| Summary: | The aviation sector has seen several modifications, most of which are connected to autopilot systems. In the current era of artificial intelligence (AI), robust control algorithms can be tuned for optimal accuracy by utilizing AI's power. To adjust the control system for controlling the drone, this publication suggests combining the Reinforcement Learning algorithm with the ability to recognize 3D hand gestures as input. This study uses Deep Deterministic Policy Gradient (DDPG) in conjunction with 3D hand gesture recognition to determine the optimal reward and transfer the processed input to proportional integral derivative (PID) control. This method demonstrated how an AI-powered control algorithm can enhance an underactuated quadrotor unmanned aerial vehicle's (UAV) ability to fly and hover. In addition to simulation data, the study includes hardware results that were verified with a DJI Tello Drone. When compared to a typical PID flight controller, the examination of the findings shows that the new design framework provides better accuracy and computing time. Six reward functions normalized between 0 and -4000, have been estimated for training episodes of 2500, 5000, 7500, and 10,000. The greatest observation, wherein the rewards are computed for maximum value, has been recorded on 2500 episodes. © 2024 IEEE.
|
|---|