Improving the Behavior of Evasive Targets in Cooperative Target Observation

The Cooperative Targets Observation (CTO) problem consists of two groups of agents: observers and targets. The observer agents aim to maximize the Average Number of Observed Targets (ANOT) in environments where there are more targets than observers. In most of the approaches to this problem, the behavior of the target agents is very simple, out of reality in competitive multiagent environments. More recently, two strategies improved the behavior of the targets when under observation, i.e., the straight-line strategy and controlled randomization. However, in a surveillance scenario, it is reasonable to assume that targets can be modeled as an organization, with rules, structures of authorities and relationships, and rationality to try to predict the behavior of observers. The objective of this work is to propose and validate four strategies for the team of target agents in the CTO problem, three involving clustering algorithms and two organizational paradigms and one using neural networks. The approaches were implemented and tested on the NetLogo agent-based simulation platform. The results showed that target team performance increased considerably when these were modeled as rational agents in an organization and able to try to predict the behavior of their observers.