Autonomous Control of a Quadcopter using Machine Learning Algorithm

During the recent years, there has been an increase in the potential use of UAVs all engineering domains such as commercial photography, aerial reconnaissance, payload delivery, etc. UAVs/Quadcopters are generally designed to operate in known and stable environmental conditions where environment dynamics are well known or can be easily linearized. But most practical problems contain unknown or non-linear dynamics of the system or the environment. Machine Learning (ML) provides the techniques for using intelligent control systems that can perform desired tasks in such unknown conditions. This paper provides a framework using a Machine Learning algorithm to enable UAV navigation in such environments through the implementation of an intelligent reinforcement learning controller. Research started with detailed mathematical modelling of a quadcopter, based on the Newton-Euler equations of forces and moments, later quadcopter model was employed with PID controller as well as ML controller. A conventional PID controller was used to find the linearized response of the quadcopter. The results obtained by both controllers were then compared using 6 DoF simulations. Furthermore, the quadcopter is made to follow certain trajectories to determine the accuracy of the ML controller.