2022 - Present
Ph.D. Candidate, Artificial Intelligence
The Hong Kong University of Science and Technology (Guangzhou)
He Zhang 张贺 · Zhang He · Zhanghe
Ph.D.Ph.D. Candidate in Artificial Intelligence, HKUST(GZ)
EducationB.E. in Automation, Tsinghua University
Hi! I am a Ph.D. Candidate in Artificial Intelligence at the Hong Kong University of Science and Technology (Guangzhou), advised by Prof. Hui Xiong.
I received my B.E. in Automation from Tsinghua University in 2022. My research combines reinforcement learning, multimodal behavior learning, robot control, and embodied intelligence.
Education
Academic training in artificial intelligence and automation.
2018 - 2022
B.E., Automation
Tsinghua University
Research
Learning agents that acquire knowledge and skills through physical interaction.
I aim to build embodied agents that explore autonomously, discover reusable physical and causal structure, and improve continuously through interaction.
Autonomous Exploration & Skill Discovery
Reusable skills and informative interactions without dense supervision.
Intervenable Multimodal Behavior
Efficient policies whose behavior can be selected, adapted, and improved.
World Models & Lifelong Robot Learning
Action-conditioned world models that accumulate knowledge without forgetting.
Current research plan
I am exploring embodied scientific discovery: robots autonomously conduct physical experiments, recover relationships among force, mass, acceleration, friction, and contact, then use them for prediction and control.
Interests: Reinforcement Learning · Robot Learning · Multimodal Imitation · World Models · Continual Learning · Embodied Intelligence
Publications
Selected first-author and collaborative research. My name is shown in bold.
-
Source-Lifted Flow Matching for Intervenable Multimodal Imitation
Source-intervenable flow policies for controllable multimodal robot behavior.
-
Spatial Memory for Out-of-Vision Manipulation in Vision-Language-Action
SOMA equips VLA policies with persistent spatial-semantic memory for manipulating targets beyond the current camera view.
-
GoldenStart: Q-Guided Priors and Entropy Control for Distilling Flow Policies
Q-guided priors and explicit entropy control for efficient continuous-control policies.
-
A Brain-Inspired Embodied Intelligence for Fluid and Fast Reflexive Robotics Control
NeuroVLA organizes semantic planning, adaptive motor control, and fast reflexes into a cortex-cerebellum-spinal hierarchy.
-
Efficient Skill Discovery via Regret-Aware Optimization
A regret-aware min-max framework for improving skill-learning efficiency and diversity.
-
Interactive Interior Design Recommendation via Coarse-to-fine Multimodal Reinforcement Learning
Multi-turn recommendation that learns latent user preferences through reinforcement learning.
Open source
Selected open-source work in embodied intelligence.
AlphaBrain Platform
Core contributor to an open-source embodied-intelligence framework covering VLA architectures, world models, continual learning, reinforcement learning, shared training, and evaluation infrastructure.
- Developed RLT/RLT_a and PPO/GRPO trainers.
- Integrated Qwen and π₀.₅ for VLA fine-tuning and online RL.
- Refactored LIBERO evaluation and distributed execution.
43
public commits · first by commit count
View on GitHub
Experience
Research and engineering experience, ordered from most recent.
-
2025 — 2026
Algorithm Engineer
AI² Robotics · Shenzhen
Vision-language-action models, reinforcement learning, humanoid deployment, and open-source research.
-
2024 — 2025
Algorithm Intern
Shanghai AI Laboratory · Shanghai
Continual learning, robotic VLA deployment, and unsupervised skill discovery.
-
2022 — 2023
Algorithm Intern
OPPO · Shenzhen
Multimodal interactive recommendation with reinforcement learning.
-
2022 — 2023
Remote Research Intern
Microsoft Research Asia · Remote
Machine learning, decentralized finance security, and blockchain applications.
-
2021
Algorithm Intern
Xiaomi · Beijing
Deep click-through-rate prediction and advertising algorithms.
Awards
Selected recognition.
- CAST Young Science and Technology Talent Development Program Doctoral Student Track · 2025
- Comprehensive Excellence Scholarship Tsinghua University · 2020