Shortcut-enhanced Multimodal Backdoor Attack in Vision-guided Robot Grasping
Published in IEEE Transactions on Automation Science and Engineering, 2025
Overview
This work introduces SEMBA (Shortcut-Enhanced Multimodal Backdoor Attack), a novel backdoor attack method designed for multimodal vision-guided robot grasping systems in high-clutter scenarios. The method leverages Multimodal Shortcut Searching Algorithm and Multimodal Trigger Generation to effectively compromise robotic grasping systems.
Project Website
For more details, visualizations, and additional resources about this project, please visit the SEMBA Project Page.
Publication Details
Authors: Cheng Li, Ziyan Gao, Nak Young Chong
Venue: IEEE Transactions on Automation Science and Engineering
Publication Date: 2025-07-25
DOI: Read on TechRxiv
Recommended citation: Li, C., Gao, Z., & Chong, N. Y. (2025). Shortcut-enhanced Multimodal Backdoor Attack in Vision-guided Robot Grasping. IEEE Transactions on Automation Science and Engineering.
Download Paper