Hybrid Model-Based and Data-Driven Trajectory Tracking Control for Flapping-Wing Microaerial Vehicles

Trajectory tracking for flapping-wing microaerial vehicles (FWMAVs) is challenging due to inherent nonlinear dynamics and unsteady aerodynamics. While incremental model predictive control (IMPC) effectively manages constraints, its reliance on fixed weighting matrices limits performance during aggressive maneuvers. Conversely, reinforcement learning (RL) offers adaptability but lacks necessary stability guarantees. To address this, we propose a hybrid RL-based IMPC (RLIMPC) architecture. We first establish a quasi-steady aerodynamic model using blade element theory to improve dynamic fidelity. The proposed framework then integrates IMPC as a low-level controller for constraint satisfaction, while a high-level RL agent dynamically optimizes the IMPC weighting matrices in real time. This hierarchical approach compensates for linearization errors and adapts to varying flight envelopes. Validation results demonstrate that RLIMPC outperforms fixed-weight schemes in tracking accuracy, robustness, and control smoothness.