DNN regression-based rotor control strategy for coaxial drone in indoor environment

Abstract

Research on enabling autonomous flight and missions in both outdoor and indoor environments has been steadily increasing. Due to the open propeller configuration of traditional quadcopters, there is a higher likelihood of mission failure in complex and obstacle-rich indoor spaces. Drones with propellers positioned inside the platform have an advantage in consistently performing missions in such environments. Additionally, for navigating through narrow indoor spaces, a drone with coaxial contra rotating rotors is advantageous due to its compact design. The effect of thrust between two propellers is reduced to less than twice that of a single propeller due to coaxial contra rotating rotors, making precise yaw control challenging during altitude adjustments. In addition, mission in indoor spaces ground effect occurs and combined with the contra rotating rotor effect makes yaw control more difficult than when there is no ground effect. In this context, this paper derives the nonlinear model of a coaxial contra rotating drone, constructs a simulator, and analyzes altitude and yaw control considering the interference effect between the two contra rotating rotors and the ground effect. Uncertainties arising from ground effect and losses in thrust and moments due to contra rotating rotors are obtained through PID controller. These uncertainties are then compensated using a deep neural network regression method.