Fuzzy Q learning based UAV autopilot

Abstract

Navigation and control of an unmanned aerial vehicle (UAV) is a challenging problem and could be framed as a Reinforcement Learning (RL) task. Herein, we propose to use reinforcement learning for designing a UAV autopilot based on the Fuzzy Q Learning (FQL) approach. Proposed control scheme envisages an amalgamation of proportional (P) control that stabilizes the UAV and an action triggering Fuzzy Inference system (FIS) control that learns the correct control action to achieve the desired flight trajectory for a UAV flight. We test the proposed RL based UAV control for three cases: (i) Altitude control (ii) Trajectory Tracking, and (iii) Reconnaissance flight of a UAV. Results demonstrate the viability and effectiveness of a UAV autopilot designed using FQL.