Creation of an RL Environment to Monitor Ocean Features with Autonomous Vehicles

The ability to detect and monitor dynamic ocean features is crucial for understanding and mitigating environmental impacts. These features, which can include variations in temperature, salinity, and biological phenomena, play a significant role in marine ecosystems and can have substantial ecological and economic consequences. This thesis explores the application of Reinforcement Learning (RL) for monitoring ocean features through the simulation of an autonomous agent operating in a dynamic and uncertain environment. The work builds on a pre-existing project, modifying the environment to integrate various ocean features and implementing advanced detection techniques supported by sensors and an optimized reward function to guide the agent’s behavior. Two different simulations have been developed and compared, each characterized by a distinct reward function. The first simulation relies on sensors capable of measuring the distance between the agent and the ocean feature, while the second utilizes a sensor that detects only the presence or absence of ocean features in the surrounding area. The obtained results highlight the potential of the RL approach in providing an effective method for detecting and monitoring ocean features, paving the way for future real-world implementations. This study can be applied to various environmental monitoring tasks, such as detecting oil spills, Posidonia meadow, and algal blooms, contributing to advancements in autonomous environmental surveillance systems.