Mengfei Zhao
Research map
Portrait of Mengfei Zhao

Mengfei Zhao

CTO at Axis Robotics

Research Assistant at NTU
Undergraduate at SUSTech

Building self-improving robotic systems that decide what to learn next.

My research lies at the intersection of data-centric learning, self-improving systems, and sim-to-real reinforcement learning. I study how robots can decide what experience to acquire, what data to retain, and what to learn next.

I am building robot agents and skill harnesses that connect simulation-trained experts, visual skills, and real-world feedback. These systems support skill selection, composition, and execution evaluation, enabling robots to continually refine their skills and the experience they learn from.

At Axis Robotics, backed by a $12 million seed round, I lead a team of approximately 15 researchers and engineers. As CTO, I set our research direction and oversee team organization, departmental budgeting, and resource allocation.

I am mentored by Jianfei Yang and Jindou Jia at NTU, and collaborate with Jiachen Li at Georgia Tech.

Research Map

Data-Mediated Self-ImprovingTUCOProbing Data Value
Axis Dataset V1
Co-Adaptive DAggerEvolving Skills
Co-Adaptive DAggerTUCO
Evolving Skills
AURAAPEXΔ Flow MatchingCODE
General
Model

Recent Publication

  1. TUCO — overview of the research method

    TUCO: Curating Simulation Demonstrations for Sim-to-Real Robot Policy Co-Training

    Ning Zhu*, Mengfei Zhao*, Yikai Tang*, Zhangyujie Sun, Peihao Li, Dongyue Ni, Jindou Jia†, Jianfei Yang†

    Co-first author, listed second

    * Co-first authors · Equal contribution · † Corresponding author

    In submission · ICLR

  2. AXIS — overview of the research method

    AXIS: A Growable Community-Driven Data Engine for Scalable Robot Manipulation

    Mengfei Zhao*†, Dihong Huang*, Yikai Tang*, Peihao Li*, Mingxuan Yan*, Ruiqi Zhuang*, Yanjia Huang*, Jie Wang, Hai Zhai, Tony Zhou, Rui Zhang, Zhexi Luo, Yuchen Huang, Jianfei Yang†, Jiachen Li†

    Co-first author, listed first · Corresponding author

    * Co-first authors · Equal contribution · † Corresponding author

    IROS Workshop

  3. APEX — overview of the research method

    APEX: Adaptive Policy Execution for Precise Manipulation

    Mengfei Zhao*, Chenxi Jiang*, Tuo An, Jindou Jia†, Jianfei Yang†

    Co-first author, listed first

    * Co-first authors · Equal contribution · † Corresponding author

    In submission · ICRA

  4. AURA: Adaptive Uncertainty Enables Emergent Robot Behaviors Without Scaling

    Co-first author, listed fourth

    Targeted: Science Robotics

    Use action history and local action diversity to adapt uncertainty in generative policies, adjusting history, inference, and execution schedules.

  5. Δ Flow Matching — overview of the research method

    Δ Flow Matching for Robot Foundation Models

    Kuangji Zuo*, Tuo An*, Xinying Guo, Gen Li, Mengfei Zhao, Jiaqi Bai, Bofan Lyu, Zhuoyuan Yu, Jinghan Yang, Jiayi Li, Jindou Jia†, Jianfei Yang†

    Fifth author; no co-first designation in the current manuscript · Project leader

    * Co-first authors · Equal contribution · † Corresponding author

    Start generation from recent action and visual history with light noise, learning changes from past to future to study training efficiency, single-step inference, and action smoothness.

  6. AXIS-Bench: simulation demonstrations, state-based quality labels, and quality-aware VLA pretraining findings

    What Matters in Quality-Aware VLA Pretraining: A Benchmark and Empirical Study

    Mingxuan Yan, Litian Gong, Dihong Huang, Zhixu Li, Tianyu Zhang, Wenqian Zhang, Hefeifei Jiang, Peihao Li, Jian Zhang, Mengfei Zhao, Yikai Tang, Hai Zhai, Zehao Wang, Ruijian Liang, Yanjia Huang, Lin Shao, Changliu Liu, Kaiyu Hang, Zhiwen Fan, Masayoshi Tomizuka, Jianfei Yang, Jiachen Li†

    Tenth author

    · † Corresponding author

    In submission · ICRA