Multi-Agent System that Attains Longevity via Death

Megan Olsen, Hava Siegelmann

We propose a novel approach to self-regenerating systems which require continuous operation, such as security surveillance. For that aim we introduce HADES, a self-regenerating cooperative multi-agent system with local monitoring. When agents of HADES find local failures they repair them. However, in extreme cases repair may not be possible and irregular aggressive agents will multiply. These irregular agents may use all of the system's resources and thus take over the system. To optimize system longevity, we identify protocols for killing these irregular agents. Our primary contribution is a double communication protocol of alert and death signals among the agents, making the multi-agent system robust to failures and attacks.