Quadrotor Control using Reinforcement Learning under Wind Disturbance

Quadrotors usually suffer various external disturbances when flying in the real-world environment. A robust and adaptive controller is required to maintain the stability. Traditional control methods require to handcraft the model of the disturbance, making them difficult to adapt to various scenarios. In this paper, a reinforcement-learning-based (RL-based) quadrotor control method is proposed to counter external disturbances. By using Proximal Policy Optimization (PPO), a policy network is trained to directly output the desired 3-axis Euler angles and throttle for feedback control. In simulated experiments and real-world experiments, we show quantitative results via the comparison between the traditional method and our proposed method.