Personal Profile

01

Education Experience

Since Feb 2025, I have worked with the IWIN-FINS Lab at Shanghai Jiao Tong University on undergraduate research in underwater robot learning and control.

Since Sep 2025, I have also worked with the RHOS Lab on embodied-AI evaluation, VLA pretraining, and robot learning.

02

Research

Robotics, embodied AI and VLA.

03

Publications

  1. Yu Zhang, Yuanmingqing Song, et al. “FinsSim: A reality-aligned integrated simulation platform for underwater robot learning.” Manuscript under review at IEEE International Conference on Robotics and Automation (ICRA). [First author]
  2. Mingwei Li, Dehao Ye, Yu Zhang, Ziyu Wang, Yong-Lu Li. “Data That Selects Itself: Scalable Self-Curation for Robot Learning.” Manuscript under review at International Conference on Learning Representations (ICLR 2027), 2026. [Third author]
  3. Ziyu Wang, Chenyuan Liu, Yushun Xiang, Runhao Zhang, Yu Zhang, et al. “The Great March 100: 100 Detail-oriented Tasks for Evaluating Embodied AI Agents.” arXiv:2601.11421, 2026. [Fifth author; manuscript under review]

Research Directions

This is a collection of ongoing and long-term researches.

Embodied AI is a field that studies how intelligent agents interact with the real world. It involves multiple aspects such as perception, decision-making, and action, aiming to build intelligent systems that can operate autonomously in complex environments.

“Artificial intelligence makes us think what is artificial and what is intelligence.”

Embodied AI and robotics

VLA model pretraining with human-hand keypoints

Actual deployment on edge robotics platforms

Locomotion of humanoid robots

Open this research
Embodied AI, focusing on how intelligent agents interact with the real world.

Focus on how to apply Learning-based methods to the motion control problem of underwater robots, including simulation platform construction, real machine deployment and inference verification. Including single robot and multi-robot. The robot is the carrier of research, and the underwater is a special scene for robot applications. Through the research of underwater robots, accumulate basic engineering experience and general principles of robot system development, and explore cutting-edge neural network methods in combination with reinforcement learning.

“Chase the sea, and become the sea.”

Reinforcement learning for underwater robots

Control Theory for Underwater Systems

Simulation-to-Hardware Framework

Multiagent Systems

Open this research
Underwater Reinforcement Learning, for the motion control of vehicles in complex aquatic environments.

Articles and notes

A lighter writing for research notes and everyday thinking beside the more formal Research and Publications indexes.