Deployment of Algorithms for Autonomous Drone Racing in a Real Environment

Autonomous drone racing (ADR) has proven to be a good training ground for developing and testing algorithms for autonomous drones. It usually requires specific conditions not available to many researchers. In this paper, we use the affordable Tello drone with support for ROS to implement and deploy computationally efficient algorithms for ADR. We propose the use of a control system based on a combination of the Feedback Linearization and the Model Predictive Control, localization and gate detection using a convolutional neural network and images from the drone’s front camera as inputs, and a motion planning algorithm described by a state machine. We tested our system in simulations and in an adapted outdoor environment, demonstrating the system is capable of flying through all gates without collisions.