Attitude Control of Flying-Wing UAV based on Deep Deterministic Policy Gradient

This paper explored the the attitude coupling control of the fly-wing UAV based on the deep reinforcement learning, and the six-DOF nonlinear dynamic model for a UAV is constructed as the environment for agent training. The input state and reward function containing attitude error, attitude integration error, angular velocity and other information are designed to realize the attitude stability coupling control of longitudinal and lateral. The simulation results show that the agent can be extended to various flight missions, and is superior to the PID in terms of overshoot. The agent can adapt to various working conditions such as turbulent wind disturbance, sensor noise, model parameter deviation, and has good generalization ability and robustness.