Dynamic Data Integration for Resilience to Sensor Attacks in Multi-Agent Systems

Abstract

In recent years the number of security incidents affecting control systems has increased. These incidents have shown the need to develop strategies to improve system resilience to cyber-attacks. This paper presents a practical implementation of a strategy to detect cyber-attacks and mitigate their effects on sensors of a multi-agent system. The proposed approach computes, in real-time, a convex combination of measurements from main and redundant sensors, producing a trust value of the measurement and feeding it to the controller. We implemented this approach on a testbed of ground robots in formation. Experimental results to various kinds of attacks and a key performance index show that the proposed strategy reduces the effects of attacks not only on the affected agent, but also prevents the propagation of the attack over the remaining agents.