کنترل لندینگ پهپاد با دینامیک نامشخص با استفاده از یادگیری تقویتی

The landing phase remains one of the most critical and challenging maneuvers within a drone's flight envelope. This study presents a proof-of-concept controller utilizing DDPG (Deep Deterministic Policy Gradient), a DRL (Deep Reinforcement Learning) approach, to determine optimal landing policies based on predefined requirements or rewards. The research introduces novel techniques in reward shaping and investigates the impact of hyperparameters neural network architectures on the training of drone landing control. Results from outer loop control and direct control using DDPG demonstrate the efficacy and potential of this DRL method in drone landing, showcasing robust performance across various wind disturbance conditions. Additionally, the study reveals insights into drone control and state dynamics during landing and controller design.

کلیدواژه ها:

Deep deterministic policy gradient ، deep reinforcement learning ، Neural network