Sailing Boat Route Finder with Deep Reinforcement Learning

In this thesis I will implement a system for accomplishing autonomous sailing boat route planning. For the training I will use the public historical wind data from satellites with a 10 km resolution. I will describe a deep reinforcement learning system where states are fake images of wind velocity around the boat, the boat position and orientation and the target location. For evaluating my results I will use the Vendée Globe 2020/21 competition: a sailing boats regatta in which competitors have to accomplish a round of the world trip without any help or breaks. I will use wind data of the race period and I will compare my autonomous agent's track with the GPSes of human competitors'.