Swarm Robotic Flocking With Aggregation Ability Privacy

Shuai Zhang, Yunke Huang, Weizi Li, Jia Pan

Abstract: We address the challenge of achieving flocking behavior in swarm robotic systems without compromising the privacy of individual robots’ aggregation capabilities. Traditional flocking algorithms are susceptible to privacy breaches, as adversaries can deduce the identity and aggregation abilities of robots by observing their movements. We introduce a novel control mechanism for privacy-preserving flocking, leveraging the Laplace mechanism within the framework of differential privacy. Our method mitigates privacy breaches by introducing a controlled level of noise, thus obscuring sensitive information. We explore the trade-off between privacy and utility by varying the differential privacy parameter $\epsilon $ . Our quantitative analysis reveals that $\epsilon \leq 0.13$ represents a lower threshold where private information is almost completely protected, whereas $\epsilon \geq 0.85$ marks an upper threshold where private information cannot be protected at all. Empirical results validate that our approach effectively maintains privacy of the robots’ aggregation abilities throughout the flocking process. Note to Practitioners—This paper was motivated by the problem of preserving privacy of individual robots in a swarm robotic system. Existing approaches to address this issue generally consider that accomplishing complex tasks requiring explicit information sharing between robots, while explicit communication in public channel carries the risk of information leakage. It is not always like this in real adversarial environments, and this assumption restricts the investigation of privacy in autonomous systems. This paper suggests that an individual robot can use its sensors onboard to perceive states of other neighbors in a distributed way without explicit communication. Despite avoiding information leakage during explicit information sharing between robots, the configuration of swarm can still reveal sensitive information about the ability of each robot. In this paper, we propose a privacy-preserving approach for flocking control using the Laplace mechanism based on the concept of differential privacy. The solution prevents an adversary with full knowledge of the swarm’s configuration from learning the sensitive information of individual robots, thus ensuring the security of swarm robots in terms of sensitive information during ongoing missions.