Data-Driven System Identification and Model Predictive Control of a Multirotor with an Unknown Suspended Payload

Transporting a suspended payload with a multirotor has many applications. Knowledge of the payload dynamics is required to adjust the control of the system to damp payload oscillations. Often, the dynamics of the payload are unknown and cannot be represented with an a priori model before a flight. This paper proposes a Model Predictive Control (MPC) architecture that controls a multirotor carrying an unknown suspended payload using a plant model from data-driven system identification techniques. Dynamic Mode Decomposition with Control (DMDc) and Hankel Alternative View Of Koopman with Control (HAVOKc) are the regression techniques used to identify system models without relying on modelling assumptions and by using only time series measurements. The standard Hankel Alternative View Of Koopman (HAVOK) is adapted slightly in this work for use with controlled systems. These two techniques are combined in the MPC architecture and compared against a conventional Proportional Integral Derivative (PID) system to control a multirotor with an unknown suspended payload within simulation. The results show that both MPC systems outperform the conventional system and achieve velocity control while simultaneously damping the payload swing angle. The proposed systems also show good adaptability with different payload parameters. Both system identification methods perform well with the presence of measurement noise.