Learning-Based External Wrench Estimation for Quadrotors

In this work, we present a quadrotor external disturbance estimator without any preliminary knowledge about system dynamic using machine learning. Traditionally, measuring the external wrench directly requires large demands on sensor accuracy and would increase the take-off weight of quadrotors. This paper presents an estimation method based on machine learning using system states measured by sensors such as position, velocity, attitude and angular velocity. This method can train a highly accurate estimator without further information about the quadrotors dynamic to predict external wrench under simulation environment. Our proposed method is verified through simulation and compare to a traditional wrench estimator.