Portrait of Haozhuo Zhang

Robot learning

Haozhuo Zhang

张浩卓

PhD Student in Robot Learning

Department of Computer Science, The University of Manchester

Cognitive Robotics Lab (CoRoLab)

Supervisor: Prof. Angelo Cangelosi

haozhuo.zhang@postgrad.manchester.ac.uk

Google Scholar Email

Research

I work on robot learning for humanoid and cognitive robots. The goal is a general-purpose system for long-horizon, physically complex whole-body control, in which a humanoid interacts with objects and its environment. This covers whole-body control and loco-manipulation, physics-based motion synthesis for 3D humanoids, vision-language-action models for humanoids, world-action models, and human–object and human–scene interaction.

Publications

Independent first author

Diagram of Humanoid Horizon: parallel training, dynamic starting, and reward gating.

Humanoid Horizon: Extending Task Horizon in Whole-Body Loco-Manipulation via Parallel Training, Dynamic Starting, and Reward Gating

Advances in Neural Information Processing Systems (NeurIPS), 2026. Accepted.

arXivProject

Figure from LHM-Humanoid on long-horizon object transport.

LHM-Humanoid: Long-Horizon Human Motion Control for Continuous Object Transport in Cluttered Scenes

Asian Conference on Computer Vision (ACCV), 2026. Accepted.

arXivProject

Examples of hands generated by Hand1000.

Hand1000: Generating Realistic Hands from Text with Only 1,000 Images

AAAI Conference on Artificial Intelligence (AAAI), 2025. Accepted.

ProceedingsarXivProject

IEEE Robotics and Automation Society emblem.

Incremental Reinforcement Learning for Multi-Task Humanoid-Object-Interaction

IEEE Robotics and Automation Letters (RA-L). Under review.

WACV 2027 logo.

PCD-Humanoid: Point-Cloud Driven Whole-Body Humanoid Object Transport across 2,000 Warehouse Scenes

IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2027. Under review.

Second author

Figure from FLIP, a flow-centric manipulation world model.

FLIP: Flow-Centric Generative Planning as General-Purpose Manipulation World Model

International Conference on Learning Representations (ICLR), 2025. Accepted.

OpenReviewProject

Figure from the continual person re-identification paper.

Mitigate Catastrophic Remembering via Continual Knowledge Purification for Noisy Lifelong Person Re-Identification

ACM International Conference on Multimedia (ACM MM), 2024. Accepted.

DOI

Fourth author

Figure from DF3 on decoder-free feature forecasting.

DF3: World Modeling via Decoder-Free Feature Forecasting in Autonomous Navigation

arXiv:2608.02428, 2026. Preprint.

arXiv

Education

2025–present
The University of Manchester

The University of Manchester

PhD student in Computer Science. Cognitive Robotics Lab (CoRoLab). Supervisor: Prof. Angelo Cangelosi. Manchester, United Kingdom.

2021–2025
Peking University

Peking University

BSc in Computer Science. Beijing, China. Undergraduate research with Prof. Shanghang Zhang and Prof. Jiahuan Zhou (Peking University), Prof. Lin Shao (National University of Singapore), and Prof. Bin Zhu (Singapore Management University).

Industry experience

2025–2026
X-Humanoid

X-Humanoid, Beijing Humanoid Robot Innovation Center

Research intern, on-site and remote. Beijing, China. Long-horizon whole-body loco-manipulation, including reinforcement learning teachers distilled into vision-language-action policies for continuous object transport in cluttered scenes.

Awards

2025

Dean’s Doctoral Scholarship

Faculty of Science and Engineering, The University of Manchester. Full scholarship. About 10 to 20 awards each year across the whole Faculty, after department nomination and a faculty interview.

2025

Outstanding Undergraduate Thesis

Peking University. University outstanding thesis, and one of the ten best theses in the School of EECS.