Reinforcement Learning Approach for Autonomous UAV Navigation in 3D Space

Abstract

In the last two decades, the rapid development of unmanned aerial vehicles (UAVs) resulted in their usage for a wide range of applications. Miniaturization and cost reduction of electrical components have led to their commercialization, and today they can be utilized for various tasks in an unknown environment. Finding the optimal path based on the start and target pose information is one of the most complex demands for any intelligent UAV system. As this problem requires a high level of adaptability and learning capability of the UAV, the framework based on reinforcement learning is proposed for the localization and navigation tasks. In this paper, Q-learning algorithm for the autonomous navigation of the UAV in 3D space is implemented. To test the proposed methodology for UAV intelligent control, the simulation is conducted in ROS-Gazebo environment. The obtained simulation results have shown that the UAV can reach the target pose autonomously in an efficient way.