Perpustakaan Digital ITB

Controlling tilt-rotor Vertical Take-Off and Landing (VTOL) UAVs presents a significant challenge due to their complex and rapidly changing flight dynamics, particularly during the transition between hover and forward flight. Traditional controllers often struggle with these nonlinearities and can be rendered ineffective by environmental disturbances and model inaccuracies [1]. Although neural networks can model these complex dynamics, they often require large datasets which are risky and expensive to obtain. Hence in this thesis, a controller for a VTOL tilt rotor UAV is proposed, utilizing the reinforcement learning Proximal Policy Optimization (PPO) algorithm. PPO algorithm is chosen for its superior performance compared to other reinforcement learning algorithms [2]. In SIMULINK, a model of the UAV and the environment is created. The agent (controller) is created and trained in MATLAB. The results show that the trained controller is able to stabilize the UAV and track ground speed and altitude reference signals. Batch simulations also show small performance degradation under varied system parameters. In conclusion, reinforcement learning, particularly PPO has shown to be a worthy alternative to other nonlinear control methods in controlling the Raybe Tilt Rotor UAV.