CDA: Covert Deception Attacks in Multi-Agent Resource Scheduling

Abstract

In this letter, we address the critical security concerns in multi-agent systems, where illegal infiltration is commonly used to convert agents into malicious entities. Existing research predominantly focuses on explicit malicious attack patterns. Our work introduces a covert deception attack framework in the context of multi-agent resource scheduling scenarios. We first highlight vulnerabilities in scheduling strategies based on time and path costs. Exploiting these weaknesses, an infiltrated agent clandestinely gathers motion characteristics of other agents while posing as a teammate. Using these motion characteristics, the infiltrated agent employs an LSTM architecture to learn and predict congestion areas, thereby designing attack paths with greater time efficiency. This approach allows the infiltrated agent to secure additional resources and evade capture more effectively. Validation through simulation and real-world experiments demonstrates the feasibility and effectiveness of our approach, underscoring the importance of evaluating covert attacks in risk assessments within multi-agent systems.