Comparison of Reinforcement Learning controller with a classical controller for an UAV

Presently Reinforcement Learning (RL) poses a stiff challenge to analytical control techniques due to its model-free controller design using data and advanced statistical methods. This work attempts to compare the performance of Model Predictive Control (MPC) and a controller trained in RL for an Unmanned Aerial Vehicle (UAV). The RL agent is trained on a Soft Actor-Critic (SAC) algorithm while the PyTorch framework is used for building the Deep Neural Networks. The simulation results on the actual nonlinear UAV model indicate that the RL controller can match and even outperform MPC.