Neural-Network for Position Estimation of a Cable-Suspended Payload Using Inertial Quadrotor Sensing

This paper considers a standard quadrotor drone with a cable-suspended payload and minimal sensor configuration. A neural network estimator is proposed to perform accurate real-time payload position estimation. A novel proprioceptive feedback measurement method is proposed, and a neural network has been trained with domain randomization. The network shows accurate zero-shot estimation, even with excitations never seen by the system before. This preliminary work has been tested in a simulated environment and aims to show that only onboard inertial sensing is enough to achieve the sought task. The presented work may open new applications for drone transportation in real environments subject to several perturbations.