Augmentation of the Self-Guided Learning Capabilities of a Reinforcement Learning based Design Optimization Algorithm for electric Fixed-Wing VTOL UAV Propellers

A Monte-Carlo Tree Search Optimization Algorithm that is trained using a reinforcement learning approach is enhanced with increased self-learning capabilities. These enable a dynamic and adaptive transition between exploration and exploitation on gained knowledge and a parameter selection process based on probability distribution functions. The algorithm is able to converge to the minimum of commonly used ontimization test functions and can devise an optimal parameter set of a propeller geometry for a fixed-wing VTOL UAV or other design geometric tasks.