Decentralized Robot Swarm Clustering: Adding Resilience to Malicious Masquerade Attacks

Mitali Gandhe, Michael W. Otte

Abstract: We compare the resilience of four distributed robot swarm clustering algorithms to masquerade attacks launched from malicious robots within the swarm. The clustering algorithms are distributed variants of DBSCAN and k-Means that have been modified for use on a distributed robot swarm that only has access to local communication and local distance measurements. We subject these distributed variants of k-Means and DBSCAN to malicious masquerade attacks and observe how clustering performance is affected. We then modify each variant to include a distributed Intrusion Detection and Response System (IDRS) to detect malicious robots and maintain the swarm’s integrity despite an attack. We evaluate all four variants both in simulation and in a hardware testbed containing a swarm of 25 Kilobot robots. We find that centralizing data within the swarm makes the swarm more vulnerable to malicious attacks, and that distributed IDRS relying on local message passing can effectively identify malicious robots and reduce their negative effects on swarm clustering performance.