Haodong Cui | 崔浩东

Biography

I am a master's student at the School of Artificial Intelligence and Automation, Huazhong University of Science and Technology (HUST), advised by Prof. Zhigang Zeng (IEEE Fellow).

My research interests include Embodied Intelligence, World Models, Multimodal Large Language Models, and Video Generation, with a focus on how learned representations support planning, control, and generalization in robots.

I am currently a World Models Research Intern at Xiaomi, exploring in-context learning for World-Action Models. Previously, I worked on world-model reinforcement learning for real robots at Department of Precision Instrument, Tsinghua University. I received my B.S. in Mathematics and Applied Mathematics from Xinjiang University, ranking 1/289 with a GPA of 4.28/5.0.

Always happy to discuss research and collaboration.

News

[2026/09/27] 🎉 Our paper HaM-World: Soft-Hamiltonian World Models with Selective Memory for Planning has been accepted by NeurIPS 2026.
[2026/01/07] 🎉 Our paper Spatial Sampling-based Passivity and Synchronization of Multiweighted Coupled Reaction-diffusion Neural Networks was accepted by CNSNS.
[2025/09/01] 🎉 Our paper Adaptive Passivity-based Synchronization of Spatiotemporal Neural Networks with Multi-weighted Coupling under Spatially Point Measurements was accepted by CISC 2025.

Selected Publications

Selected work in world models and mathematical control.

NeurIPS 2026

HaM-World: Soft-Hamiltonian World Models with Selective Memory for Planning Haoyun Tang*, Haodong Cui*, Keyao Xu, Zhandong Mei†, Kun Wang† Conference on Neural Information Processing Systems (NeurIPS), 2026

PDF arXiv alphaXiv17 Code14

@article{tang2026hamworld,
  title={HaM-World: Soft-Hamiltonian World Models with Selective Memory for Planning},
  author={Tang, Haoyun and Cui, Haodong and Xu, Keyao and Mei, Zhandong and Wang, Kun},
  journal={arXiv preprint arXiv:2605.05951},
  year={2026}
}

CNSNS 2026

Spatial Sampling-based Passivity and Synchronization of Multiweighted Coupled Reaction-diffusion Neural Networks Haodong Cui, Haoyun Tang, Mingyu Ma, Cheng Hu†, Tingting Shi Communications in Nonlinear Science and Numerical Simulation, 2026

PDF DOI

@article{cui2026spatial,
  title={Spatial Sampling-based Passivity and Synchronization of Multiweighted Coupled Reaction-diffusion Neural Networks},
  author={Cui, Haodong and Tang, Haoyun and Ma, Mingyu and Hu, Cheng and Shi, Tingting},
  journal={Communications in Nonlinear Science and Numerical Simulation},
  volume={157},
  pages={109666},
  year={2026},
  doi={10.1016/j.cnsns.2026.109666}
}

CISC 2025

Adaptive Passivity-based Synchronization of Spatiotemporal Neural Networks with Multi-weighted Coupling under Spatially Point Measurements Haoyun Tang, Haodong Cui, Mingyu Ma, Cheng Hu†, Tingting Shi Chinese Intelligent Systems Conference (CISC 2025), 2026

PDF DOI

@inproceedings{tang2026adaptive,
  title={Adaptive Passivity-based Synchronization of Spatiotemporal Neural Networks with Multi-weighted Coupling under Spatially Point Measurements},
  author={Tang, Haoyun and Cui, Haodong and Ma, Mingyu and Hu, Cheng and Shi, Tingting},
  booktitle={Proceedings of 2025 Chinese Intelligent Systems Conference},
  pages={266--276},
  year={2026},
  publisher={Springer Nature Singapore},
  doi={10.1007/978-981-95-6813-0_26}
}

Research in progress

Beyond Final Answers: Diagnosing Deep Research Agents through Observable Trajectories

Co-first author · Under review at EACL 2027

ReliDRA diagnoses process failures through task alignment, transition quality, and evidence in observable agent trajectories.

Education

Huazhong University of Science and Technology (HUST) 2026.09 – Present

Master's Student School of Artificial Intelligence and Automation (AIA)

M.S. in Control Science. Advisor: Prof. Zhigang Zeng (IEEE Fellow).
Xinjiang University (XJU) 2022.09 – 2026.06

Undergraduate Student College of Mathematics and System Science

B.S. in Mathematics and Applied Mathematics. GPA: 4.28/5.0 · Rank: 1/289.

Research & Industry Experience

Xiaomi2026.09 – Present

World Models Research Intern

Researching how in-context learning improves generalization to unseen tasks in World-Action Models (WAMs).
Tsinghua University, Center for Brain Inspired Computing Research 2025.11 – 2026.05

Reinforcement Learning Algorithm Intern Advisors: Shangqi Guo and Fangwen Yu
- Worked on DreamerV3-based reinforcement learning systems for real-world robot navigation and locomotion on Clearpath Jackal UGV and Unitree Go1 platforms.
- Built real-data-to-policy workflows across keyboard data collection, offline training on curated trajectories, and pure inference deployment; selected projects: [Jackal Navigation], [Go1 DreamerV3-WTW].
- Designed reward, perception, and deployment modules for target progress, ArUco-based distance estimation, obstacle avoidance, odometry-depth consistency, and high-level gait/style adaptation.

Selected Projects

Jackal-DreamerV3 Navigation [GitHub] Keywords: DreamerV3, ROS2, JAX, RealSense, ArUco, ZeroMQ Built a real-robot multi-target indoor navigation workflow on Clearpath Jackal, covering keyboard data collection, offline training, reward/perception design, and pure inference deployment.
Go1-DreamerV3-WTW [GitHub] Keywords: DreamerV3, Walk These Ways, Unitree Go1, Isaac Gym, LCM Integrated DreamerV3 as a high-level gait/style adaptation policy over the Walk These Ways low-level controller, with training/deployment scripts and real-robot bring-up utilities.

Honors and Awards

National Scholarship (国家奖学金), 2025
Merit Student Pacemaker (三好学生标兵), 2025
Meritorious Winner, American Mathematical Contest in Modeling (美国大学生数学建模竞赛 M 奖), 2025
First Prize (Provincial), Chinese Mathematical Contest (全国大学生数学竞赛省一等奖), 2024
Academic Excellence Scholarship, Xinjiang Uygur Autonomous Region (学业优秀奖学金), 2024