Xirui Kang · Miracle Kang

Xirui Kang

Research in deep learning and vision-language-action models for embodied AI.

AffiliationCity University of Hong Kong

Research DirectionsRepresentation Learning · Transfer Learning · Embodied Intelligence

Portrait of Xirui Kang
Portrait Xirui Kang · Miracle Kang

About

I build models that connect perception, language, and action.

I am Xirui "Miracle" Kang. As an Embodied AI researcher, my work aims to give artificial intelligence the ability to understand and interact with the physical world. But beyond the algorithms, my true drive is deeply personal.

To me, life is an endless quest for understanding. I believe that through the quiet persistence of lifelong learning, we can gradually transcend the endless complexities of "how" (Shu) to grasp the profound clarity of "why" (Dao). Yet, the summit of this intellectual journey is never solitary wisdom. Ultimately, knowledge is a vessel meant to be emptied—shared freely to illuminate the path for others, translating our collective progress into genuine, enduring happiness.

Experience

Units and schools.

2025-12 ~ 2026.07

X Square Robot

Foundation model pretraining and action tokenizer design.

2025.09 ~ 2026.10

City University of Hong Kong

Research and study in deep learning, VLM models, and VLA.

2025 - Present

Tsinghua University (RA)

Research collaboration in embodied intelligence and robot foundation models.

2020.09 ~ 2024.06

Nanjing University of Posts and Telecommunications

Undergraduate studies and early research training in AI and related fields.

Publications

Publications by year.

2026

Wall-OSS-0.5 Technical Report

R Yu, P Zhang, S Liu, B Liu, M Kang, S Li, L Shi, E Ma, P Yang, C Pan, et al. X Square Robot Technical Report.

Report

2026

X-Tokenizer: A Multimodal Action Tokenizer for Vision-Language-Action Pretraining

X Kang, Y Shi, L Liang, R Gan, D Liu, P Zhang, D Chen, X Qin, Y Zheng, et al. arXiv preprint arXiv:2606.14752.

arXiv Project

2026

SpecPL: Disentangling Spectral Granularity for Prompt Learning

J Zhou, X Kang, F Huang, L. M. Po.

Paper

2025

X-VLA: Soft-prompted Transformer as Scalable Cross-Embodiment Vision-Language-Action Model

J Zheng, J Li, Z Wang, D Liu, X Kang, Y Feng, Y Zheng, J Zou, Y Chen, et al. 1st at Agibot World Challenge, IROS 2025.

Paper / Project

2024

LiteSOD-DINO: Leveraging Large Kernel Convolutions and Attention for Enhanced Small Object Detection in Infrared Imagery Under Resource Constraints

X Kang. International Conference on Neural Information Processing, 305-319.

Paper