, Guanya Shi, Xiaolong Wang, Linxi 'Jim' Fan†, Yuke Zhu†

HOVER: Versatile Neural Whole-Body Controller for Humanoid Robots

ICRA 2025
TL;DR: HOVER is a 1.5M-parameter neural network to control the body of a humanoid robot. It takes a lot of subconscious processing for us humans to walk, maintain balance, and maneuver our arms and legs into desired positions. We capture this 'subconsciousness' in HOVER, a single model that learns how to coordinate the motors of a humanoid robot to support locomotion and manipulation.
AnyCar to Anywhere: Learning Universal Dynamics Model
Co-Lead

Authors

Wenli Xiao*, Haoru Xue*, Tony Tao, Dvij Kalaria, John Dolan, Guanya Shi

AnyCar to Anywhere: Learning Universal Dynamics Model

CoRL 2024, X-Embodiment workshop
ICRA 2025
TL;DR: AnyCar is a generalist vehicle dynamics model for agile mobility. It can adapt to various cars, tasks, and envs via in-context adaptation, outperforming well-tuned generalist models up to 54%.

Background

NVIDIA GEAR Lab logo

Research Intern

NVIDIA GEAR Lab

2024 - Present

Advisor:

Dr. Jim Fan

Prof. Yuke Zhu

Carnegie Mellon Univ Robotics Institute Summer Scholar, RISS logo

Research Intern

Carnegie Mellon Univ Robotics Institute Summer Scholar, RISS

June 2022 - Aug 2023

Advisor:

Prof. John Dolan

Yiwei Lyu

RISE Lab, UC Berkeley logo

Research Intern

RISE Lab, UC Berkeley

March 2022 - May 2022

Advisor:

Prof. Joseph E. Gonzalez

Tianjun Zhang

NCEL Lab, Shenzhen AIRS logo

Research Intern

NCEL Lab, Shenzhen AIRS

Aug 2020 - March 2022

Advisor:

Prof. Jianwei Huang

Prof. Bing Luo