Adaptive Data-Driven Model Predictive Control for Quadrotor UAVs Subject to Model Uncertainties and Gust Disturbances
TL;DR
The DMPC adopts an augmented state-space formulation based on control increments, which incorporates the estimated disturbances to eliminate steady-state tracking offsets while strictly enforcing physical actuator constraints, demonstrating reliable flight operation under realistic conditions.
Abstract
This paper presents a robust and adaptive Data-driven Model Predictive Control (DMPC) strategy for quadrotors subject to model uncertainties including unmodeled dynamics, actuator efficiency variations, and sudden mass loss as well as disturbances induced by transient wind gusts. To obtain a suitable system model for MPC implementation, the proposed approach performs continuous local linearization of the highly nonlinear dynamics at each sampling instant using a Recursive Least Squares (RLS) algorithm. The RLS estimator dynamically identifies the unknown system matrices and lumped disturbances, effectively rendering the MPC as an adaptive control framework. Moreover, the DMPC adopts an augmented state-space formulation based on control increments, which incorporates the estimated disturbances to eliminate steady-state tracking offsets while strictly enforcing physical actuator constraints. Extensive numerical simulations validate the performance and robustness of the proposed strategy. The system consistently achieves precise 3D trajectory tracking in the presence of unmodeled dynamics and effectively rejects severe wind gusts, demonstrating reliable flight operation under realistic conditions.