Autonomous Sailing Behavioral Learning high level Control Architecture

We propose a high-level, behavior-based control architecture for an autonomous sailboat that enables efficacy in mission execution under diverse weather conditions, maintaining energy autonomy, while allowing the use of data driven models for learning behavioral tasks. Inspired by subsumption, this novel architecture employs hierarchical behaviors acquired through the use of the proximal policy optimization, mostly known as PPO, which is a reinforcement learning based technique. Its validity and efficacy is assessed through digital emulation of the vessel’s behaviors in the Gazebo simulation environment, combined with the ROS framework and GYM Gazebo, thus mitigating complexities and costs associated with real-world sailing operations. The successful results facilitate the creation of a resilient and versatile sailing vessel capable of handling missions without requiring the user to master navigation specifics, naval procedures, or corner cases.