Level Curve Tracking via Robust RL-Guided Model Predictive Control

Abstract

Dear Editor, This letter investigates the level curve tracking problem of unknown scalar fields using an unmanned aerial vehicle (UAV) and presents a robust reinforcement learning (RL)-guided model predictive control (MPC) scheme for the UAV. Specifically, we formulate the MPC trajectory tracking problem, wherein an RL-based trajectory planning algorithm provides the reference trajectory to guide the UAV towards the desired concentration. Notably, we introduce random noise during the training of the planning policy in the RL algorithm, demonstrating its efficacy in enhancing policy robustness. The proposed RL-guided MPC scheme's effectiveness is validated through simulations.