The paper discusses the vulnerability of multi-agent collaborative (MAC) perception, particularly in applications like autonomous driving. MAC perception, despite its advantages, is more susceptible to adversarial attacks due to the information exchange between agents. The authors propose a reactive defense mechanism, Malicious Agent Detection (MADE), that each agent can deploy to detect and remove potential malicious agents in its local collaboration network. The paper presents comprehensive evaluations of this method, showing that it significantly reduces the performance drops caused by adversarial attacks.
Publication date: 19 Oct 2023
Project Page: Not provided
Paper: https://arxiv.org/pdf/2310.11901
