Siyu Xu is a Ph.D. student at The University of Sydney, supervised by A/Prof. Chang Xu. His research focuses on Embodied AI, specifically Vision-Language-Action (VLA) models and World-Action (WAM) models. His goal is to develop efficient and general-purpose robotic systems capable of understanding and interacting with the world through vision, language, and actions, enabling them to seamlessly assist humans in various real-world tasks and everyday life.

Recently, he has been particularly interested in improving the out-of-distribution (OOD) generalization and efficiency of VLA/WAM systems, with the goal of making embodied foundation models more reliable and practical for real-world deployment.

🔥 News

  • 2026.06:  🎉🎉 One paper is accepted to ECCV 2026.
  • 2026.05:  🎉🎉 One paper is accepted to ICML 2026.
  • 2026.02:  🎉🎉 One paper is accepted to CVPR 2026.
  • 2026.01:  🎉🎉 One paper is accepted to ICLR 2026.
  • 2025.09:  🎉🎉 VLA-Cache is accepted to NeurIPS 2025.
  • 2025.01:  🎉🎉 One paper is accepted to NAACL 2025 Findings.

📝 Publications

ECCV 2026
sym

Revisiting Parameter Redundancy in Vision-Language-Action Models: Insights from VLM-to-VLA Adaptation

Fengnian Zhang, Tao Huang, Siyu Xu, Zhong Jin, Chang Xu

[Code]

European Conference on Computer Vision (ECCV), 2026

ICML 2026
sym

Seeing Realism from Simulation: Efficient Video Transfer for Vision-Language-Action Data Augmentation

Chenyu Hui, Xiaodi Huang, Siyu Xu, Yunke Wang, Shan You, Fei Wang, Tao Huang, Chang Xu

[Code] [Poster]

International Conference on Machine Learning (ICML), 2026

CVPR 2026
sym

Affordance Field Intervention: Enabling VLAs to Escape Memory Traps in Robotic Manipulation

Siyu Xu, Zijian Wang, Yunke Wang, Chenghao Xia, Tao Huang, Chang Xu

[Poster]

Conference of the Computer Vision and Pattern Recognition (CVPR), 2026

ICLR 2026
sym

Action-aware Dynamic Pruning for Efficient Vision-Language-Action Manipulation

Xiaohuan Pei, Yuxing Chen, Siyu Xu, Yunke Wang, Yuheng Shi, Chang Xu

[Project] [Code]

International Conference on Learning Representations (ICLR), 2026

NeurIPS 2025
sym

VLA-Cache: Towards Efficient Vision-Language-Action Model via Adaptive Token Caching in Robotic Manipulation

Siyu Xu, Yunke Wang, Chenghao Xia, Dihao Zhu, Tao Huang, Chang Xu

[Project] [Code] [Nvidia GTC Poster]

Conference of the Thirty-Ninth Annual Conference on Neural Information Processing Systems (NeurIPS), 2025

NAACL 2025
sym

CollagePrompt: A Benchmark for Budget-Friendly Visual Recognition with GPT-4V

Siyu Xu, Yunke Wang, Daochang Liu, Bo Du, Chang Xu

[Project] [Code]

Conference of the Nations of the Americas Chapter of the ACL (NAACL), 2025

🎖 Honors and Awards

  • 2026.03 Poster Presenter - NVIDIA GTC 2026 Conference San Jose
  • 2025.11 Invention Patent - China National Intellectual Property Administration
  • 2025.10 NeurIPS 2025 Scholar Award - Neural Information Processing Systems Foundation Board
  • 2022.05 Second Class Prize - ASC22 Student Supercomputer Challenge, Asia Supercomputer Community
  • 2020.10 First Place Award - Pre-training for Video Captioning Challenge, ACM International Conference on Multimedia
  • 2020.10 7th Place - Artificial Intelligence Competition for Video Generation Challenge, ZHEJIANG LAB
  • 2020.05 9th Place - International Audio and Video Algorithm Optimization Competition, Mongo Media

🧑‍🏫 Teaching

  • 2024 S1, Tutor of COMP5329, Deep Learning, USYD
  • 2023 S2, Tutor of COMP5328, Advanced Machine Learning, USYD

📖 Educations

  • 2025.07 - Present, Ph.D in Intelligent Robotics, University of Sydney
  • 2023.07 - 2025.06, M.Phil in Computer Vision, University of Sydney
  • 2022.08 - 2023.06, Master of Information Technology, University of Sydney

💻 Internships