Human-Robot Collaboration Attack Based on Perception Deception
Yajie Liu, Jie Ren, Xianglong Meng, Chuanbin Sun, Yanzhi Dong, Xiaohu Yuan · 2024
Human-robot collaboration (HRC) plays a crucial role in various fields, highly relying on the perception capabilities of robots. While some literature explores HRC, research on potential attacks against HRC systems remains lacking. This research gap could lead to system vulnerabilities being exploited, affecting productivity and worker safety, and hindering the further adoption and application of HRC technology. We propose a perception-based deception attack method for HRC, which misguides the robot's perception to attack its decision-making, thereby impacting the effectiveness of human-robot collaboration. Our experiments involved a widely-used HRC task and demonstrated that perception deception could reduce the human-robot collaboration completion rate by over 36.67% and increase decision latency by 87.58%. The experimental results indicate that these attack methods can significantly reduce the robot's perception success rate. Through this research, we lay the foundation for further studies on improving the efficiency and robustness of human-robot collaboration.