Optimized Threat Evaluation in Maritime Environment through Reinforcement Learning Techniques

Maritime security is essential for protecting vessels, ports and coastal areas from threats such as piracy, terrorism and other criminal activities. In this context, developing intelligent systems capable of detecting threats in real time and with high accuracy is of strategic importance. The thesis focuses on the design and implementation of Reinforcement Learning (RL) algorithms for real-time threat evaluation in maritime environments. RL techniques enhance system adaptability and accuracy in dynamic and unpredictable scenarios, enabling agents to learn directly from interactions with the environment and make autonomous decisions. In collaboration with Leonardo S.p.A., this project investigates the use of RL agents to monitor maritime entities using only AIS-derived data, such as position, velocity and orientation, to identify suspicious behaviors and classify potential threats. The results demonstrate that RL offers a promising approach for the development of intelligent maritime surveillance systems capable of learning and reacting in real time.