IMAGE QUALITY-DRIVEN OCTOROTOR FLIGHT CONTROL VIA REINFORCEMENT LEARNING

This article presents the design of a reinforcement learning method based flight controller to enhance the qualities of image taken from an octorotor platform. Concerning the effect of a low resolution and a high blur rate of target images on feature extraction and target detection, we started by analyzing the relationship between these two kinds of image qualities and altitude and velocity of the octorotor. This leads to the generation of corresponding control commands. We then applied a reinforcement learning technique to automatically design the altitude and velocity controllers of the octorotor. The image analysis and the control command generation algorithms are successfully tested on the octorotor platform, and the controllers demonstrate a satisfactory performance in simulations.