Hybrid Modeling Approach for Optimization Based Control of Multirotor Unmanned Aerial Vehicles

Model Predictive Control, Sparse Identification of Nonlinear Dynamics, Unmanned Aerial Vehicle

First-principle based modeling is a fundamental approach with regards to Guidance, Navigation and Control (GNC) system development and integration. Optimization based techniques such as Model Predictive Control (MPC) often rely on simplified governing equations of the system omitting complex interactions which are difficult to accurately model or pose numerical difficulties for the optimization problem solver. This paper investigates a hybrid modeling approach based on Sparse Identification of Nonlinear Dynamics (SINDy) for local model improvement within the MPC framework. Presented hybrid modeling approach benefits from known structure of physics-based model such that the learning process in computationally lightweight. Numerical experiments assume a multirotor Unmanned Aerial Vehicle (UAV) subject to external phenomena such as ground effect or wind gusts which are typically encountered in urban environments. Model adaptation to changing mass of the vehicle is also studied within the approach.