Multi-agent Reward-Based Intruder Capture

Patrolling surveillance systems are critical for guarding and controlling premises of any size. The primary objective of patrolling is to efficiently explore the entire environment, create a map, and detect intruders while preventing their escape. In this work, we propose a novel algorithm that employs a Finite State Machine to control the agents' behaviors, enabling them to explore the map as quickly as possible and coordinate to locate and surround a moving intruder. Our approach encourages agents to follow the most interesting paths and facilitates communication among agents within a defined communication radius of N nodes. Furthermore, our approach supports dynamic coalitions that agents can form during simulation. Despite having no prior knowledge of the environment, our algorithm can effectively coordinate agents to explore various topologically diverse environments. We evaluate our approach using different environments to validate its effectiveness.