Reinforcement learning based flight controller capable of controlling a quadcopter with four, three and two working motors

In this research, we show how a reinforcement learning based algorithm called Fault-Tolerant Bio-inspired Flight Controller (FT-BFC) is capable of training a single neural network based model to fly a quadcopter with two, three, and four working rotors. Our algorithm can learn a low-level flight controller that directly controls angular velocities of motors to fly a quadcopter when it has four fully functional motors, and also, despite having one or two motor failures (That is, our proposed flight controller is a fault-tolerant controller as well). In the training and running of our controller, we do not use any conventional flight controller, such as a PID or SMC controller. We test our algorithm in a simulation environment, Gazebo simulator, and illustrate our simulation results that backing up our algorithm capabilities. Finally, before concluding our paper, we discuss the implementation of our algorithm in a real quadcopter.