Hongliang Zeng (Xavier)

AI & Embodied Intelligence Researcher

Download CV
Hongliang Zeng

I hold a Ph.D. from South China University of Technology (advised by Prof. Ping Zhang), with research spanning embodied intelligence, robotic manipulation, active perception, reinforcement learning, and 3D point cloud understanding. I have published 7 first-author papers at IJCAI, AAAI, TNNLS, ICASSP, and ICME. I now work at sudo, focusing on robot foundation-model development.

Education

Sep 2020 — Jun 2025
  • Research on embodied intelligence, including robotic manipulation, active perception, and reinforcement learning, published at IJCAI, AAAI, TNNLS, etc.
  • Research on 3D point cloud self-supervised learning and generation, published at ICASSP, ICME, etc.
  • Granted patent: A Method and System for Robotic Manipulation of Articulated Objects.
Robotic Manipulation Computer Vision
Sep 2014 — Jun 2018

Undergraduate studies in mechanical engineering.

SolidWorks AutoCAD

Work Experience

Jul 2026 — Present

Foundation model algorithm development.

Nov 2024 — May 2026

Embodied Intelligence Algorithm Engineer

Astribot, Shenzhen

  • Model: Led the development of DuoCore-FS, a fast-slow dual-system VLA for whole-body manipulation, reaching 30 Hz action generation (4B, RTX 3090); published a technical report; optimized π0.6* value-model training for stability and less overfitting, and attempted a real-robot RL pipeline; validated VLA + grounding-loss co-training (+40% unseen-object generalization).
  • Demo: Contributed to the Astribot 0426 VLA system demo and the ICRA 2026 cloth-folding demo, responsible for data collection and model training.
  • Data: Collected and organized 4,000+ hours of heterogeneous robot data for pretraining, and co-developed a DAgger-based interactive data-collection pipeline accelerating policy-driven data gathering.
  • Baseline: Reproduced and validated baseline models and related techniques (π0-small, π0-fast, π0, π0.5, π0.6*, RTC, fast-wam, lingbot-va, Qwen3VL + expert, etc.).
  • AI Infra: Built the team's large robot-model distributed training framework and unified data-format standard, improving training and iteration efficiency.

Honors

  • 5-Star Performance (Top Rating)
  • Outstanding Contribution Partner
  • Astribot Navigator
VLA RL World Model
Jul 2018 — Jun 2019

Mechanical design and engineering.

SolidWorks AutoCAD Mechanical Design

Publications