Hi there 
Welcome to my academic base!
I am a junior at Xidian University, currently pursuing research in robot learning with particular attention to perception and inference about the real world under the guidance of Prof. Lixin Yang and Prof. Cewu Lu at the MVIG Lab, Shanghai Jiao Tong University. My overarching objective is to develop autonomous robots capable of independent exploration and self-directed learning through dynamic interaction with real world.
Previously, during my sophomore year, I contributed to research on adversarial attacks against computer vision systems at the Key Laboratory of Cooperative Intelligent Systems, Ministry of Education, under the supervision of Prof. Hao Li and Prof. Maoguo Gong. This experience instilled in me a deep appreciation for the importance of robustness and security in vision-based systems, particularly for applications in agents’ autonomous exploration.
News
- My first work on robot learning:MBA, about object motion for robots manipulation
- In charge of Microsoft Club. Feel free to reach out if you’d like to join.
- I have set up a Personal Blog, welcome everyone to visit!
Research Experience
Publications
Adversarial Attack
Yue Su, Hao Li✉, Maoguo Gong✉
A generative adversarial attack on VI-ReID models perturbs modality-invariant features, creating patches that expose SOTA vulnerabilities and highlight the need for enhanced feature extraction.
[arxiv]
Hao Li✉, Fanggao Wan, Yue Su, Yue Wu, Mingyang Zhang, Maoguo Gong✉
Historically, infrared adversarial attacks were single-use and tough to deploy. Using TEC, we implemented efficient attacks adaptable to hardware scenarios, causing pedestrian detection models in infrared settings to misjudge. In Submission to AAAI2025. 🚀
Robot Learning
Yue Su*, Xinyu Zhan*, Hongjie Fang, Yong-Lu Li, Cewu Lu, Lixin Yang ✉
Propose MBA, a novel module that employs two cascaded diffusion processes for object motion generation and robot action generation under object motion guidance. Designed as a plug-and-play component, MBA can be flexibly integrated into existing robotic manipulation policies with diffusion action heads.
[arxiv] [website] [code]
Projects
Time series forecasting is a regression problem suited for U-Net's architecture due to its consistent input-output distributions and strong mathematical alignment. U-Net1D outperformed transformer-based models at 2022 on some tasks, showing its surprising potential. Combining U-Net with Bert-Encoder improved performance by incorporating both local and global attention.
[code] [report-cn]
The problem of finding the maximum flow in a network has been well solved in the Ford-Fulkerson Method. However, the problem lies in how to design better heuristic information to help us find the augmenting path in each iteration to minimize the number of iterations. We boldly challenge this problem through the ant colony algorithm.
[code] [report-cn]
Existing adversarial attacks on point clouds are mainly based on physical generation rather than gradient perturbation. We tried to extend FGSM to the 3D field and achieved significant success within a certain gradient range, but the sampling method of 3D models tells us that things seem to be not that simple...
[code] [report-cn]
This project uses the Google Gemini API to create a simple chatbot application simulating two crosstalk performers (Dougen and Penggen) performing based on user-provided topics.
[code] [report-cn]