Reinforcement-learning-based robust optimal control for underactuated USV with prescribed performance | Synapse