Vertical Take-Off and Landing System Control Using Deep Reinforcement Learning

This paper introduces the utilization of a controller based on the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm for the Vertical Take-Off and Landing (VTOL) system control. The TD3 algorithm combines deep neural networks and actor-critic reinforcement learning techniques, enabling it to learn optimal policies directly from interactions with the environment. MATLAB and Simulink are used to provide an efficient platform for training and analyzing the performance of the TD3-based controller. This study compares the performance of the TD3-based controller to that of the PID controller subject to step changes and external disturbances. Results obtained by applying the TD3-based controller to one degree of freedom VTOL system demonstrate the effectiveness of the proposed control. The results, also, indicate that the proposed controller performance is better than the PID controller, with no overshoot and approximately four times faster settling time. Also, the proposed controller shows robustness against disturbances under various conditions.