Design and Implementation of an Autonomous Control System for Simulated Fixed-Wing Unmanned Aerial Vehicles

With the increasing popularity of unmanned aerial vehicles (UAVs), the demand for enhancing UAV intelligence has become a prominent trend. To improve UAV automation, flight control systems are required to enable autonomous flight without human operators. Currently, UAV flight control primarily relies on two methods: PID control and rule-based control, each with its own limitations. With advancements in reinforcement learning (RL) algorithms and computational capabilities, researchers have increasingly explored the potential of agents trained through RL to perform complex flight control tasks. This paper is dedicated to the design and implementation of an autonomous flight control system for fixed-wing UAVs. It leverages JSBSim’s aerodynamic simulator and the OpenAI Gym interface, allowing users to create various flight control tasks and design different deep RL models using RLlib, to learn corresponding optimal control strategies. The specific task we implemented to keep F16 in a steady level flight showcases the system’s capabilities to build a simulation environment, model and train the agent, and learn optimal policies using PPO. Experimental findings confirm our system’s scalability and adaptability to a wide range of flight control tasks and demonstrate the above task’s robust performance throughout the learning and testing phases.